Update HISTORY.md -- release RocksDB 3.0

BackupableDBTest thread-safe
Summary: We need to lock accesses to some TestEnv variables. Otherwise we get failures like http://ci-builds.fb.com/job/rocksdb_asan_check/657/console Test Plan: make check Reviewers: dhruba, haobo, sdong Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D18489
2026-11-06 02:52:35 -05:00 · 2014-05-05 14:44:57 -07:00 · 2014-05-05 14:30:24 -07:00 · 2014-05-05 12:57:47 -07:00 · 2014-05-05 11:35:40 -07:00 · 2014-05-05 11:11:48 -07:00
308 changed files with 42156 additions and 11391 deletions
@@ -6,6 +6,5 @@
    "linters"
  ],
  "lint.engine" : "FacebookFbcodeLintEngine",
-  "lint.engine.single.linter" : "FbcodeCppLinter",
-  "lint.cpplint.prefix" : "linters"
+  "lint.engine.single.linter" : "FbcodeCppLinter"
 }
@@ -1,3 +1,4 @@
+TARGETS
 build_config.mk

 *.a
@@ -12,6 +13,12 @@ build_config.mk
 *_test
 *_bench
 *_stress
+*.out
+*.class
+*.jar
+*.*jnilib*
+*.d-e
+*.o-*

 ldb
 manifest_dump
@@ -21,3 +28,6 @@ build_tools/VALGRIND_LOGS/
 coverage/COVERAGE_REPORT
 .gdbhistory
 .phutil_module_cache
+tags
+java/*.log
+java/include/org_rocksdb_*.h
@@ -8,7 +8,11 @@ open source project, you're good to go. If you are submitting a pull
 request for the first time, just let us know that you have completed
 the CLA and we can cross-check with your GitHub username.

-Complete your CLA here: <https://developers.facebook.com/opensource/cla>
+Complete your CLA here: <https://code.facebook.com/cla>
+
+If you don't have a Facebook account, we can send you a PDF that you can
+sign offline. Send us an e-mail or create a new github issue to
+request the CLA in PDF format.

 ## License

@@ -1,5 +1,45 @@
 # Rocksdb Change Log

+## 3.0.0 (05/05/2014)
+
+### Public API changes
+* Added _LEVEL to all InfoLogLevel enums
+* Deprecated ReadOptions.prefix and ReadOptions.prefix_seek. Seek() defaults to prefix-based seek when Options.prefix_extractor is supplied. More detail is documented in https://github.com/facebook/rocksdb/wiki/Prefix-Seek-API-Changes
+
+### New Features
+* Column family support
+* Added an option to use different checksum functions in BlockBasedTableOptions
+* Added ApplyToAllCacheEntries() function to Cache
+
+## 2.8.0 (04/04/2014)
+
+* Removed arena.h from public header files.
+* By default, checksums are verified on every read from database
+* Change default value of several options, including: paranoid_checks=true, max_open_files=5000, level0_slowdown_writes_trigger=20, level0_stop_writes_trigger=24, disable_seek_compaction=true, max_background_flushes=1 and allow_mmap_writes=false
+* Added is_manual_compaction to CompactionFilter::Context
+* Added "virtual void WaitForJoin()" in class Env. Default operation is no-op.
+* Removed BackupEngine::DeleteBackupsNewerThan() function
+* Added new option -- verify_checksums_in_compaction
+* Changed Options.prefix_extractor from raw pointer to shared_ptr (take ownership)
+  Changed HashSkipListRepFactory and HashLinkListRepFactory constructor to not take SliceTransform object (use Options.prefix_extractor implicitly)
+* Added Env::GetThreadPoolQueueLen(), which returns the waiting queue length of thread pools
+* Added a command "checkconsistency" in ldb tool, which checks
+  if file system state matches DB state (file existence and file sizes)
+* Separate options related to block based table to a new struct BlockBasedTableOptions.
+* WriteBatch has a new function Count() to return total size in the batch, and Data() now returns a reference instead of a copy
+* Add more counters to perf context.
+* Supports several more DB properties: compaction-pending, background-errors and cur-size-active-mem-table.
+
+### New Features
+* If we find one truncated record at the end of the MANIFEST or WAL files,
+  we will ignore it. We assume that writers of these records were interrupted
+  and that we can safely ignore it.
+* A new SST format "PlainTable" is added, which is optimized for memory-only workloads. It can be created through NewPlainTableFactory() or NewTotalOrderPlainTableFactory().
+* A new mem table implementation hash linked list optimizing for the case that there are only few keys for each prefix, which can be created through NewHashLinkListRepFactory().
+* Merge operator supports a new function PartialMergeMulti() to allow users to do partial merges against multiple operands.
+* Now compaction filter has a V2 interface. It buffers the kv-pairs sharing the same key prefix, process them in batches, and return the batched results back to DB. The new interface uses a new structure CompactionFilterContext for the same purpose as CompactionFilter::Context in V1.
+* Geo-spatial support for locations and radial-search.
+
 ## 2.7.0 (01/28/2014)

 ### Public API changes
@@ -17,15 +17,42 @@ libraries. You are on your own.

 ## Supported platforms

-* **Linux**
+* **Linux - Ubuntu**
    * Upgrade your gcc to version at least 4.7 to get C++11 support.
-    * Install gflags. First, try: `sudo apt-get install libgflags-dev`.
+    * Install gflags. First, try: `sudo apt-get install libgflags-dev`
      If this doesn't work and you're using Ubuntu, here's a nice tutorial:
      (http://askubuntu.com/questions/312173/installing-gflags-12-04)
    * Install snappy. This is usually as easy as:
      `sudo apt-get install libsnappy-dev`.
    * Install zlib. Try: `sudo apt-get install zlib1g-dev`.
    * Install bzip2: `sudo apt-get install libbz2-dev`.
+* **Linux - CentOS**
+    * Upgrade your gcc to version at least 4.7 to get C++11 support:
+      `yum install gcc47-c++`
+    * Install gflags:
+
+              wget https://gflags.googlecode.com/files/gflags-2.0-no-svn-files.tar.gz
+              tar -xzvf gflags-2.0-no-svn-files.tar.gz
+              cd gflags-2.0
+              ./configure && make && sudo make install
+
+    * Install snappy:
+
+              wget https://snappy.googlecode.com/files/snappy-1.1.1.tar.gz
+              tar -xzvf snappy-1.1.1.tar.gz
+              cd snappy-1.1.1
+              ./configure && make && sudo make install
+
+    * Install zlib:
+
+              sudo yum install zlib
+              sudo yum install zlib-devel
+
+    * Install bzip2:
+
+              sudo yum install bzip2
+              sudo yum install bzip2-devel
+
 * **OS X**:
    * Install latest C++ compiler that supports C++ 11:
        * Update XCode:  run `xcode-select --install` (or install it from XCode App's settting).
@@ -40,11 +67,14 @@ libraries. You are on your own.
    * Please note that some of the optimizations/features are disabled in OSX.
    We did not run any production workloads on it.

+* **iOS**:
+  * Run: `TARGET_OS=IOS make static_lib`
+
 ## Compilation
-`make clean; make` will compile librocksdb.a (RocskDB static library) and all
+`make clean; make` will compile librocksdb.a (RocksDB static library) and all
 the unit tests. You can run all unit tests with `make check`.

-For shared library builds, exec `make librocksdb.so` instead.
+For shared library builds, exec `make shared_lib` instead.

 If you followed the above steps and your compile or unit tests fail,
 please submit an issue: (https://github.com/facebook/rocksdb/issues)
@@ -2,7 +2,7 @@ BSD License

 For rocksdb software

-Copyright (c) 2013, Facebook, Inc.
+Copyright (c) 2014, Facebook, Inc.
 All rights reserved.
 ---------------------------------------------------------------------

@@ -6,11 +6,23 @@
 INSTALL_PATH ?= $(CURDIR)

 #-----------------------------------------------
-# Uncomment exactly one of the lines labelled (A), (B), and (C) below
-# to switch between compilation modes.

-# OPT ?= -DNDEBUG     # (A) Production use (optimized mode)
+ifneq ($(MAKECMDGOALS),dbg)
 OPT += -O2 -fno-omit-frame-pointer -momit-leaf-frame-pointer
+else
+# intentionally left blank
+endif
+
+ifeq ($(MAKECMDGOALS),shared_lib)
+PLATFORM_SHARED_LDFLAGS=-fPIC
+OPT += -DNDEBUG
+endif
+
+ifeq ($(MAKECMDGOALS),static_lib)
+PLATFORM_SHARED_LDFLAGS=-fPIC
+OPT += -DNDEBUG
+endif
+
 #-----------------------------------------------

 # detect what platform we're building on
@@ -18,6 +30,14 @@ $(shell (export ROCKSDB_ROOT=$(CURDIR); $(CURDIR)/build_tools/build_detect_platf
 # this file is generated by the previous line to set build flags and sources
 include build_config.mk

+ifneq ($(PLATFORM), IOS)
+CFLAGS += -g
+CXXFLAGS += -g
+else
+# no debug info for IOS, that will make our library big
+OPT += -DNDEBUG
+endif
+
 # ASAN doesn't work well with jemalloc. If we're compiling with ASAN, we should use regular malloc.
 ifdef COMPILE_WITH_ASAN
 	# ASAN compile flags
@@ -32,8 +52,8 @@ else
 endif

 WARNING_FLAGS = -Wall -Werror
-CFLAGS += -g $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
-CXXFLAGS += -g $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual
+CFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
+CXXFLAGS += $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -Woverloaded-virtual

 LDFLAGS += $(PLATFORM_LDFLAGS)

@@ -43,6 +63,7 @@ MEMENVOBJECTS = $(MEMENV_SOURCES:.cc=.o)

 TESTUTIL = ./util/testutil.o
 TESTHARNESS = ./util/testharness.o $(TESTUTIL)
+BENCHHARNESS = ./util/benchharness.o
 VALGRIND_ERROR = 2
 VALGRIND_DIR = build_tools/VALGRIND_LOGS
 VALGRIND_VER := $(join $(VALGRIND_VER),valgrind)
@@ -50,12 +71,16 @@ VALGRIND_OPTS = --error-exitcode=$(VALGRIND_ERROR) --leak-check=full

 TESTS = \
 	db_test \
+	block_hash_index_test \
 	autovector_test \
+	column_family_test \
 	table_properties_collector_test \
 	arena_test \
 	auto_roll_logger_test \
+	benchharness_test \
 	block_test \
 	bloom_test \
+	dynamic_bloom_test \
 	c_test \
 	cache_test \
 	coding_test \
@@ -74,6 +99,8 @@ TESTS = \
 	merge_test \
 	redis_test \
 	reduce_levels_test \
+	plain_table_db_test \
+	prefix_test \
 	simple_table_db_test \
 	skiplist_test \
 	stringappend_test \
@@ -81,23 +108,28 @@ TESTS = \
 	backupable_db_test \
 	version_edit_test \
 	version_set_test \
+	file_indexer_test \
 	write_batch_test\
 	deletefile_test \
-	table_test
+	table_test \
+	thread_local_test \
+        geodb_test

 TOOLS = \
        sst_dump \
+	db_sanity_test \
        db_stress \
        ldb \
 	db_repl_stress \
 	blob_store_bench

-PROGRAMS = db_bench signal_test $(TESTS) $(TOOLS)
-BENCHMARKS = db_bench_sqlite3 db_bench_tree_db table_reader_bench
+PROGRAMS = db_bench signal_test table_reader_bench log_and_apply_bench $(TOOLS)

 # The library name is configurable since we are maintaining libraries of both
 # debug/release mode.
-LIBNAME = librocksdb
+ifeq ($(LIBNAME),)
+        LIBNAME=librocksdb
+endif
 LIBRARY = ${LIBNAME}.a
 MEMENVLIBRARY = libmemenv.a

@@ -115,7 +147,7 @@ SHARED3 = $(SHARED1)
 SHARED = $(SHARED1)
 else
 # Update db.h if you change these.
-SHARED_MAJOR = 2
+SHARED_MAJOR = 3
 SHARED_MINOR = 0
 SHARED1 = ${LIBNAME}.$(PLATFORM_SHARED_EXT)
 SHARED2 = $(SHARED1).$(SHARED_MAJOR)
@@ -127,36 +159,47 @@ $(SHARED2): $(SHARED3)
 	ln -fs $(SHARED3) $(SHARED2)
 endif

-$(SHARED3): $(LIBOBJECTS)
-	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED2) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(LDFLAGS) $(SOURCES)-o $@
+$(SHARED3):
+	$(CXX) $(PLATFORM_SHARED_LDFLAGS)$(SHARED2) $(CXXFLAGS) $(PLATFORM_SHARED_CFLAGS) $(SOURCES) $(LDFLAGS) -o $@

 endif  # PLATFORM_SHARED_EXT

-all: $(LIBRARY) $(PROGRAMS)
-
 .PHONY: blackbox_crash_test check clean coverage crash_test ldb_tests \
-	release tags valgrind_check whitebox_crash_test format shared_lib
+	release tags valgrind_check whitebox_crash_test format static_lib shared_lib all \
+	dbg

-# Will also generate shared libraries. 
+all: $(LIBRARY) $(PROGRAMS) $(TESTS)
+
+static_lib: $(LIBRARY)
+
+shared_lib: $(SHARED)
+
+dbg: $(LIBRARY) $(PROGRAMS) $(TESTS)
+
+# creates static library and programs
 release:
 	$(MAKE) clean
-	OPT=-DNDEBUG $(MAKE) all -j32
+	OPT="-DNDEBUG -O2" $(MAKE) static_lib $(PROGRAMS) -j32
+
+release_shared_lib:
+	$(MAKE) clean
+	OPT="-DNDEBUG -O2" $(MAKE) shared_lib -j32

 coverage:
 	$(MAKE) clean
-	COVERAGEFLAGS="-fprofile-arcs -ftest-coverage" LDFLAGS+="-lgcov" $(MAKE) all check
+	COVERAGEFLAGS="-fprofile-arcs -ftest-coverage" LDFLAGS+="-lgcov" $(MAKE) all check -j32
 	(cd coverage; ./coverage_test.sh)
 	# Delete intermediate files
 	find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;

-check: all $(PROGRAMS) $(TESTS) $(TOOLS)
+check: $(PROGRAMS) $(TESTS) $(TOOLS)
 	for t in $(TESTS); do echo "***** Running $$t"; ./$$t || exit 1; done
 	python tools/ldb_test.py

-ldb_tests: all $(PROGRAMS) $(TOOLS)
+ldb_tests: all $(PROGRAMS) $(TESTS) $(TOOLS)
 	python tools/ldb_test.py

-crash_test: blackbox_crash_test whitebox_crash_test
+crash_test: whitebox_crash_test blackbox_crash_test

 blackbox_crash_test: db_stress
 	python -u tools/db_crashtest.py
@@ -171,7 +214,7 @@ asan_check:

 asan_crash_test:
 	$(MAKE) clean
-	COMPILE_WITH_ASAN=1 $(MAKE) crash_test -j32
+	COMPILE_WITH_ASAN=1 $(MAKE) crash_test
 	$(MAKE) clean

 valgrind_check: all $(PROGRAMS) $(TESTS)
@@ -189,7 +232,7 @@ valgrind_check: all $(PROGRAMS) $(TESTS)
 	done

 clean:
-	-rm -f $(PROGRAMS) $(BENCHMARKS) $(LIBRARY) $(SHARED) $(MEMENVLIBRARY) build_config.mk
+	-rm -f $(PROGRAMS) $(TESTS) $(LIBRARY) $(SHARED) $(MEMENVLIBRARY) build_config.mk
 	-rm -rf ios-x86/* ios-arm/*
 	-find . -name "*.[od]" -exec rm {} \;
 	-find . -type f -regex ".*\.\(\(gcda\)\|\(gcno\)\)" -exec rm {} \;
@@ -200,8 +243,6 @@ tags:
 format:
 	build_tools/format-diff.sh

-shared_lib: $(SHARED)
-
 # ---------------------------------------------------------------------------
 # 	Unit tests and tools
 # ---------------------------------------------------------------------------
@@ -212,21 +253,21 @@ $(LIBRARY): $(LIBOBJECTS)
 db_bench: db/db_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(CXX) db/db_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)
+
 db_stress: tools/db_stress.o $(LIBOBJECTS) $(TESTUTIL)
 	$(CXX) tools/db_stress.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)

+db_sanity_test: tools/db_sanity_test.o $(LIBOBJECTS) $(TESTUTIL)
+	$(CXX) tools/db_sanity_test.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)
+
 db_repl_stress: tools/db_repl_stress.o $(LIBOBJECTS) $(TESTUTIL)
 	$(CXX) tools/db_repl_stress.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)

 blob_store_bench: tools/blob_store_bench.o $(LIBOBJECTS) $(TESTUTIL)
 	$(CXX) tools/blob_store_bench.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS) $(COVERAGEFLAGS)

-db_bench_sqlite3: doc/bench/db_bench_sqlite3.o $(LIBOBJECTS) $(TESTUTIL)
-	$(CXX) doc/bench/db_bench_sqlite3.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) -lsqlite3 $(COVERAGEFLAGS)
-
-db_bench_tree_db: doc/bench/db_bench_tree_db.o $(LIBOBJECTS) $(TESTUTIL)
-	$(CXX) doc/bench/db_bench_tree_db.o $(LIBOBJECTS) $(TESTUTIL) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) -lkyotocabinet $(COVERAGEFLAGS)
-
 signal_test: util/signal_test.o $(LIBOBJECTS)
 	$(CXX) util/signal_test.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

@@ -236,12 +277,18 @@ arena_test: util/arena_test.o $(LIBOBJECTS) $(TESTHARNESS)
 autovector_test: util/autovector_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/autovector_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

+column_family_test: db/column_family_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) db/column_family_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 table_properties_collector_test: db/table_properties_collector_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/table_properties_collector_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

 bloom_test: util/bloom_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/bloom_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

+dynamic_bloom_test: util/dynamic_bloom_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) util/dynamic_bloom_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 c_test: db/c_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/c_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

@@ -260,9 +307,15 @@ stringappend_test: utilities/merge_operators/string_append/stringappend_test.o $
 redis_test: utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

+benchharness_test: util/benchharness_test.o $(LIBOBJECTS) $(TESTHARNESS) $(BENCHHARNESS)
+	$(CXX) util/benchharness_test.o $(LIBOBJECTS) $(TESTHARNESS) $(BENCHHARNESS) $(EXEC_LDFLAGS) -o$@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 histogram_test: util/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o$@ $(LDFLAGS) $(COVERAGEFLAGS)

+thread_local_test: util/thread_local_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) util/thread_local_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 corruption_test: db/corruption_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/corruption_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

@@ -272,11 +325,20 @@ 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)

+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)
+	$(CXX) table/table_reader_bench.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) -pg
+
+log_and_apply_bench: db/log_and_apply_bench.o $(LIBOBJECTS) $(TESTHARNESS) $(BENCHHARNESS)
+	$(CXX) db/log_and_apply_bench.o $(LIBOBJECTS) $(TESTHARNESS) $(BENCHHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS) -pg

 perf_context_test: db/perf_context_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/perf_context_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
@@ -320,6 +382,9 @@ version_edit_test: db/version_edit_test.o $(LIBOBJECTS) $(TESTHARNESS)
 version_set_test: db/version_set_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/version_set_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

+file_indexer_test : db/file_indexer_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) db/file_indexer_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+
 reduce_levels_test: tools/reduce_levels_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) tools/reduce_levels_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

@@ -332,12 +397,15 @@ merge_test: db/merge_test.o $(LIBOBJECTS) $(TESTHARNESS)
 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)
+
 $(MEMENVLIBRARY) : $(MEMENVOBJECTS)
 	rm -f $@
 	$(AR) -rs $@ $(MEMENVOBJECTS)

-memenv_test : helpers/memenv/memenv_test.o $(MEMENVLIBRARY) $(LIBRARY) $(TESTHARNESS)
-	$(CXX) helpers/memenv/memenv_test.o $(MEMENVLIBRARY) $(LIBRARY) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
+memenv_test : helpers/memenv/memenv_test.o $(MEMENVOBJECTS) $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) helpers/memenv/memenv_test.o $(MEMENVOBJECTS) $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

 manual_compaction_test: util/manual_compaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/manual_compaction_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)
@@ -360,6 +428,35 @@ sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 ldb: tools/ldb.o $(LIBOBJECTS)
 	$(CXX) tools/ldb.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS) $(COVERAGEFLAGS)

+# ---------------------------------------------------------------------------
+# Jni stuff
+# ---------------------------------------------------------------------------
+
+JNI_NATIVE_SOURCES = ./java/rocksjni/*.cc
+JAVA_INCLUDE = -I$(JAVA_HOME)/include/ -I$(JAVA_HOME)/include/linux
+ROCKSDBJNILIB = ./java/librocksdbjni.so
+
+ifeq ($(PLATFORM), OS_MACOSX)
+ROCKSDBJNILIB = ./java/librocksdbjni.jnilib
+JAVA_INCLUDE = -I/System/Library/Frameworks/JavaVM.framework/Headers/
+endif
+
+rocksdbjava: clean
+	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) $(COVERAGEFLAGS)
+
+jclean:
+	cd java;$(MAKE) clean;
+	rm -f $(ROCKSDBJNILIB)
+
+jtest:
+	cd java;$(MAKE) sample;$(MAKE) test;
+
+jdb_bench:
+	cd java;$(MAKE) db_bench;
+
 # ---------------------------------------------------------------------------
 #  	Platform-specific compilation
 # ---------------------------------------------------------------------------
@@ -370,20 +467,20 @@ ifeq ($(PLATFORM), IOS)
 PLATFORMSROOT=/Applications/Xcode.app/Contents/Developer/Platforms
 SIMULATORROOT=$(PLATFORMSROOT)/iPhoneSimulator.platform/Developer
 DEVICEROOT=$(PLATFORMSROOT)/iPhoneOS.platform/Developer
-IOSVERSION=$(shell defaults read $(PLATFORMSROOT)/iPhoneOS.platform/versionCFBundleShortVersionString)
+IOSVERSION=$(shell defaults read $(PLATFORMSROOT)/iPhoneOS.platform/version CFBundleShortVersionString)

 .cc.o:
 	mkdir -p ios-x86/$(dir $@)
-	$(SIMULATORROOT)/usr/bin/$(CXX) $(CXXFLAGS) -isysroot $(SIMULATORROOT)/SDKs/iPhoneSimulator$(IOSVERSION).sdk -arch i686 -c $< -o ios-x86/$@ $(COVERAGEFLAGS)
+	$(CXX) $(CXXFLAGS) -isysroot $(SIMULATORROOT)/SDKs/iPhoneSimulator$(IOSVERSION).sdk -arch i686 -arch x86_64 -c $< -o ios-x86/$@
 	mkdir -p ios-arm/$(dir $@)
-	$(DEVICEROOT)/usr/bin/$(CXX) $(CXXFLAGS) -isysroot $(DEVICEROOT)/SDKs/iPhoneOS$(IOSVERSION).sdk -arch armv6 -arch armv7 -c $< -o ios-arm/$@ $(COVERAGEFLAGS)
+	xcrun -sdk iphoneos $(CXX) $(CXXFLAGS) -isysroot $(DEVICEROOT)/SDKs/iPhoneOS$(IOSVERSION).sdk -arch armv6 -arch armv7 -arch armv7s -arch arm64 -c $< -o ios-arm/$@
 	lipo ios-x86/$@ ios-arm/$@ -create -output $@

 .c.o:
 	mkdir -p ios-x86/$(dir $@)
-	$(SIMULATORROOT)/usr/bin/$(CC) $(CFLAGS) -isysroot $(SIMULATORROOT)/SDKs/iPhoneSimulator$(IOSVERSION).sdk -arch i686 -c $< -o ios-x86/$@
+	$(CC) $(CFLAGS) -isysroot $(SIMULATORROOT)/SDKs/iPhoneSimulator$(IOSVERSION).sdk -arch i686 -arch x86_64 -c $< -o ios-x86/$@
 	mkdir -p ios-arm/$(dir $@)
-	$(DEVICEROOT)/usr/bin/$(CC) $(CFLAGS) -isysroot $(DEVICEROOT)/SDKs/iPhoneOS$(IOSVERSION).sdk -arch armv6 -arch armv7 -c $< -o ios-arm/$@
+	xcrun -sdk iphoneos $(CC) $(CFLAGS) -isysroot $(DEVICEROOT)/SDKs/iPhoneOS$(IOSVERSION).sdk -arch armv6 -arch armv7 -arch armv7s -arch arm64 -c $< -o ios-arm/$@
 	lipo ios-x86/$@ ios-arm/$@ -create -output $@

 else
@@ -423,6 +520,10 @@ depend: $(DEPFILES)
 # working solution.
 ifneq ($(MAKECMDGOALS),clean)
 ifneq ($(MAKECMDGOALS),format)
+ifneq ($(MAKECMDGOALS),jclean)
+ifneq ($(MAKECMDGOALS),jtest)
 -include $(DEPFILES)
 endif
 endif
+endif
+endif
@@ -1,3 +0,0 @@
-* Detailed instructions on how to compile using fbcode and jemalloc
-
-* Latest release is 2.7.fb
@@ -0,0 +1,20 @@
+# RocksDBLite
+
+RocksDBLite is a project focused on mobile use cases, which don't need a lot of fancy things we've built for server workloads and they are very sensitive to binary size. For that reason, we added a compile flag ROCKSDB_LITE that comments out a lot of the nonessential code and keeps the binary lean.
+
+Some examples of the features disabled by ROCKSDB_LITE:
+* compiled-in support for LDB tool
+* No backupable DB
+* No support for replication (which we provide in form of TrasactionalIterator)
+* No advanced monitoring tools
+* No special-purpose memtables that are highly optimized for specific use cases
+
+When adding a new big feature to RocksDB, please add ROCKSDB_LITE compile guard if:
+* Nobody from mobile really needs your feature,
+* Your feature is adding a lot of weight to the binary.
+
+Don't add ROCKSDB_LITE compile guard if:
+* It would introduce a lot of code complexity. Compile guards make code harder to read. It's a trade-off.
+* Your feature is not adding a lot of weight.
+
+If unsure, ask. :)
@@ -19,7 +19,8 @@
 #
 #       -DLEVELDB_PLATFORM_POSIX if cstdatomic is present
 #       -DLEVELDB_PLATFORM_NOATOMIC if it is not
-#       -DSNAPPY                     if the Snappy library is present
+#       -DSNAPPY                    if the Snappy library is present
+#       -DLZ4                       if the LZ4 library is present
 #
 # Using gflags in rocksdb:
 # Our project depends on gflags, which requires users to take some extra steps
@@ -38,7 +39,7 @@ if test -z "$OUTPUT"; then
 fi

 # we depend on C++11
-PLATFORM_CXXFLAGS="-std=gnu++11"
+PLATFORM_CXXFLAGS="-std=c++11"
 # we currently depend on POSIX platform
 COMMON_FLAGS="-DROCKSDB_PLATFORM_POSIX"

@@ -86,7 +87,7 @@ PLATFORM_SHARED_CFLAGS="-fPIC"
 PLATFORM_SHARED_VERSIONED=false

 # generic port files (working on all platform by #ifdef) go directly in /port
-GENERIC_PORT_FILES=`find $ROCKSDB_ROOT/port -name '*.cc' | tr "\n" " "`
+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 "$TARGET_OS" in
@@ -97,6 +98,13 @@ case "$TARGET_OS" in
        PLATFORM_SHARED_LDFLAGS="-dynamiclib -install_name "
        # PORT_FILES=port/darwin/darwin_specific.cc
        ;;
+    IOS)
+        PLATFORM=IOS
+        COMMON_FLAGS="$COMMON_FLAGS -DOS_MACOSX -DIOS_CROSS_COMPILE -DROCKSDB_LITE"
+        PLATFORM_SHARED_EXT=dylib
+        PLATFORM_SHARED_LDFLAGS="-dynamiclib -install_name "
+        CROSS_COMPILE=true
+        ;;
    Linux)
        PLATFORM=OS_LINUX
        COMMON_FLAGS="$COMMON_FLAGS -DOS_LINUX"
@@ -148,7 +156,9 @@ case "$TARGET_OS" in
        exit 1
 esac

-$PWD/build_tools/build_detect_version
+if test -z "$DO_NOT_RUN_BUILD_DETECT_VERSION"; then
+  $PWD/build_tools/build_detect_version
+fi

 # 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
@@ -158,7 +168,7 @@ 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"
+PRUNE_BENCH="-name *bench*.cc -prune"
 PORTABLE_FILES=`cd $ROCKSDB_ROOT; find $DIRS $PRUNE_TEST -o $PRUNE_BENCH -o -name '*.cc' -print | sort | tr "\n" " "`
 PORTABLE_CPP=`cd $ROCKSDB_ROOT; find $DIRS $PRUNE_TEST -o $PRUNE_BENCH -o -name '*.cpp' -print | sort | tr "\n" " "`
 set +f # re-enable globbing
@@ -244,6 +254,17 @@ EOF
        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lbz2"
    fi

+    # Test whether lz4 library is installed
+    $CXX $CFLAGS $COMMON_FLAGS -x c++ - -o /dev/null 2>/dev/null  <<EOF
+      #include <lz4.h>
+      #include <lz4hc.h>
+      int main() {}
+EOF
+    if [ "$?" = 0 ]; then
+        COMMON_FLAGS="$COMMON_FLAGS -DLZ4"
+        PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -llz4"
+    fi
+
    # Test whether tcmalloc is available
    $CXX $CFLAGS -x c++ - -o /dev/null -ltcmalloc 2>/dev/null  <<EOF
      int main() {}
@@ -43,6 +43,10 @@ ZLIB_LIBS=" $TOOLCHAIN_LIB_BASE/zlib/zlib-1.2.5/c3f970a/lib/libz.a"
 BZIP_INCLUDE=" -I $TOOLCHAIN_LIB_BASE/bzip2/bzip2-1.0.6/c3f970a/include"
 BZIP_LIBS=" $TOOLCHAIN_LIB_BASE/bzip2/bzip2-1.0.6/c3f970a/lib/libbz2.a"

+LZ4_REV=065ec7e38fe83329031f6668c43bef83eff5808b
+LZ4_INCLUDE=" -I /mnt/gvfs/third-party2/lz4/$LZ4_REV/r108/gcc-4.8.1-glibc-2.17/c3f970a/include"
+LZ4_LIBS=" /mnt/gvfs/third-party2/lz4/$LZ4_REV/r108/gcc-4.8.1-glibc-2.17/c3f970a/lib/liblz4.a"
+
 # location of gflags headers and libraries
 GFLAGS_INCLUDE=" -I $TOOLCHAIN_LIB_BASE/gflags/gflags-1.6/c3f970a/include"
 GFLAGS_LIBS=" $TOOLCHAIN_LIB_BASE/gflags/gflags-1.6/c3f970a/lib/libgflags.a"
@@ -55,22 +59,22 @@ JEMALLOC_LIB=" -Wl,--whole-archive $TOOLCHAIN_LIB_BASE/jemalloc/jemalloc-3.4.1/4
 export USE_SSE=" -msse -msse4.2 "

 CC="$TOOLCHAIN_EXECUTABLES/gcc/gcc-4.8.1/cc6c9dc/bin/gcc"
-CXX="$TOOLCHAIN_EXECUTABLES/gcc/gcc-4.8.1/cc6c9dc/bin/g++ $JINCLUDE $SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $GFLAGS_INCLUDE"
+CXX="$TOOLCHAIN_EXECUTABLES/gcc/gcc-4.8.1/cc6c9dc/bin/g++ $JINCLUDE $SNAPPY_INCLUDE $ZLIB_INCLUDE $BZIP_INCLUDE $LZ4_INCLUDE $GFLAGS_INCLUDE"
 AR=$TOOLCHAIN_EXECUTABLES/binutils/binutils-2.21.1/da39a3e/bin/ar
 RANLIB=$TOOLCHAIN_EXECUTABLES/binutils/binutils-2.21.1/da39a3e/bin/ranlib

 CFLAGS="-B$TOOLCHAIN_EXECUTABLES/binutils/binutils-2.21.1/da39a3e/bin/gold -m64 -mtune=generic"
 CFLAGS+=" $LIBGCC_INCLUDE $GLIBC_INCLUDE"
 CFLAGS+=" -DROCKSDB_PLATFORM_POSIX -DROCKSDB_ATOMIC_PRESENT -DROCKSDB_FALLOCATE_PRESENT"
-CFLAGS+=" -DSNAPPY -DGFLAGS -DZLIB -DBZIP2"
+CFLAGS+=" -DSNAPPY -DGFLAGS -DZLIB -DBZIP2 -DLZ4"

 EXEC_LDFLAGS="-Wl,--dynamic-linker,/usr/local/fbcode/gcc-4.8.1-glibc-2.17/lib/ld.so"
 EXEC_LDFLAGS+=" -Wl,--no-whole-archive $TOOLCHAIN_LIB_BASE/libunwind/libunwind-1.0.1/675d945/lib/libunwind.a"
-EXEC_LDFLAGS+=" $HDFSLIB $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $GFLAGS_LIBS"
+EXEC_LDFLAGS+=" $HDFSLIB $SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $GFLAGS_LIBS"

 PLATFORM_LDFLAGS="$LIBGCC_LIBS $GLIBC_LIBS "

-EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $GFLAGS_LIBS"
+EXEC_LDFLAGS_SHARED="$SNAPPY_LIBS $ZLIB_LIBS $BZIP_LIBS $LZ4_LIBS $GFLAGS_LIBS"

 VALGRIND_VER="$TOOLCHAIN_LIB_BASE/valgrind/valgrind-3.8.1/c3f970a/bin/"

@@ -47,7 +47,6 @@ fi
 #     ln -s `git rev-parse --show-toplevel`/build_tools/format-diff.sh $PRE_COMMIT_SCRIPT_PATH
 #   fi
 # fi
-
 set -e

 uncommitted_code=`git diff HEAD`
@@ -55,7 +54,6 @@ uncommitted_code=`git diff HEAD`
 # If there's no uncommitted changes, we assume user are doing post-commit
 # format check, in which case we'll check the modified lines from latest commit.
 # Otherwise, we'll check format of the uncommitted code only.
-format_last_commit=0
 if [ -z "$uncommitted_code" ]
 then
  # Check the format of last commit
@@ -5,13 +5,18 @@
 #  of patent rights can be found in the PATENTS file in the same directory.

 set -e
+if [ -z "$GIT" ]
+then
+  GIT="git"
+fi
+
 # Print out the colored progress info so that it can be brainlessly 
 # distinguished by users.
 function title() {
  echo -e "\033[1;32m$*\033[0m"
 }

-usage="Create new rocksdb version and prepare it for the release process\n"
+usage="Create new RocksDB version and prepare it for the release process\n"
 usage+="USAGE: ./make_new_version.sh <version>"

 # -- Pre-check
@@ -23,18 +28,19 @@ fi
 ROCKSDB_VERSION=$1

 GIT_BRANCH=`git rev-parse --abbrev-ref HEAD`
+echo $GIT_BRANCH
+
 if [ $GIT_BRANCH != "master" ]; then
  echo "Error: Current branch is '$GIT_BRANCH', Please switch to master branch."
+  exit 1
 fi

 title "Adding new tag for this release ..."
-git tag -a "$ROCKSDB_VERSION.fb" -m "Rocksdb $ROCKSDB_VERSION"
+BRANCH="$ROCKSDB_VERSION.fb"
+$GIT co -b $BRANCH

 # Setting up the proxy for remote repo access
-export http_proxy=http://172.31.255.99:8080
-export https_proxy="$http_proxy";
+title "Pushing new branch to remote repo ..."
+git push origin --set-upstream $BRANCH

-title "Pushing new tag to remote repo ..."
-proxycmd.sh git push origin --tags
-
-title "Done!"
+title "Branch $BRANCH is pushed to github;"
@@ -117,6 +117,27 @@ make release
    --sync=0 \
    --threads=16 > ${STAT_FILE}.readrandom

+# measure readrandom with 6GB block cache and tailing iterator
+./db_bench \
+    --benchmarks=readrandom \
+    --db=$DATA_DIR \
+    --use_existing_db=1 \
+    --bloom_bits=10 \
+    --num=$NUM \
+    --reads=$((NUM / 5)) \
+    --cache_size=6442450944 \
+    --cache_numshardbits=6 \
+    --table_cache_numshardbits=4 \
+    --open_files=55000 \
+    --disable_seek_compaction=1 \
+    --use_tailing_iterator=1 \
+    --statistics=1 \
+    --histogram=1 \
+    --disable_data_sync=1 \
+    --disable_wal=1 \
+    --sync=0 \
+    --threads=16 > ${STAT_FILE}.readrandomtailing
+
 # measure readrandom with 100MB block cache
 ./db_bench \
    --benchmarks=readrandom \
@@ -300,6 +321,7 @@ function send_benchmark_to_ods {
 send_benchmark_to_ods overwrite overwrite $STAT_FILE.overwrite
 send_benchmark_to_ods fillseq fillseq $STAT_FILE.fillseq
 send_benchmark_to_ods readrandom readrandom $STAT_FILE.readrandom
+send_benchmark_to_ods readrandom readrandom_tailing $STAT_FILE.readrandomtailing
 send_benchmark_to_ods readrandom readrandom_smallblockcache $STAT_FILE.readrandomsmallblockcache
 send_benchmark_to_ods readrandom readrandom_memtable_sst $STAT_FILE.readrandom_mem_sst
 send_benchmark_to_ods readrandom readrandom_fillunique_random $STAT_FILE.readrandom_filluniquerandom
@@ -44,6 +44,11 @@ $GCOV --preserve-paths --relative-only --no-output $GCNO_FILES 2>/dev/null |
  tee -a $RECENT_REPORT &&
 echo -e "Generated coverage report for recently updated files: $RECENT_REPORT\n"

+# Unless otherwise specified, we'll not generate html report by default
+if [ -z "$HTML" ]; then
+  exit 0
+fi
+
 # Generate the html report. If we cannot find lcov in this machine, we'll simply
 # skip this step.
 echo "Generating the html coverage report..."
@@ -9,16 +9,16 @@

 #include "db/builder.h"

-#include "db/filename.h"
 #include "db/dbformat.h"
+#include "db/filename.h"
 #include "db/merge_helper.h"
 #include "db/table_cache.h"
 #include "db/version_edit.h"
 #include "rocksdb/db.h"
-#include "rocksdb/table.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
 #include "rocksdb/options.h"
+#include "rocksdb/table.h"
 #include "table/block_based_table_builder.h"
 #include "util/stop_watch.h"

@@ -26,20 +26,18 @@ namespace rocksdb {

 class TableFactory;

-TableBuilder* GetTableBuilder(const Options& options, WritableFile* file,
+TableBuilder* NewTableBuilder(const Options& options,
+                              const InternalKeyComparator& internal_comparator,
+                              WritableFile* file,
                              CompressionType compression_type) {
-  return options.table_factory->GetTableBuilder(options, file,
-                                                compression_type);
+  return options.table_factory->NewTableBuilder(options, internal_comparator,
+                                                file, compression_type);
 }

-Status BuildTable(const std::string& dbname,
-                  Env* env,
-                  const Options& options,
-                  const EnvOptions& soptions,
-                  TableCache* table_cache,
-                  Iterator* iter,
-                  FileMetaData* meta,
-                  const Comparator* user_comparator,
+Status BuildTable(const std::string& dbname, Env* env, const Options& options,
+                  const EnvOptions& soptions, TableCache* table_cache,
+                  Iterator* iter, FileMetaData* meta,
+                  const InternalKeyComparator& internal_comparator,
                  const SequenceNumber newest_snapshot,
                  const SequenceNumber earliest_seqno_in_memtable,
                  const CompressionType compression) {
@@ -64,8 +62,8 @@ Status BuildTable(const std::string& dbname,
      return s;
    }

-    TableBuilder* builder = GetTableBuilder(options, file.get(),
-                                            compression);
+    TableBuilder* builder =
+        NewTableBuilder(options, internal_comparator, file.get(), compression);

    // the first key is the smallest key
    Slice key = iter->key();
@@ -73,8 +71,9 @@ Status BuildTable(const std::string& dbname,
    meta->smallest_seqno = GetInternalKeySeqno(key);
    meta->largest_seqno = meta->smallest_seqno;

-    MergeHelper merge(user_comparator, options.merge_operator.get(),
-                      options.info_log.get(),
+    MergeHelper merge(internal_comparator.user_comparator(),
+                      options.merge_operator.get(), options.info_log.get(),
+                      options.min_partial_merge_operands,
                      true /* internal key corruption is not ok */);

    if (purge) {
@@ -103,8 +102,8 @@ Status BuildTable(const std::string& dbname,
        // If the key is the same as the previous key (and it is not the
        // first key), then we skip it, since it is an older version.
        // Otherwise we output the key and mark it as the "new" previous key.
-        if (!is_first_key && !user_comparator->Compare(prev_ikey.user_key,
-                                                       this_ikey.user_key)) {
+        if (!is_first_key && !internal_comparator.user_comparator()->Compare(
+                                  prev_ikey.user_key, this_ikey.user_key)) {
          // seqno within the same key are in decreasing order
          assert(this_ikey.sequence < prev_ikey.sequence);
        } else {
@@ -202,10 +201,8 @@ Status BuildTable(const std::string& dbname,

    if (s.ok()) {
      // Verify that the table is usable
-      Iterator* it = table_cache->NewIterator(ReadOptions(),
-                                              soptions,
-                                              meta->number,
-                                              meta->file_size);
+      Iterator* it = table_cache->NewIterator(ReadOptions(), soptions,
+                                              internal_comparator, *meta);
      s = it->status();
      delete it;
    }
@@ -24,23 +24,20 @@ class VersionEdit;
 class TableBuilder;
 class WritableFile;

-
-extern TableBuilder* GetTableBuilder(const Options& options, WritableFile* file,
-                                     CompressionType compression_type);
+extern TableBuilder* NewTableBuilder(
+    const Options& options, const InternalKeyComparator& internal_comparator,
+    WritableFile* file, CompressionType compression_type);

 // Build a Table file from the contents of *iter.  The generated file
 // will be named according to meta->number.  On success, the rest of
 // *meta will be filled with metadata about the generated table.
 // If no data is present in *iter, meta->file_size will be set to
 // zero, and no Table file will be produced.
-extern Status BuildTable(const std::string& dbname,
-                         Env* env,
-                         const Options& options,
-                         const EnvOptions& soptions,
-                         TableCache* table_cache,
-                         Iterator* iter,
+extern Status BuildTable(const std::string& dbname, Env* env,
+                         const Options& options, const EnvOptions& soptions,
+                         TableCache* table_cache, Iterator* iter,
                         FileMetaData* meta,
-                         const Comparator* user_comparator,
+                         const InternalKeyComparator& internal_comparator,
                         const SequenceNumber newest_snapshot,
                         const SequenceNumber earliest_seqno_in_memtable,
                         const CompressionType compression);
@@ -7,6 +7,8 @@
 // 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.

+#ifndef ROCKSDB_LITE
+
 #include "rocksdb/c.h"

 #include <stdlib.h>
@@ -17,21 +19,28 @@
 #include "rocksdb/env.h"
 #include "rocksdb/filter_policy.h"
 #include "rocksdb/iterator.h"
+#include "rocksdb/merge_operator.h"
 #include "rocksdb/options.h"
 #include "rocksdb/status.h"
 #include "rocksdb/write_batch.h"
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/universal_compaction.h"
+#include "rocksdb/statistics.h"
+#include "rocksdb/slice_transform.h"
+#include "rocksdb/table.h"

 using rocksdb::Cache;
 using rocksdb::Comparator;
 using rocksdb::CompressionType;
 using rocksdb::DB;
 using rocksdb::Env;
+using rocksdb::InfoLogLevel;
 using rocksdb::FileLock;
 using rocksdb::FilterPolicy;
+using rocksdb::FlushOptions;
 using rocksdb::Iterator;
 using rocksdb::Logger;
+using rocksdb::MergeOperator;
 using rocksdb::NewBloomFilterPolicy;
 using rocksdb::NewLRUCache;
 using rocksdb::Options;
@@ -40,29 +49,33 @@ using rocksdb::Range;
 using rocksdb::ReadOptions;
 using rocksdb::SequentialFile;
 using rocksdb::Slice;
+using rocksdb::SliceTransform;
 using rocksdb::Snapshot;
 using rocksdb::Status;
 using rocksdb::WritableFile;
 using rocksdb::WriteBatch;
 using rocksdb::WriteOptions;
+using rocksdb::LiveFileMetaData;

 using std::shared_ptr;

 extern "C" {

-struct rocksdb_t              { DB*               rep; };
-struct rocksdb_iterator_t     { Iterator*         rep; };
-struct rocksdb_writebatch_t   { WriteBatch        rep; };
-struct rocksdb_snapshot_t     { const Snapshot*   rep; };
-struct rocksdb_readoptions_t  { ReadOptions       rep; };
-struct rocksdb_writeoptions_t { WriteOptions      rep; };
-struct rocksdb_options_t      { Options           rep; };
-struct rocksdb_seqfile_t      { SequentialFile*   rep; };
-struct rocksdb_randomfile_t   { RandomAccessFile* rep; };
-struct rocksdb_writablefile_t { WritableFile*     rep; };
-struct rocksdb_filelock_t     { FileLock*         rep; };
-struct rocksdb_logger_t       { shared_ptr<Logger>  rep; };
-struct rocksdb_cache_t        { shared_ptr<Cache>   rep; };
+struct rocksdb_t                 { DB*               rep; };
+struct rocksdb_iterator_t        { Iterator*         rep; };
+struct rocksdb_writebatch_t      { WriteBatch        rep; };
+struct rocksdb_snapshot_t        { const Snapshot*   rep; };
+struct rocksdb_flushoptions_t    { FlushOptions      rep; };
+struct rocksdb_readoptions_t     { ReadOptions       rep; };
+struct rocksdb_writeoptions_t    { WriteOptions      rep; };
+struct rocksdb_options_t         { Options           rep; };
+struct rocksdb_seqfile_t         { SequentialFile*   rep; };
+struct rocksdb_randomfile_t      { RandomAccessFile* rep; };
+struct rocksdb_writablefile_t    { WritableFile*     rep; };
+struct rocksdb_filelock_t        { FileLock*         rep; };
+struct rocksdb_logger_t          { shared_ptr<Logger>  rep; };
+struct rocksdb_cache_t           { shared_ptr<Cache>   rep; };
+struct rocksdb_livefiles_t       { std::vector<LiveFileMetaData> rep; };

 struct rocksdb_comparator_t : public Comparator {
  void* state_;
@@ -103,6 +116,9 @@ struct rocksdb_filterpolicy_t : public FilterPolicy {
      void*,
      const char* key, size_t length,
      const char* filter, size_t filter_length);
+  void (*delete_filter_)(
+      void*,
+      const char* filter, size_t filter_length);

  virtual ~rocksdb_filterpolicy_t() {
    (*destructor_)(state_);
@@ -122,7 +138,12 @@ struct rocksdb_filterpolicy_t : public FilterPolicy {
    size_t len;
    char* filter = (*create_)(state_, &key_pointers[0], &key_sizes[0], n, &len);
    dst->append(filter, len);
-    free(filter);
+
+    if (delete_filter_ != nullptr) {
+      (*delete_filter_)(state_, filter, len);
+    } else {
+      free(filter);
+    }
  }

  virtual bool KeyMayMatch(const Slice& key, const Slice& filter) const {
@@ -131,21 +152,156 @@ struct rocksdb_filterpolicy_t : public FilterPolicy {
  }
 };

+struct rocksdb_mergeoperator_t : public MergeOperator {
+  void* state_;
+  void (*destructor_)(void*);
+  const char* (*name_)(void*);
+  char* (*full_merge_)(
+      void*,
+      const char* key, size_t key_length,
+      const char* existing_value, size_t existing_value_length,
+      const char* const* operands_list, const size_t* operands_list_length,
+      int num_operands,
+      unsigned char* success, size_t* new_value_length);
+  char* (*partial_merge_)(void*, const char* key, size_t key_length,
+                          const char* const* operands_list,
+                          const size_t* operands_list_length, int num_operands,
+                          unsigned char* success, size_t* new_value_length);
+  void (*delete_value_)(
+      void*,
+      const char* value, size_t value_length);
+
+  virtual ~rocksdb_mergeoperator_t() {
+    (*destructor_)(state_);
+  }
+
+  virtual const char* Name() const {
+    return (*name_)(state_);
+  }
+
+  virtual bool FullMerge(
+      const Slice& key,
+      const Slice* existing_value,
+      const std::deque<std::string>& operand_list,
+      std::string* new_value,
+      Logger* logger) const {
+
+    size_t n = operand_list.size();
+    std::vector<const char*> operand_pointers(n);
+    std::vector<size_t> operand_sizes(n);
+    for (size_t i = 0; i < n; i++) {
+      Slice operand(operand_list[i]);
+      operand_pointers[i] = operand.data();
+      operand_sizes[i] = operand.size();
+    }
+
+    const char* existing_value_data = nullptr;
+    size_t existing_value_len = 0;
+    if (existing_value != nullptr) {
+      existing_value_data = existing_value->data();
+      existing_value_len = existing_value->size();
+    }
+
+    unsigned char success;
+    size_t new_value_len;
+    char* tmp_new_value = (*full_merge_)(
+        state_,
+        key.data(), key.size(),
+        existing_value_data, existing_value_len,
+        &operand_pointers[0], &operand_sizes[0], n,
+        &success, &new_value_len);
+    new_value->assign(tmp_new_value, new_value_len);
+
+    if (delete_value_ != nullptr) {
+      (*delete_value_)(state_, tmp_new_value, new_value_len);
+    } else {
+      free(tmp_new_value);
+    }
+
+    return success;
+  }
+
+  virtual bool PartialMergeMulti(const Slice& key,
+                                 const std::deque<Slice>& operand_list,
+                                 std::string* new_value, Logger* logger) const {
+    size_t operand_count = operand_list.size();
+    std::vector<const char*> operand_pointers(operand_count);
+    std::vector<size_t> operand_sizes(operand_count);
+    for (size_t i = 0; i < operand_count; ++i) {
+      Slice operand(operand_list[i]);
+      operand_pointers[i] = operand.data();
+      operand_sizes[i] = operand.size();
+    }
+
+    unsigned char success;
+    size_t new_value_len;
+    char* tmp_new_value = (*partial_merge_)(
+        state_, key.data(), key.size(), &operand_pointers[0], &operand_sizes[0],
+        operand_count, &success, &new_value_len);
+    new_value->assign(tmp_new_value, new_value_len);
+
+    if (delete_value_ != nullptr) {
+      (*delete_value_)(state_, tmp_new_value, new_value_len);
+    } else {
+      free(tmp_new_value);
+    }
+
+    return success;
+  }
+};
+
 struct rocksdb_env_t {
  Env* rep;
  bool is_default;
 };

+struct rocksdb_slicetransform_t : public SliceTransform {
+  void* state_;
+  void (*destructor_)(void*);
+  const char* (*name_)(void*);
+  char* (*transform_)(
+      void*,
+      const char* key, size_t length,
+      size_t* dst_length);
+  unsigned char (*in_domain_)(
+      void*,
+      const char* key, size_t length);
+  unsigned char (*in_range_)(
+      void*,
+      const char* key, size_t length);
+
+  virtual ~rocksdb_slicetransform_t() {
+    (*destructor_)(state_);
+  }
+
+  virtual const char* Name() const {
+    return (*name_)(state_);
+  }
+
+  virtual Slice Transform(const Slice& src) const {
+    size_t len;
+    char* dst = (*transform_)(state_, src.data(), src.size(), &len);
+    return Slice(dst, len);
+  }
+
+  virtual bool InDomain(const Slice& src) const {
+    return (*in_domain_)(state_, src.data(), src.size());
+  }
+
+  virtual bool InRange(const Slice& src) const {
+    return (*in_range_)(state_, src.data(), src.size());
+  }
+};
+
 struct rocksdb_universal_compaction_options_t {
  rocksdb::CompactionOptionsUniversal *rep;
 };

-
 static bool SaveError(char** errptr, const Status& s) {
-  assert(errptr != NULL);
+  assert(errptr != nullptr);
  if (s.ok()) {
    return false;
-  } else if (*errptr == NULL) {
+  } else if (*errptr == nullptr) {
    *errptr = strdup(s.ToString().c_str());
  } else {
    // TODO(sanjay): Merge with existing error?
@@ -167,7 +323,21 @@ rocksdb_t* rocksdb_open(
    char** errptr) {
  DB* db;
  if (SaveError(errptr, DB::Open(options->rep, std::string(name), &db))) {
-    return NULL;
+    return nullptr;
+  }
+  rocksdb_t* result = new rocksdb_t;
+  result->rep = db;
+  return result;
+}
+
+rocksdb_t* rocksdb_open_for_read_only(
+    const rocksdb_options_t* options,
+    const char* name,
+    unsigned char error_if_log_file_exist,
+    char** errptr) {
+  DB* db;
+  if (SaveError(errptr, DB::OpenForReadOnly(options->rep, std::string(name), &db, error_if_log_file_exist))) {
+    return nullptr;
  }
  rocksdb_t* result = new rocksdb_t;
  result->rep = db;
@@ -197,6 +367,15 @@ void rocksdb_delete(
  SaveError(errptr, db->rep->Delete(options->rep, Slice(key, keylen)));
 }

+void rocksdb_merge(
+    rocksdb_t* db,
+    const rocksdb_writeoptions_t* options,
+    const char* key, size_t keylen,
+    const char* val, size_t vallen,
+    char** errptr) {
+  SaveError(errptr,
+            db->rep->Merge(options->rep, Slice(key, keylen), Slice(val, vallen)));
+}

 void rocksdb_write(
    rocksdb_t* db,
@@ -212,7 +391,7 @@ char* rocksdb_get(
    const char* key, size_t keylen,
    size_t* vallen,
    char** errptr) {
-  char* result = NULL;
+  char* result = nullptr;
  std::string tmp;
  Status s = db->rep->Get(options->rep, Slice(key, keylen), &tmp);
  if (s.ok()) {
@@ -257,7 +436,7 @@ char* rocksdb_property_value(
    // We use strdup() since we expect human readable output.
    return strdup(tmp.c_str());
  } else {
-    return NULL;
+    return nullptr;
  }
 }

@@ -276,15 +455,48 @@ void rocksdb_approximate_sizes(
  delete[] ranges;
 }

+void rocksdb_delete_file(
+    rocksdb_t* db,
+    const char* name) {
+  db->rep->DeleteFile(name);
+}
+
+const rocksdb_livefiles_t* rocksdb_livefiles(
+    rocksdb_t* db) {
+  rocksdb_livefiles_t* result = new rocksdb_livefiles_t;
+  db->rep->GetLiveFilesMetaData(&result->rep);
+  return result;
+}
+
 void rocksdb_compact_range(
    rocksdb_t* db,
    const char* start_key, size_t start_key_len,
    const char* limit_key, size_t limit_key_len) {
  Slice a, b;
  db->rep->CompactRange(
-      // Pass NULL Slice if corresponding "const char*" is NULL
-      (start_key ? (a = Slice(start_key, start_key_len), &a) : NULL),
-      (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : NULL));
+      // Pass nullptr Slice if corresponding "const char*" is nullptr
+      (start_key ? (a = Slice(start_key, start_key_len), &a) : nullptr),
+      (limit_key ? (b = Slice(limit_key, limit_key_len), &b) : nullptr));
+}
+
+void rocksdb_flush(
+    rocksdb_t* db,
+    const rocksdb_flushoptions_t* options,
+    char** errptr) {
+  SaveError(errptr, db->rep->Flush(options->rep));
+}
+
+void rocksdb_disable_file_deletions(
+    rocksdb_t* db,
+    char** errptr) {
+  SaveError(errptr, db->rep->DisableFileDeletions());
+}
+
+void rocksdb_enable_file_deletions(
+    rocksdb_t* db,
+    unsigned char force,
+    char** errptr) {
+  SaveError(errptr, db->rep->EnableFileDeletions(force));
 }

 void rocksdb_destroy_db(
@@ -358,6 +570,10 @@ void rocksdb_writebatch_clear(rocksdb_writebatch_t* b) {
  b->rep.Clear();
 }

+int rocksdb_writebatch_count(rocksdb_writebatch_t* b) {
+  return b->rep.Count();
+}
+
 void rocksdb_writebatch_put(
    rocksdb_writebatch_t* b,
    const char* key, size_t klen,
@@ -365,6 +581,13 @@ void rocksdb_writebatch_put(
  b->rep.Put(Slice(key, klen), Slice(val, vlen));
 }

+void rocksdb_writebatch_merge(
+    rocksdb_writebatch_t* b,
+    const char* key, size_t klen,
+    const char* val, size_t vlen) {
+  b->rep.Merge(Slice(key, klen), Slice(val, vlen));
+}
+
 void rocksdb_writebatch_delete(
    rocksdb_writebatch_t* b,
    const char* key, size_t klen) {
@@ -395,6 +618,11 @@ void rocksdb_writebatch_iterate(
  b->rep.Iterate(&handler);
 }

+const char* rocksdb_writebatch_data(rocksdb_writebatch_t* b, size_t* size) {
+  *size = b->rep.GetDataSize();
+  return b->rep.Data().c_str();
+}
+
 rocksdb_options_t* rocksdb_options_create() {
  return new rocksdb_options_t;
 }
@@ -409,6 +637,12 @@ void rocksdb_options_set_comparator(
  opt->rep.comparator = cmp;
 }

+void rocksdb_options_set_merge_operator(
+    rocksdb_options_t* opt,
+    rocksdb_mergeoperator_t* merge_operator) {
+  opt->rep.merge_operator = std::shared_ptr<MergeOperator>(merge_operator);
+}
+
 void rocksdb_options_set_filter_policy(
    rocksdb_options_t* opt,
    rocksdb_filterpolicy_t* policy) {
@@ -431,7 +665,7 @@ void rocksdb_options_set_paranoid_checks(
 }

 void rocksdb_options_set_env(rocksdb_options_t* opt, rocksdb_env_t* env) {
-  opt->rep.env = (env ? env->rep : NULL);
+  opt->rep.env = (env ? env->rep : nullptr);
 }

 void rocksdb_options_set_info_log(rocksdb_options_t* opt, rocksdb_logger_t* l) {
@@ -440,6 +674,11 @@ void rocksdb_options_set_info_log(rocksdb_options_t* opt, rocksdb_logger_t* l) {
  }
 }

+void rocksdb_options_set_info_log_level(
+    rocksdb_options_t* opt, int v) {
+  opt->rep.info_log_level = static_cast<InfoLogLevel>(v);
+}
+
 void rocksdb_options_set_write_buffer_size(rocksdb_options_t* opt, size_t s) {
  opt->rep.write_buffer_size = s;
 }
@@ -454,6 +693,12 @@ void rocksdb_options_set_cache(rocksdb_options_t* opt, rocksdb_cache_t* c) {
  }
 }

+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;
 }
@@ -492,6 +737,18 @@ void rocksdb_options_set_max_grandparent_overlap_factor(
  opt->rep.max_grandparent_overlap_factor = n;
 }

+void rocksdb_options_set_max_bytes_for_level_multiplier_additional(
+    rocksdb_options_t* opt, int* level_values, size_t num_levels) {
+  opt->rep.max_bytes_for_level_multiplier_additional.resize(num_levels);
+  for (size_t i = 0; i < num_levels; ++i) {
+    opt->rep.max_bytes_for_level_multiplier_additional[i] = level_values[i];
+  }
+}
+
+void rocksdb_options_enable_statistics(rocksdb_options_t* opt) {
+  opt->rep.statistics = rocksdb::CreateDBStatistics();
+}
+
 void rocksdb_options_set_num_levels(rocksdb_options_t* opt, int n) {
  opt->rep.num_levels = n;
 }
@@ -537,6 +794,16 @@ void rocksdb_options_set_compression_options(
  opt->rep.compression_opts.strategy = strategy;
 }

+void rocksdb_options_set_prefix_extractor(
+    rocksdb_options_t* opt, rocksdb_slicetransform_t* 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;
@@ -557,6 +824,11 @@ void rocksdb_options_set_db_log_dir(
  opt->rep.db_log_dir = db_log_dir;
 }

+void rocksdb_options_set_wal_dir(
+    rocksdb_options_t* opt, const char* v) {
+  opt->rep.wal_dir = v;
+}
+
 void rocksdb_options_set_WAL_ttl_seconds(rocksdb_options_t* opt, uint64_t ttl) {
  opt->rep.WAL_ttl_seconds = ttl;
 }
@@ -566,6 +838,99 @@ void rocksdb_options_set_WAL_size_limit_MB(
  opt->rep.WAL_size_limit_MB = limit;
 }

+void rocksdb_options_set_manifest_preallocation_size(
+    rocksdb_options_t* opt, size_t v) {
+  opt->rep.manifest_preallocation_size = v;
+}
+
+void rocksdb_options_set_purge_redundant_kvs_while_flush(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.purge_redundant_kvs_while_flush = v;
+}
+
+void rocksdb_options_set_allow_os_buffer(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.allow_os_buffer = v;
+}
+
+void rocksdb_options_set_allow_mmap_reads(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.allow_mmap_reads = v;
+}
+
+void rocksdb_options_set_allow_mmap_writes(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.allow_mmap_writes = v;
+}
+
+void rocksdb_options_set_is_fd_close_on_exec(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.is_fd_close_on_exec = v;
+}
+
+void rocksdb_options_set_skip_log_error_on_recovery(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.skip_log_error_on_recovery = v;
+}
+
+void rocksdb_options_set_stats_dump_period_sec(
+    rocksdb_options_t* opt, unsigned int v) {
+  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;
+}
+
+void rocksdb_options_set_access_hint_on_compaction_start(
+    rocksdb_options_t* opt, int v) {
+  switch(v) {
+    case 0:
+      opt->rep.access_hint_on_compaction_start = rocksdb::Options::NONE;
+      break;
+    case 1:
+      opt->rep.access_hint_on_compaction_start = rocksdb::Options::NORMAL;
+      break;
+    case 2:
+      opt->rep.access_hint_on_compaction_start = rocksdb::Options::SEQUENTIAL;
+      break;
+    case 3:
+      opt->rep.access_hint_on_compaction_start = rocksdb::Options::WILLNEED;
+      break;
+  }
+}
+
+void rocksdb_options_set_use_adaptive_mutex(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.use_adaptive_mutex = v;
+}
+
+void rocksdb_options_set_bytes_per_sync(
+    rocksdb_options_t* opt, uint64_t v) {
+  opt->rep.bytes_per_sync = v;
+}
+
+void rocksdb_options_set_verify_checksums_in_compaction(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.verify_checksums_in_compaction = v;
+}
+
+void rocksdb_options_set_filter_deletes(
+    rocksdb_options_t* opt, unsigned char v) {
+  opt->rep.filter_deletes = v;
+}
+
+void rocksdb_options_set_max_sequential_skip_in_iterations(
+    rocksdb_options_t* opt, uint64_t v) {
+  opt->rep.max_sequential_skip_in_iterations = v;
+}
+
 void rocksdb_options_set_max_write_buffer_number(rocksdb_options_t* opt, int n) {
  opt->rep.max_write_buffer_number = n;
 }
@@ -582,6 +947,56 @@ void rocksdb_options_set_max_background_flushes(rocksdb_options_t* opt, int n) {
  opt->rep.max_background_flushes = n;
 }

+void rocksdb_options_set_max_log_file_size(rocksdb_options_t* opt, size_t v) {
+  opt->rep.max_log_file_size = v;
+}
+
+void rocksdb_options_set_log_file_time_to_roll(rocksdb_options_t* opt, size_t v) {
+  opt->rep.log_file_time_to_roll = v;
+}
+
+void rocksdb_options_set_keep_log_file_num(rocksdb_options_t* opt, size_t v) {
+  opt->rep.keep_log_file_num = v;
+}
+
+void rocksdb_options_set_soft_rate_limit(rocksdb_options_t* opt, double v) {
+  opt->rep.soft_rate_limit = v;
+}
+
+void rocksdb_options_set_hard_rate_limit(rocksdb_options_t* opt, double v) {
+  opt->rep.hard_rate_limit = v;
+}
+
+void rocksdb_options_set_rate_limit_delay_max_milliseconds(
+    rocksdb_options_t* opt, unsigned int v) {
+  opt->rep.rate_limit_delay_max_milliseconds = v;
+}
+
+void rocksdb_options_set_max_manifest_file_size(
+    rocksdb_options_t* opt, size_t v) {
+  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;
+}
+
+void rocksdb_options_set_table_cache_remove_scan_count_limit(
+    rocksdb_options_t* opt, int v) {
+  opt->rep.table_cache_remove_scan_count_limit = v;
+}
+
+void rocksdb_options_set_arena_block_size(
+    rocksdb_options_t* opt, size_t v) {
+  opt->rep.arena_block_size = v;
+}
+
 void rocksdb_options_set_disable_auto_compactions(rocksdb_options_t* opt, int disable) {
  opt->rep.disable_auto_compactions = disable;
 }
@@ -590,6 +1005,11 @@ void rocksdb_options_set_disable_seek_compaction(rocksdb_options_t* opt, int dis
  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;
+}
+
 void rocksdb_options_set_source_compaction_factor(
    rocksdb_options_t* opt, int n) {
  opt->rep.expanded_compaction_factor = n;
@@ -607,6 +1027,78 @@ void rocksdb_options_set_memtable_vector_rep(rocksdb_options_t *opt) {
  opt->rep.memtable_factory.reset(factory);
 }

+void rocksdb_options_set_memtable_prefix_bloom_bits(
+    rocksdb_options_t* opt, uint32_t v) {
+  opt->rep.memtable_prefix_bloom_bits = v;
+}
+
+void rocksdb_options_set_memtable_prefix_bloom_probes(
+    rocksdb_options_t* opt, uint32_t v) {
+  opt->rep.memtable_prefix_bloom_probes = v;
+}
+
+void rocksdb_options_set_hash_skip_list_rep(
+    rocksdb_options_t *opt, size_t bucket_count,
+    int32_t skiplist_height, int32_t skiplist_branching_factor) {
+  static rocksdb::MemTableRepFactory* factory = 0;
+  if (!factory) {
+    factory = rocksdb::NewHashSkipListRepFactory(
+        bucket_count, skiplist_height, skiplist_branching_factor);
+  }
+  opt->rep.memtable_factory.reset(factory);
+}
+
+void rocksdb_options_set_hash_link_list_rep(
+    rocksdb_options_t *opt, size_t bucket_count) {
+  static rocksdb::MemTableRepFactory* factory = 0;
+  if (!factory) {
+    factory = rocksdb::NewHashLinkListRepFactory(bucket_count);
+  }
+  opt->rep.memtable_factory.reset(factory);
+}
+
+void rocksdb_options_set_plain_table_factory(
+    rocksdb_options_t *opt, uint32_t user_key_len, int bloom_bits_per_key,
+    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);
+  }
+  opt->rep.table_factory.reset(factory);
+}
+
+void rocksdb_options_set_max_successive_merges(
+    rocksdb_options_t* opt, size_t v) {
+  opt->rep.max_successive_merges = v;
+}
+
+void rocksdb_options_set_min_partial_merge_operands(
+    rocksdb_options_t* opt, uint32_t v) {
+  opt->rep.min_partial_merge_operands = v;
+}
+
+void rocksdb_options_set_bloom_locality(
+    rocksdb_options_t* opt, uint32_t v) {
+  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;
+}
+
+void rocksdb_options_set_inplace_update_num_locks(
+    rocksdb_options_t* opt, size_t v) {
+  opt->rep.inplace_update_num_locks = v;
+}
+
 void rocksdb_options_set_compaction_style(rocksdb_options_t *opt, int style) {
  opt->rep.compaction_style = static_cast<rocksdb::CompactionStyle>(style);
 }
@@ -617,42 +1109,18 @@ void rocksdb_options_set_universal_compaction_options(rocksdb_options_t *opt, ro

 /*
 TODO:
-merge_operator
+DB::OpenForReadOnly
+DB::MultiGet
+DB::KeyMayExist
+DB::GetOptions
+DB::GetSortedWalFiles
+DB::GetLatestSequenceNumber
+DB::GetUpdatesSince
+DB::GetDbIdentity
+DB::RunManualCompaction
+custom cache
 compaction_filter
-prefix_extractor
-whole_key_filtering
-max_bytes_for_level_multiplier_additional
-delete_obsolete_files_period_micros
-max_log_file_size
-log_file_time_to_roll
-keep_log_file_num
-soft_rate_limit
-hard_rate_limit
-rate_limit_delay_max_milliseconds
-max_manifest_file_size
-no_block_cache
-table_cache_numshardbits
-table_cache_remove_scan_count_limit
-arena_block_size
-manifest_preallocation_size
-purge_redundant_kvs_while_flush
-allow_os_buffer
-allow_mmap_reads
-allow_mmap_writes
-is_fd_close_on_exec
-skip_log_error_on_recovery
-stats_dump_period_sec
-block_size_deviation
-advise_random_on_open
-access_hint_on_compaction_start
-use_adaptive_mutex
-bytes_per_sync
-filter_deletes
-max_sequential_skip_in_iterations
-table_factory
 table_properties_collectors
-inplace_update_support
-inplace_update_num_locks
 */

 rocksdb_comparator_t* rocksdb_comparator_create(
@@ -687,12 +1155,16 @@ rocksdb_filterpolicy_t* rocksdb_filterpolicy_create(
        void*,
        const char* key, size_t length,
        const char* filter, size_t filter_length),
+    void (*delete_filter)(
+        void*,
+        const char* filter, size_t filter_length),
    const char* (*name)(void*)) {
  rocksdb_filterpolicy_t* result = new rocksdb_filterpolicy_t;
  result->state_ = state;
  result->destructor_ = destructor;
  result->create_ = create_filter;
  result->key_match_ = key_may_match;
+  result->delete_filter_ = delete_filter;
  result->name_ = name;
  return result;
 }
@@ -719,11 +1191,40 @@ rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(int bits_per_key) {
  };
  Wrapper* wrapper = new Wrapper;
  wrapper->rep_ = NewBloomFilterPolicy(bits_per_key);
-  wrapper->state_ = NULL;
+  wrapper->state_ = nullptr;
+  wrapper->delete_filter_ = nullptr;
  wrapper->destructor_ = &Wrapper::DoNothing;
  return wrapper;
 }

+rocksdb_mergeoperator_t* rocksdb_mergeoperator_create(
+    void* state, void (*destructor)(void*),
+    char* (*full_merge)(void*, const char* key, size_t key_length,
+                        const char* existing_value,
+                        size_t existing_value_length,
+                        const char* const* operands_list,
+                        const size_t* operands_list_length, int num_operands,
+                        unsigned char* success, size_t* new_value_length),
+    char* (*partial_merge)(void*, const char* key, size_t key_length,
+                           const char* const* operands_list,
+                           const size_t* operands_list_length, int num_operands,
+                           unsigned char* success, size_t* new_value_length),
+    void (*delete_value)(void*, const char* value, size_t value_length),
+    const char* (*name)(void*)) {
+  rocksdb_mergeoperator_t* result = new rocksdb_mergeoperator_t;
+  result->state_ = state;
+  result->destructor_ = destructor;
+  result->full_merge_ = full_merge;
+  result->partial_merge_ = partial_merge;
+  result->delete_value_ = delete_value;
+  result->name_ = name;
+  return result;
+}
+
+void rocksdb_mergeoperator_destroy(rocksdb_mergeoperator_t* merge_operator) {
+  delete merge_operator;
+}
+
 rocksdb_readoptions_t* rocksdb_readoptions_create() {
  return new rocksdb_readoptions_t;
 }
@@ -746,7 +1247,17 @@ void rocksdb_readoptions_set_fill_cache(
 void rocksdb_readoptions_set_snapshot(
    rocksdb_readoptions_t* opt,
    const rocksdb_snapshot_t* snap) {
-  opt->rep.snapshot = (snap ? snap->rep : NULL);
+  opt->rep.snapshot = (snap ? snap->rep : nullptr);
+}
+
+void rocksdb_readoptions_set_read_tier(
+    rocksdb_readoptions_t* opt, int v) {
+  opt->rep.read_tier = static_cast<rocksdb::ReadTier>(v);
+}
+
+void rocksdb_readoptions_set_tailing(
+    rocksdb_readoptions_t* opt, unsigned char v) {
+  opt->rep.tailing = v;
 }

 rocksdb_writeoptions_t* rocksdb_writeoptions_create() {
@@ -767,6 +1278,19 @@ void rocksdb_writeoptions_disable_WAL(rocksdb_writeoptions_t* opt, int disable)
 }


+rocksdb_flushoptions_t* rocksdb_flushoptions_create() {
+  return new rocksdb_flushoptions_t;
+}
+
+void rocksdb_flushoptions_destroy(rocksdb_flushoptions_t* opt) {
+  delete opt;
+}
+
+void rocksdb_flushoptions_set_wait(
+    rocksdb_flushoptions_t* opt, unsigned char v) {
+  opt->rep.wait = v;
+}
+
 rocksdb_cache_t* rocksdb_cache_create_lru(size_t capacity) {
  rocksdb_cache_t* c = new rocksdb_cache_t;
  c->rep = NewLRUCache(capacity);
@@ -797,6 +1321,57 @@ void rocksdb_env_destroy(rocksdb_env_t* env) {
  delete env;
 }

+rocksdb_slicetransform_t* rocksdb_slicetransform_create(
+    void* state,
+    void (*destructor)(void*),
+    char* (*transform)(
+        void*,
+        const char* key, size_t length,
+        size_t* dst_length),
+    unsigned char (*in_domain)(
+        void*,
+        const char* key, size_t length),
+    unsigned char (*in_range)(
+        void*,
+        const char* key, size_t length),
+    const char* (*name)(void*)) {
+  rocksdb_slicetransform_t* result = new rocksdb_slicetransform_t;
+  result->state_ = state;
+  result->destructor_ = destructor;
+  result->transform_ = transform;
+  result->in_domain_ = in_domain;
+  result->in_range_ = in_range;
+  result->name_ = name;
+  return result;
+}
+
+void rocksdb_slicetransform_destroy(rocksdb_slicetransform_t* st) {
+  delete st;
+}
+
+rocksdb_slicetransform_t* rocksdb_slicetransform_create_fixed_prefix(size_t prefixLen) {
+  struct Wrapper : public rocksdb_slicetransform_t {
+    const SliceTransform* rep_;
+    ~Wrapper() { delete rep_; }
+    const char* Name() const { return rep_->Name(); }
+    Slice Transform(const Slice& src) const {
+      return rep_->Transform(src);
+    }
+    bool InDomain(const Slice& src) const {
+      return rep_->InDomain(src);
+    }
+    bool InRange(const Slice& src) const {
+      return rep_->InRange(src);
+    }
+    static void DoNothing(void*) { }
+  };
+  Wrapper* wrapper = new Wrapper;
+  wrapper->rep_ = rocksdb::NewFixedPrefixTransform(prefixLen);
+  wrapper->state_ = nullptr;
+  wrapper->destructor_ = &Wrapper::DoNothing;
+  return wrapper;
+}
+
 rocksdb_universal_compaction_options_t* rocksdb_universal_compaction_options_create() {
  rocksdb_universal_compaction_options_t* result = new rocksdb_universal_compaction_options_t;
  result->rep = new rocksdb::CompactionOptionsUniversal;
@@ -839,4 +1414,63 @@ void rocksdb_universal_compaction_options_destroy(
  delete uco;
 }

+void rocksdb_options_set_min_level_to_compress(rocksdb_options_t* opt, int level) {
+  if (level >= 0) {
+    assert(level <= opt->rep.num_levels);
+    opt->rep.compression_per_level.resize(opt->rep.num_levels);
+    for (int i = 0; i < level; i++) {
+      opt->rep.compression_per_level[i] = rocksdb::kNoCompression;
+    }
+    for (int i = level; i < opt->rep.num_levels; i++) {
+      opt->rep.compression_per_level[i] = opt->rep.compression;
+    }
+  }
+}
+
+int rocksdb_livefiles_count(
+  const rocksdb_livefiles_t* lf) {
+  return lf->rep.size();
+}
+
+const char* rocksdb_livefiles_name(
+  const rocksdb_livefiles_t* lf,
+  int index) {
+  return lf->rep[index].name.c_str();
+}
+
+int rocksdb_livefiles_level(
+  const rocksdb_livefiles_t* lf,
+  int index) {
+  return lf->rep[index].level;
+}
+
+size_t rocksdb_livefiles_size(
+  const rocksdb_livefiles_t* lf,
+  int index) {
+  return lf->rep[index].size;
+}
+
+const char* rocksdb_livefiles_smallestkey(
+  const rocksdb_livefiles_t* lf,
+  int index,
+  size_t* size) {
+  *size = lf->rep[index].smallestkey.size();
+  return lf->rep[index].smallestkey.data();
+}
+
+const char* rocksdb_livefiles_largestkey(
+  const rocksdb_livefiles_t* lf,
+  int index,
+  size_t* size) {
+  *size = lf->rep[index].largestkey.size();
+  return lf->rep[index].largestkey.data();
+}
+
+extern void rocksdb_livefiles_destroy(
+  const rocksdb_livefiles_t* lf) {
+  delete lf;
+}
+
 }  // end extern "C"
+
+#endif  // ROCKSDB_LITE
@@ -145,7 +145,7 @@ static char* FilterCreate(
  memcpy(result, "fake", 4);
  return result;
 }
-unsigned char FilterKeyMatch(
+static unsigned char FilterKeyMatch(
    void* arg,
    const char* key, size_t length,
    const char* filter, size_t filter_length) {
@@ -154,6 +154,37 @@ unsigned char FilterKeyMatch(
  return fake_filter_result;
 }

+// Custom merge operator
+static void MergeOperatorDestroy(void* arg) { }
+static const char* MergeOperatorName(void* arg) {
+  return "TestMergeOperator";
+}
+static char* MergeOperatorFullMerge(
+    void* arg,
+    const char* key, size_t key_length,
+    const char* existing_value, size_t existing_value_length,
+    const char* const* operands_list, const size_t* operands_list_length,
+    int num_operands,
+    unsigned char* success, size_t* new_value_length) {
+  *new_value_length = 4;
+  *success = 1;
+  char* result = malloc(4);
+  memcpy(result, "fake", 4);
+  return result;
+}
+static char* MergeOperatorPartialMerge(
+    void* arg,
+    const char* key, size_t key_length,
+    const char* const* operands_list, const size_t* operands_list_length,
+    int num_operands,
+    unsigned char* success, size_t* new_value_length) {
+  *new_value_length = 4;
+  *success = 1;
+  char* result = malloc(4);
+  memcpy(result, "fake", 4);
+  return result;
+}
+
 int main(int argc, char** argv) {
  rocksdb_t* db;
  rocksdb_comparator_t* cmp;
@@ -342,7 +373,7 @@ int main(int argc, char** argv) {
    rocksdb_filterpolicy_t* policy;
    if (run == 0) {
      policy = rocksdb_filterpolicy_create(
-          NULL, FilterDestroy, FilterCreate, FilterKeyMatch, FilterName);
+          NULL, FilterDestroy, FilterCreate, FilterKeyMatch, NULL, FilterName);
    } else {
      policy = rocksdb_filterpolicy_create_bloom(10);
    }
@@ -376,6 +407,79 @@ int main(int argc, char** argv) {
    rocksdb_filterpolicy_destroy(policy);
  }

+  StartPhase("merge_operator");
+  {
+    rocksdb_mergeoperator_t* merge_operator;
+    merge_operator = rocksdb_mergeoperator_create(
+        NULL, MergeOperatorDestroy, MergeOperatorFullMerge,
+        MergeOperatorPartialMerge, NULL, MergeOperatorName);
+    // Create new database
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbname, &err);
+    rocksdb_options_set_merge_operator(options, merge_operator);
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo", 3, "foovalue", 8, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "foo", "foovalue");
+    rocksdb_merge(db, woptions, "foo", 3, "barvalue", 8, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "foo", "fake");
+
+    // Merge of a non-existing value
+    rocksdb_merge(db, woptions, "bar", 3, "barvalue", 8, &err);
+    CheckNoError(err);
+    CheckGet(db, roptions, "bar", "fake");
+
+  }
+
+  StartPhase("prefix");
+  {
+    // Create new database
+    rocksdb_close(db);
+    rocksdb_destroy_db(options, dbname, &err);
+
+    rocksdb_filterpolicy_t* policy = rocksdb_filterpolicy_create_bloom(10);
+    rocksdb_options_set_filter_policy(options, policy);
+    rocksdb_options_set_prefix_extractor(options, rocksdb_slicetransform_create_fixed_prefix(3));
+    rocksdb_options_set_hash_skip_list_rep(options, 50000, 4, 4);
+    rocksdb_options_set_plain_table_factory(options, 4, 10, 0.75, 16);
+
+    db = rocksdb_open(options, dbname, &err);
+    CheckNoError(err);
+
+    rocksdb_put(db, woptions, "foo1", 4, "foo", 3, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo2", 4, "foo", 3, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "foo3", 4, "foo", 3, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "bar1", 4, "bar", 3, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "bar2", 4, "bar", 3, &err);
+    CheckNoError(err);
+    rocksdb_put(db, woptions, "bar3", 4, "bar", 3, &err);
+    CheckNoError(err);
+
+    rocksdb_iterator_t* iter = rocksdb_create_iterator(db, roptions);
+    CheckCondition(!rocksdb_iter_valid(iter));
+
+    rocksdb_iter_seek(iter, "bar", 3);
+    rocksdb_iter_get_error(iter, &err);
+    CheckNoError(err);
+    CheckCondition(rocksdb_iter_valid(iter));
+
+    CheckIter(iter, "bar1", "bar");
+    rocksdb_iter_next(iter);
+    CheckIter(iter, "bar2", "bar");
+    rocksdb_iter_next(iter);
+    CheckIter(iter, "bar3", "bar");
+    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);
@@ -0,0 +1,583 @@
+//  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/column_family.h"
+
+#include <vector>
+#include <string>
+#include <algorithm>
+
+#include "db/db_impl.h"
+#include "db/version_set.h"
+#include "db/internal_stats.h"
+#include "db/compaction_picker.h"
+#include "db/table_properties_collector.h"
+#include "util/autovector.h"
+#include "util/hash_skiplist_rep.h"
+
+namespace rocksdb {
+
+ColumnFamilyHandleImpl::ColumnFamilyHandleImpl(ColumnFamilyData* cfd,
+                                               DBImpl* db, port::Mutex* mutex)
+    : cfd_(cfd), db_(db), mutex_(mutex) {
+  if (cfd_ != nullptr) {
+    cfd_->Ref();
+  }
+}
+
+ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
+  if (cfd_ != nullptr) {
+    DBImpl::DeletionState deletion_state;
+    mutex_->Lock();
+    if (cfd_->Unref()) {
+      delete cfd_;
+    }
+    db_->FindObsoleteFiles(deletion_state, false, true);
+    mutex_->Unlock();
+    if (deletion_state.HaveSomethingToDelete()) {
+      db_->PurgeObsoleteFiles(deletion_state);
+    }
+  }
+}
+
+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;
+}
+}  // 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);
+#else
+  ClipToRange(&result.write_buffer_size,
+              ((size_t)64) << 10, ((size_t)64) << 30);
+#endif
+  // if user sets arena_block_size, we trust user to use this value. Otherwise,
+  // calculate a proper value from writer_buffer_size;
+  if (result.arena_block_size <= 0) {
+    result.arena_block_size = result.write_buffer_size / 10;
+  }
+  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;
+  }
+  if (result.soft_rate_limit > result.hard_rate_limit) {
+    result.soft_rate_limit = result.hard_rate_limit;
+  }
+  if (!result.prefix_extractor) {
+    assert(result.memtable_factory);
+    Slice name = result.memtable_factory->Name();
+    if (name.compare("HashSkipListRepFactory") == 0 ||
+        name.compare("HashLinkListRepFactory") == 0) {
+      result.memtable_factory = std::make_shared<SkipListFactory>();
+    }
+  }
+
+  // -- Sanitize the table properties collector
+  // All user defined properties collectors will be wrapped by
+  // UserKeyTablePropertiesCollector since for them they only have the
+  // knowledge of the user keys; internal keys are invisible to them.
+  auto& collectors = result.table_properties_collectors;
+  for (size_t i = 0; i < result.table_properties_collectors.size(); ++i) {
+    assert(collectors[i]);
+    collectors[i] =
+        std::make_shared<UserKeyTablePropertiesCollector>(collectors[i]);
+  }
+  // Add collector to collect internal key statistics
+  collectors.push_back(std::make_shared<InternalKeyPropertiesCollector>());
+
+  return result;
+}
+
+int SuperVersion::dummy = 0;
+void* const SuperVersion::kSVInUse = &SuperVersion::dummy;
+void* const SuperVersion::kSVObsolete = nullptr;
+
+SuperVersion::~SuperVersion() {
+  for (auto td : to_delete) {
+    delete td;
+  }
+}
+
+SuperVersion* SuperVersion::Ref() {
+  refs.fetch_add(1, std::memory_order_relaxed);
+  return this;
+}
+
+bool SuperVersion::Unref() {
+  // fetch_sub returns the previous value of ref
+  uint32_t previous_refs = refs.fetch_sub(1, std::memory_order_relaxed);
+  assert(previous_refs > 0);
+  return previous_refs == 1;
+}
+
+void SuperVersion::Cleanup() {
+  assert(refs.load(std::memory_order_relaxed) == 0);
+  imm->Unref(&to_delete);
+  MemTable* m = mem->Unref();
+  if (m != nullptr) {
+    to_delete.push_back(m);
+  }
+  current->Unref();
+}
+
+void SuperVersion::Init(MemTable* new_mem, MemTableListVersion* new_imm,
+                        Version* new_current) {
+  mem = new_mem;
+  imm = new_imm;
+  current = new_current;
+  mem->Ref();
+  imm->Ref();
+  current->Ref();
+  refs.store(1, std::memory_order_relaxed);
+}
+
+namespace {
+void SuperVersionUnrefHandle(void* ptr) {
+  // UnrefHandle is called when a thread exists or a ThreadLocalPtr gets
+  // destroyed. When former happens, the thread shouldn't see kSVInUse.
+  // When latter happens, we are in ~ColumnFamilyData(), no get should happen as
+  // well.
+  SuperVersion* sv = static_cast<SuperVersion*>(ptr);
+  if (sv->Unref()) {
+    sv->db_mutex->Lock();
+    sv->Cleanup();
+    sv->db_mutex->Unlock();
+    delete sv;
+  }
+}
+}  // anonymous namespace
+
+ColumnFamilyData::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,
+                                   ColumnFamilySet* column_family_set)
+    : id_(id),
+      name_(name),
+      dummy_versions_(dummy_versions),
+      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)),
+      mem_(nullptr),
+      imm_(options.min_write_buffer_number_to_merge),
+      super_version_(nullptr),
+      super_version_number_(0),
+      local_sv_(new ThreadLocalPtr(&SuperVersionUnrefHandle)),
+      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(dbname, &options_, storage_options, table_cache));
+    if (options_.compaction_style == kCompactionStyleUniversal) {
+      compaction_picker_.reset(
+          new UniversalCompactionPicker(&options_, &internal_comparator_));
+    } else {
+      compaction_picker_.reset(
+          new LevelCompactionPicker(&options_, &internal_comparator_));
+    }
+
+    Log(options_.info_log, "Options for column family \"%s\":\n",
+        name.c_str());
+    const ColumnFamilyOptions* cf_options = &options_;
+    cf_options->Dump(options_.info_log.get());
+  }
+}
+
+// DB mutex held
+ColumnFamilyData::~ColumnFamilyData() {
+  assert(refs_ == 0);
+  // remove from linked list
+  auto prev = prev_;
+  auto next = next_;
+  prev->next_ = next;
+  next->prev_ = prev;
+
+  // it's nullptr for dummy CFD
+  if (column_family_set_ != nullptr) {
+    // remove from column_family_set
+    column_family_set_->RemoveColumnFamily(this);
+  }
+
+  if (current_ != nullptr) {
+    current_->Unref();
+  }
+
+  if (super_version_ != nullptr) {
+    // Release SuperVersion reference kept in ThreadLocalPtr.
+    // This must be done outside of mutex_ since unref handler can lock mutex.
+    super_version_->db_mutex->Unlock();
+    local_sv_.reset();
+    super_version_->db_mutex->Lock();
+
+    bool is_last_reference __attribute__((unused));
+    is_last_reference = super_version_->Unref();
+    assert(is_last_reference);
+    super_version_->Cleanup();
+    delete super_version_;
+    super_version_ = nullptr;
+  }
+
+  if (dummy_versions_ != nullptr) {
+    // List must be empty
+    assert(dummy_versions_->next_ == dummy_versions_);
+    delete dummy_versions_;
+  }
+
+  if (mem_ != nullptr) {
+    delete mem_->Unref();
+  }
+  autovector<MemTable*> to_delete;
+  imm_.current()->Unref(&to_delete);
+  for (MemTable* m : to_delete) {
+    delete m;
+  }
+}
+
+const EnvOptions* ColumnFamilyData::soptions() const {
+  return &(column_family_set_->storage_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::CreateNewMemtable() {
+  assert(current_ != nullptr);
+  if (mem_ != nullptr) {
+    delete mem_->Unref();
+  }
+  mem_ = new MemTable(internal_comparator_, options_);
+  mem_->Ref();
+}
+
+Compaction* ColumnFamilyData::PickCompaction(LogBuffer* log_buffer) {
+  return compaction_picker_->PickCompaction(current_, log_buffer);
+}
+
+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);
+}
+
+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();
+  }
+  return sv;
+}
+
+SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
+    port::Mutex* db_mutex) {
+  SuperVersion* sv = nullptr;
+  // The SuperVersion is cached in thread local storage to avoid acquiring
+  // mutex when SuperVersion does not change since the last use. When a new
+  // SuperVersion is installed, the compaction or flush thread cleans up
+  // cached SuperVersion in all existing thread local storage. To avoid
+  // acquiring mutex for this operation, we use atomic Swap() on the thread
+  // local pointer to guarantee exclusive access. If the thread local pointer
+  // is being used while a new SuperVersion is installed, the cached
+  // SuperVersion can become stale. In that case, the background thread would
+  // have swapped in kSVObsolete. We re-check the value at when returning
+  // SuperVersion back to thread local, with an atomic compare and swap.
+  // The superversion will need to be released if detected to be stale.
+  void* ptr = local_sv_->Swap(SuperVersion::kSVInUse);
+  // Invariant:
+  // (1) Scrape (always) installs kSVObsolete in ThreadLocal storage
+  // (2) the Swap above (always) installs kSVInUse, ThreadLocal storage
+  // should only keep kSVInUse before ReturnThreadLocalSuperVersion call
+  // (if no Scrape happens).
+  assert(ptr != SuperVersion::kSVInUse);
+  sv = static_cast<SuperVersion*>(ptr);
+  if (sv == SuperVersion::kSVObsolete ||
+      sv->version_number != super_version_number_.load()) {
+    RecordTick(options_.statistics.get(), NUMBER_SUPERVERSION_ACQUIRES);
+    SuperVersion* sv_to_delete = nullptr;
+
+    if (sv && sv->Unref()) {
+      RecordTick(options_.statistics.get(), NUMBER_SUPERVERSION_CLEANUPS);
+      db_mutex->Lock();
+      // NOTE: underlying resources held by superversion (sst files) might
+      // not be released until the next background job.
+      sv->Cleanup();
+      sv_to_delete = sv;
+    } else {
+      db_mutex->Lock();
+    }
+    sv = super_version_->Ref();
+    db_mutex->Unlock();
+
+    delete sv_to_delete;
+  }
+  assert(sv != nullptr);
+  return sv;
+}
+
+bool ColumnFamilyData::ReturnThreadLocalSuperVersion(SuperVersion* sv) {
+  assert(sv != nullptr);
+  // Put the SuperVersion back
+  void* expected = SuperVersion::kSVInUse;
+  if (local_sv_->CompareAndSwap(static_cast<void*>(sv), expected)) {
+    // When we see kSVInUse in the ThreadLocal, we are sure ThreadLocal
+    // storage has not been altered and no Scrape has happend. The
+    // SuperVersion is still current.
+    return true;
+  } else {
+    // ThreadLocal scrape happened in the process of this GetImpl call (after
+    // thread local Swap() at the beginning and before CompareAndSwap()).
+    // This means the SuperVersion it holds is obsolete.
+    assert(expected == SuperVersion::kSVObsolete);
+  }
+  return false;
+}
+
+SuperVersion* ColumnFamilyData::InstallSuperVersion(
+    SuperVersion* new_superversion, port::Mutex* db_mutex) {
+  new_superversion->db_mutex = db_mutex;
+  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();
+  }
+  if (old_superversion != nullptr && old_superversion->Unref()) {
+    old_superversion->Cleanup();
+    return old_superversion;  // will let caller delete outside of mutex
+  }
+  return nullptr;
+}
+
+void ColumnFamilyData::ResetThreadLocalSuperVersions() {
+  autovector<void*> sv_ptrs;
+  local_sv_->Scrape(&sv_ptrs, SuperVersion::kSVObsolete);
+  for (auto ptr : sv_ptrs) {
+    assert(ptr);
+    if (ptr == SuperVersion::kSVInUse) {
+      continue;
+    }
+    auto sv = static_cast<SuperVersion*>(ptr);
+    if (sv->Unref()) {
+      sv->Cleanup();
+      delete sv;
+    }
+  }
+}
+
+ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
+                                 const DBOptions* db_options,
+                                 const EnvOptions& storage_options,
+                                 Cache* table_cache)
+    : max_column_family_(0),
+      dummy_cfd_(new ColumnFamilyData(dbname, 0, "", nullptr, nullptr,
+                                      ColumnFamilyOptions(), db_options,
+                                      storage_options_, nullptr)),
+      default_cfd_cache_(nullptr),
+      db_name_(dbname),
+      db_options_(db_options),
+      storage_options_(storage_options),
+      table_cache_(table_cache),
+      spin_lock_(ATOMIC_FLAG_INIT) {
+  // initialize linked list
+  dummy_cfd_->prev_ = dummy_cfd_;
+  dummy_cfd_->next_ = dummy_cfd_;
+}
+
+ColumnFamilySet::~ColumnFamilySet() {
+  while (column_family_data_.size() > 0) {
+    // cfd destructor will delete itself from column_family_data_
+    auto cfd = column_family_data_.begin()->second;
+    cfd->Unref();
+    delete cfd;
+  }
+  dummy_cfd_->Unref();
+  delete dummy_cfd_;
+}
+
+ColumnFamilyData* ColumnFamilySet::GetDefault() const {
+  assert(default_cfd_cache_ != nullptr);
+  return default_cfd_cache_;
+}
+
+ColumnFamilyData* ColumnFamilySet::GetColumnFamily(uint32_t id) const {
+  auto cfd_iter = column_family_data_.find(id);
+  if (cfd_iter != column_family_data_.end()) {
+    return cfd_iter->second;
+  } else {
+    return nullptr;
+  }
+}
+
+ColumnFamilyData* ColumnFamilySet::GetColumnFamily(const std::string& name)
+    const {
+  auto cfd_iter = column_families_.find(name);
+  if (cfd_iter != column_families_.end()) {
+    auto cfd = GetColumnFamily(cfd_iter->second);
+    assert(cfd != nullptr);
+    return cfd;
+  } else {
+    return nullptr;
+  }
+}
+
+uint32_t ColumnFamilySet::GetNextColumnFamilyID() {
+  return ++max_column_family_;
+}
+
+uint32_t ColumnFamilySet::GetMaxColumnFamily() { return max_column_family_; }
+
+void ColumnFamilySet::UpdateMaxColumnFamily(uint32_t new_max_column_family) {
+  max_column_family_ = std::max(new_max_column_family, max_column_family_);
+}
+
+// under a DB mutex
+ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
+    const std::string& name, uint32_t id, Version* dummy_versions,
+    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);
+  Lock();
+  column_families_.insert({name, id});
+  column_family_data_.insert({id, new_cfd});
+  Unlock();
+  max_column_family_ = std::max(max_column_family_, id);
+  // add to linked list
+  new_cfd->next_ = dummy_cfd_;
+  auto prev = dummy_cfd_->prev_;
+  new_cfd->prev_ = prev;
+  prev->next_ = new_cfd;
+  dummy_cfd_->prev_ = new_cfd;
+  if (id == 0) {
+    default_cfd_cache_ = new_cfd;
+  }
+  return new_cfd;
+}
+
+void ColumnFamilySet::Lock() {
+  // spin lock
+  while (spin_lock_.test_and_set(std::memory_order_acquire)) {
+  }
+}
+
+void ColumnFamilySet::Unlock() { spin_lock_.clear(std::memory_order_release); }
+
+// REQUIRES: DB mutex held
+void ColumnFamilySet::FreeDeadColumnFamilies() {
+  autovector<ColumnFamilyData*> to_delete;
+  for (auto cfd = dummy_cfd_->next_; cfd != dummy_cfd_; cfd = cfd->next_) {
+    if (cfd->refs_ == 0) {
+      to_delete.push_back(cfd);
+    }
+  }
+  for (auto cfd : to_delete) {
+    // this is very rare, so it's not a problem that we do it under a mutex
+    delete cfd;
+  }
+}
+
+// under a DB mutex
+void ColumnFamilySet::RemoveColumnFamily(ColumnFamilyData* cfd) {
+  auto cfd_iter = column_family_data_.find(cfd->GetID());
+  assert(cfd_iter != column_family_data_.end());
+  Lock();
+  column_family_data_.erase(cfd_iter);
+  column_families_.erase(cfd->GetName());
+  Unlock();
+}
+
+bool ColumnFamilyMemTablesImpl::Seek(uint32_t column_family_id) {
+  if (column_family_id == 0) {
+    // optimization for common case
+    current_ = column_family_set_->GetDefault();
+  } else {
+    // maybe outside of db mutex, should lock
+    column_family_set_->Lock();
+    current_ = column_family_set_->GetColumnFamily(column_family_id);
+    column_family_set_->Unlock();
+  }
+  handle_.SetCFD(current_);
+  return current_ != nullptr;
+}
+
+uint64_t ColumnFamilyMemTablesImpl::GetLogNumber() const {
+  assert(current_ != nullptr);
+  return current_->GetLogNumber();
+}
+
+MemTable* ColumnFamilyMemTablesImpl::GetMemTable() const {
+  assert(current_ != nullptr);
+  return current_->mem();
+}
+
+const Options* ColumnFamilyMemTablesImpl::GetOptions() const {
+  assert(current_ != nullptr);
+  return current_->options();
+}
+
+ColumnFamilyHandle* ColumnFamilyMemTablesImpl::GetColumnFamilyHandle() {
+  assert(current_ != nullptr);
+  return &handle_;
+}
+
+}  // namespace rocksdb
@@ -0,0 +1,418 @@
+//  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.
+
+#pragma once
+
+#include <unordered_map>
+#include <string>
+#include <vector>
+#include <atomic>
+
+#include "rocksdb/options.h"
+#include "rocksdb/db.h"
+#include "rocksdb/env.h"
+#include "db/memtable_list.h"
+#include "db/write_batch_internal.h"
+#include "db/table_cache.h"
+#include "util/thread_local.h"
+
+namespace rocksdb {
+
+class Version;
+class VersionSet;
+class MemTable;
+class MemTableListVersion;
+class CompactionPicker;
+class Compaction;
+class InternalKey;
+class InternalStats;
+class ColumnFamilyData;
+class DBImpl;
+class LogBuffer;
+
+// ColumnFamilyHandleImpl is the class that clients use to access different
+// column families. It has non-trivial destructor, which gets called when client
+// is done using the column family
+class ColumnFamilyHandleImpl : public ColumnFamilyHandle {
+ public:
+  // create while holding the mutex
+  ColumnFamilyHandleImpl(ColumnFamilyData* cfd, DBImpl* db, port::Mutex* mutex);
+  // destroy without mutex
+  virtual ~ColumnFamilyHandleImpl();
+  virtual ColumnFamilyData* cfd() const { return cfd_; }
+
+  virtual uint32_t GetID() const;
+
+ private:
+  ColumnFamilyData* cfd_;
+  DBImpl* db_;
+  port::Mutex* mutex_;
+};
+
+// Does not ref-count ColumnFamilyData
+// We use this dummy ColumnFamilyHandleImpl because sometimes MemTableInserter
+// calls DBImpl methods. When this happens, MemTableInserter need access to
+// ColumnFamilyHandle (same as the client would need). In that case, we feed
+// MemTableInserter dummy ColumnFamilyHandle and enable it to call DBImpl
+// methods
+class ColumnFamilyHandleInternal : public ColumnFamilyHandleImpl {
+ public:
+  ColumnFamilyHandleInternal()
+      : ColumnFamilyHandleImpl(nullptr, nullptr, nullptr) {}
+
+  void SetCFD(ColumnFamilyData* cfd) { internal_cfd_ = cfd; }
+  virtual ColumnFamilyData* cfd() const override { return internal_cfd_; }
+
+ private:
+  ColumnFamilyData* internal_cfd_;
+};
+
+// holds references to memtable, all immutable memtables and version
+struct SuperVersion {
+  MemTable* mem;
+  MemTableListVersion* imm;
+  Version* current;
+  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
+  // delete all those memtables outside of mutex, during destruction
+  autovector<MemTable*> to_delete;
+  // Version number of the current SuperVersion
+  uint64_t version_number;
+  port::Mutex* db_mutex;
+
+  // should be called outside the mutex
+  SuperVersion() = default;
+  ~SuperVersion();
+  SuperVersion* Ref();
+
+  bool Unref();
+
+  // call these two methods with db mutex held
+  // Cleanup unrefs mem, imm and current. Also, it stores all memtables
+  // that needs to be deleted in to_delete vector. Unrefing those
+  // objects needs to be done in the mutex
+  void Cleanup();
+  void Init(MemTable* new_mem, MemTableListVersion* new_imm,
+            Version* new_current);
+
+  // The value of dummy is not actually used. kSVInUse takes its address as a
+  // mark in the thread local storage to indicate the SuperVersion is in use
+  // by thread. This way, the value of kSVInUse is guaranteed to have no
+  // conflict with SuperVersion object address and portable on different
+  // platform.
+  static int dummy;
+  static void* const kSVInUse;
+  static void* const kSVObsolete;
+};
+
+extern ColumnFamilyOptions SanitizeOptions(const InternalKeyComparator* icmp,
+                                           const InternalFilterPolicy* ipolicy,
+                                           const ColumnFamilyOptions& src);
+
+class ColumnFamilySet;
+
+// This class keeps all the data that a column family needs. It's mosly dumb and
+// used just to provide access to metadata.
+// Most methods require DB mutex held, unless otherwise noted
+class ColumnFamilyData {
+ public:
+  ~ColumnFamilyData();
+
+  // thread-safe
+  uint32_t GetID() const { return id_; }
+  // thread-safe
+  const std::string& GetName() const { return name_; }
+
+  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
+  // FreeDeadColumnFamilies()
+  bool Unref() {
+    assert(refs_ > 0);
+    return --refs_ == 0;
+  }
+
+  // This can only be called from single-threaded VersionSet::LogAndApply()
+  // After dropping column family no other operation on that column family
+  // will be executed. All the files and memory will be, however, kept around
+  // until client drops the column family handle. That way, client can still
+  // access data from dropped column family.
+  // Column family can be dropped and still alive. In that state:
+  // *) Column family is not included in the iteration.
+  // *) Compaction and flush is not executed on the dropped column family.
+  // *) Client can continue writing and reading from column family. However, all
+  // writes stay in the current memtable.
+  // When the dropped column family is unreferenced, then we:
+  // *) delete all memory associated with that column family
+  // *) delete all the files associated with that column family
+  void SetDropped() {
+    // can't drop default CF
+    assert(id_ != 0);
+    dropped_ = true;
+  }
+  bool IsDropped() const { return dropped_; }
+
+  // thread-safe
+  int NumberLevels() const { return options_.num_levels; }
+
+  void SetLogNumber(uint64_t log_number) { log_number_ = log_number; }
+  uint64_t GetLogNumber() const { return log_number_; }
+
+  // thread-safe
+  const Options* options() const { return &options_; }
+  const EnvOptions* soptions() const;
+
+  InternalStats* internal_stats() { return internal_stats_.get(); }
+
+  MemTableList* imm() { return &imm_; }
+  MemTable* mem() { return mem_; }
+  Version* current() { return current_; }
+  Version* dummy_versions() { return dummy_versions_; }
+  void SetMemtable(MemTable* new_mem) { mem_ = new_mem; }
+  void SetCurrent(Version* current);
+  void CreateNewMemtable();
+
+  TableCache* table_cache() { 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);
+
+  CompactionPicker* compaction_picker() { return compaction_picker_.get(); }
+  // thread-safe
+  const Comparator* user_comparator() const {
+    return internal_comparator_.user_comparator();
+  }
+  // thread-safe
+  const InternalKeyComparator& internal_comparator() const {
+    return internal_comparator_;
+  }
+
+  SuperVersion* GetSuperVersion() { return super_version_; }
+  // thread-safe
+  // Return a already referenced SuperVersion to be used safely.
+  SuperVersion* GetReferencedSuperVersion(port::Mutex* db_mutex);
+  // thread-safe
+  // Get SuperVersion stored in thread local storage. If it does not exist,
+  // get a reference from a current SuperVersion.
+  SuperVersion* GetThreadLocalSuperVersion(port::Mutex* db_mutex);
+  // Try to return SuperVersion back to thread local storage. Retrun true on
+  // success and false on failure. It fails when the thread local storage
+  // contains anything other than SuperVersion::kSVInUse flag.
+  bool ReturnThreadLocalSuperVersion(SuperVersion* sv);
+  // thread-safe
+  uint64_t GetSuperVersionNumber() const {
+    return super_version_number_.load();
+  }
+  // will return a pointer to SuperVersion* if previous SuperVersion
+  // 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);
+
+  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_;
+  }
+
+ 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,
+                   ColumnFamilySet* column_family_set);
+
+  uint32_t id_;
+  const std::string name_;
+  Version* dummy_versions_;  // Head of circular doubly-linked list of versions.
+  Version* current_;         // == dummy_versions->prev_
+
+  int refs_;                   // outstanding references to ColumnFamilyData
+  bool dropped_;               // true if client dropped it
+
+  const InternalKeyComparator internal_comparator_;
+  const InternalFilterPolicy internal_filter_policy_;
+
+  Options const options_;
+
+  std::unique_ptr<TableCache> table_cache_;
+
+  std::unique_ptr<InternalStats> internal_stats_;
+
+  MemTable* mem_;
+  MemTableList imm_;
+  SuperVersion* super_version_;
+
+  // An ordinal representing the current SuperVersion. Updated by
+  // InstallSuperVersion(), i.e. incremented every time super_version_
+  // changes.
+  std::atomic<uint64_t> super_version_number_;
+
+  // Thread's local copy of SuperVersion pointer
+  // This needs to be destructed before mutex_
+  std::unique_ptr<ThreadLocalPtr> local_sv_;
+
+  // pointers for a circular linked list. we use it to support iterations
+  // that can be concurrent with writes
+  ColumnFamilyData* next_;
+  ColumnFamilyData* prev_;
+
+  // This is the earliest log file number that contains data from this
+  // Column Family. All earlier log files must be ignored and not
+  // 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_;
+
+  // An object that keeps all the compaction stats
+  // and picks the next compaction
+  std::unique_ptr<CompactionPicker> compaction_picker_;
+
+  ColumnFamilySet* column_family_set_;
+};
+
+// ColumnFamilySet has interesting thread-safety requirements
+// * CreateColumnFamily() or RemoveColumnFamily() -- need to protect by DB
+// mutex. Inside, column_family_data_ and column_families_ will be protected
+// by Lock() and Unlock(). CreateColumnFamily() should ONLY be called from
+// VersionSet::LogAndApply() in the normal runtime. It is also called
+// during Recovery and in DumpManifest(). RemoveColumnFamily() is called
+// from ColumnFamilyData destructor
+// * Iteration -- hold DB mutex, but you can release it in the body of
+// iteration. If you release DB mutex in body, reference the column
+// family before the mutex and unreference after you unlock, since the column
+// family might get dropped when the DB mutex is released
+// * GetDefault() -- thread safe
+// * GetColumnFamily() -- either inside of DB mutex or call Lock() <-> Unlock()
+// * GetNextColumnFamilyID(), GetMaxColumnFamily(), UpdateMaxColumnFamily() --
+// inside of DB mutex
+class ColumnFamilySet {
+ public:
+  // ColumnFamilySet supports iteration
+  class iterator {
+   public:
+    explicit iterator(ColumnFamilyData* cfd)
+        : current_(cfd) {}
+    iterator& operator++() {
+      // dummy is never dead or dropped, so this will never be infinite
+      do {
+        current_ = current_->next_;
+      } while (current_->refs_ == 0 || current_->IsDropped());
+      return *this;
+    }
+    bool operator!=(const iterator& other) {
+      return this->current_ != other.current_;
+    }
+    ColumnFamilyData* operator*() { return current_; }
+
+   private:
+    ColumnFamilyData* current_;
+  };
+
+  ColumnFamilySet(const std::string& dbname, const DBOptions* db_options,
+                  const EnvOptions& storage_options, Cache* table_cache);
+  ~ColumnFamilySet();
+
+  ColumnFamilyData* GetDefault() const;
+  // GetColumnFamily() calls return nullptr if column family is not found
+  ColumnFamilyData* GetColumnFamily(uint32_t id) const;
+  ColumnFamilyData* GetColumnFamily(const std::string& name) const;
+  // this call will return the next available column family ID. it guarantees
+  // that there is no column family with id greater than or equal to the
+  // returned value in the current running instance or anytime in RocksDB
+  // instance history.
+  uint32_t GetNextColumnFamilyID();
+  uint32_t GetMaxColumnFamily();
+  void UpdateMaxColumnFamily(uint32_t new_max_column_family);
+
+  ColumnFamilyData* CreateColumnFamily(const std::string& name, uint32_t id,
+                                       Version* dummy_version,
+                                       const ColumnFamilyOptions& options);
+
+  iterator begin() { return iterator(dummy_cfd_->next_); }
+  iterator end() { return iterator(dummy_cfd_); }
+
+  void Lock();
+  void Unlock();
+
+  // REQUIRES: DB mutex held
+  // Don't call while iterating over ColumnFamilySet
+  void FreeDeadColumnFamilies();
+
+ private:
+  friend class ColumnFamilyData;
+  // helper function that gets called from cfd destructor
+  // REQUIRES: DB mutex held
+  void RemoveColumnFamily(ColumnFamilyData* cfd);
+
+  // column_families_ and column_family_data_ need to be protected:
+  // * when mutating: 1. DB mutex locked first, 2. spinlock locked second
+  // * when reading, either: 1. lock DB mutex, or 2. lock spinlock
+  //  (if both, respect the ordering to avoid deadlock!)
+  std::unordered_map<std::string, uint32_t> column_families_;
+  std::unordered_map<uint32_t, ColumnFamilyData*> column_family_data_;
+
+  uint32_t max_column_family_;
+  ColumnFamilyData* dummy_cfd_;
+  // We don't hold the refcount here, since default column family always exists
+  // We are also not responsible for cleaning up default_cfd_cache_. This is
+  // just a cache that makes common case (accessing default column family)
+  // faster
+  ColumnFamilyData* default_cfd_cache_;
+
+  const std::string db_name_;
+  const DBOptions* const db_options_;
+  const EnvOptions storage_options_;
+  Cache* table_cache_;
+  std::atomic_flag spin_lock_;
+};
+
+// We use ColumnFamilyMemTablesImpl to provide WriteBatch a way to access
+// 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) {}
+
+  // sets current_ to ColumnFamilyData with column_family_id
+  // returns false if column family doesn't exist
+  bool Seek(uint32_t column_family_id) override;
+
+  // Returns log number of the selected column family
+  uint64_t GetLogNumber() const override;
+
+  // REQUIRES: Seek() called first
+  virtual MemTable* GetMemTable() const override;
+
+  // Returns options for selected column family
+  // REQUIRES: Seek() called first
+  virtual const Options* GetOptions() const override;
+
+  // Returns column family handle for the selected column family
+  virtual ColumnFamilyHandle* GetColumnFamilyHandle() override;
+
+ private:
+  ColumnFamilySet* column_family_set_;
+  ColumnFamilyData* current_;
+  ColumnFamilyHandleInternal handle_;
+};
+
+}  // namespace rocksdb
@@ -0,0 +1,977 @@
+//  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 <vector>
+#include <string>
+
+#include "db/db_impl.h"
+#include "rocksdb/env.h"
+#include "rocksdb/db.h"
+#include "util/testharness.h"
+#include "util/testutil.h"
+#include "util/coding.h"
+#include "utilities/merge_operators.h"
+
+namespace rocksdb {
+
+namespace {
+std::string RandomString(Random* rnd, int len) {
+  std::string r;
+  test::RandomString(rnd, len, &r);
+  return r;
+}
+}  // anonymous namespace
+
+// counts how many operations were performed
+class EnvCounter : public EnvWrapper {
+ public:
+  explicit EnvCounter(Env* base)
+      : EnvWrapper(base), num_new_writable_file_(0) {}
+  int GetNumberOfNewWritableFileCalls() {
+    return num_new_writable_file_;
+  }
+  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r,
+                         const EnvOptions& soptions) {
+    ++num_new_writable_file_;
+    return EnvWrapper::NewWritableFile(f, r, soptions);
+  }
+
+ private:
+  int num_new_writable_file_;
+};
+
+class ColumnFamilyTest {
+ public:
+  ColumnFamilyTest() : rnd_(139) {
+    env_ = new EnvCounter(Env::Default());
+    dbname_ = test::TmpDir() + "/column_family_test";
+    db_options_.create_if_missing = true;
+    db_options_.env = env_;
+    DestroyDB(dbname_, Options(db_options_, column_family_options_));
+  }
+
+  ~ColumnFamilyTest() {
+    delete env_;
+  }
+
+  void Close() {
+    for (auto h : handles_) {
+      delete h;
+    }
+    handles_.clear();
+    names_.clear();
+    delete db_;
+    db_ = nullptr;
+  }
+
+  Status TryOpen(std::vector<std::string> cf,
+                 std::vector<ColumnFamilyOptions> options = {}) {
+    std::vector<ColumnFamilyDescriptor> column_families;
+    names_.clear();
+    for (size_t i = 0; i < cf.size(); ++i) {
+      column_families.push_back(ColumnFamilyDescriptor(
+          cf[i], options.size() == 0 ? column_family_options_ : options[i]));
+      names_.push_back(cf[i]);
+    }
+    return DB::Open(db_options_, dbname_, column_families, &handles_, &db_);
+  }
+
+  Status OpenReadOnly(std::vector<std::string> cf,
+                         std::vector<ColumnFamilyOptions> options = {}) {
+    std::vector<ColumnFamilyDescriptor> column_families;
+    names_.clear();
+    for (size_t i = 0; i < cf.size(); ++i) {
+      column_families.push_back(ColumnFamilyDescriptor(
+          cf[i], options.size() == 0 ? column_family_options_ : options[i]));
+      names_.push_back(cf[i]);
+    }
+    return DB::OpenForReadOnly(db_options_, dbname_, column_families, &handles_,
+                               &db_);
+  }
+
+  void AssertOpenReadOnly(std::vector<std::string> cf,
+                    std::vector<ColumnFamilyOptions> options = {}) {
+    ASSERT_OK(OpenReadOnly(cf, options));
+  }
+
+
+  void Open(std::vector<std::string> cf,
+            std::vector<ColumnFamilyOptions> options = {}) {
+    ASSERT_OK(TryOpen(cf, options));
+  }
+
+  void Open() {
+    Open({"default"});
+  }
+
+  DBImpl* dbfull() { return reinterpret_cast<DBImpl*>(db_); }
+
+  int GetProperty(int cf, std::string property) {
+    std::string value;
+    ASSERT_TRUE(dbfull()->GetProperty(handles_[cf], property, &value));
+    return std::stoi(value);
+  }
+
+  void Destroy() {
+    for (auto h : handles_) {
+      delete h;
+    }
+    handles_.clear();
+    names_.clear();
+    delete db_;
+    db_ = nullptr;
+    ASSERT_OK(DestroyDB(dbname_, Options(db_options_, column_family_options_)));
+  }
+
+  void CreateColumnFamilies(
+      const std::vector<std::string>& cfs,
+      const std::vector<ColumnFamilyOptions> options = {}) {
+    int cfi = handles_.size();
+    handles_.resize(cfi + cfs.size());
+    names_.resize(cfi + cfs.size());
+    for (size_t i = 0; i < cfs.size(); ++i) {
+      ASSERT_OK(db_->CreateColumnFamily(
+          options.size() == 0 ? column_family_options_ : options[i], cfs[i],
+          &handles_[cfi]));
+      names_[cfi] = cfs[i];
+      cfi++;
+    }
+  }
+
+  void Reopen(const std::vector<ColumnFamilyOptions> options = {}) {
+    std::vector<std::string> names;
+    for (auto name : names_) {
+      if (name != "") {
+        names.push_back(name);
+      }
+    }
+    Close();
+    assert(options.size() == 0 || names.size() == options.size());
+    Open(names, options);
+  }
+
+  void CreateColumnFamiliesAndReopen(const std::vector<std::string>& cfs) {
+    CreateColumnFamilies(cfs);
+    Reopen();
+  }
+
+  void DropColumnFamilies(const std::vector<int>& cfs) {
+    for (auto cf : cfs) {
+      ASSERT_OK(db_->DropColumnFamily(handles_[cf]));
+      delete handles_[cf];
+      handles_[cf] = nullptr;
+      names_[cf] = "";
+    }
+  }
+
+  void PutRandomData(int cf, int num, int key_value_size) {
+    for (int i = 0; i < num; ++i) {
+      // 10 bytes for key, rest is value
+      ASSERT_OK(Put(cf, test::RandomKey(&rnd_, 10),
+                    RandomString(&rnd_, key_value_size - 10)));
+    }
+  }
+
+  void WaitForFlush(int cf) {
+    ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[cf]));
+  }
+
+  void WaitForCompaction() { ASSERT_OK(dbfull()->TEST_WaitForCompact()); }
+
+  Status Put(int cf, const std::string& key, const std::string& value) {
+    return db_->Put(WriteOptions(), handles_[cf], Slice(key), Slice(value));
+  }
+  Status Merge(int cf, const std::string& key, const std::string& value) {
+    return db_->Merge(WriteOptions(), handles_[cf], Slice(key), Slice(value));
+  }
+  Status Flush(int cf) {
+    return db_->Flush(FlushOptions(), handles_[cf]);
+  }
+
+  std::string Get(int cf, const std::string& key) {
+    ReadOptions options;
+    options.verify_checksums = true;
+    std::string result;
+    Status s = db_->Get(options, handles_[cf], Slice(key), &result);
+    if (s.IsNotFound()) {
+      result = "NOT_FOUND";
+    } else if (!s.ok()) {
+      result = s.ToString();
+    }
+    return result;
+  }
+
+  void CompactAll(int cf) {
+    ASSERT_OK(db_->CompactRange(handles_[cf], nullptr, nullptr));
+  }
+
+  void Compact(int cf, const Slice& start, const Slice& limit) {
+    ASSERT_OK(db_->CompactRange(handles_[cf], &start, &limit));
+  }
+
+  int NumTableFilesAtLevel(int level, int cf) {
+    return GetProperty(cf,
+                       "rocksdb.num-files-at-level" + std::to_string(level));
+  }
+
+  // Return spread of files per level
+  std::string FilesPerLevel(int cf) {
+    std::string result;
+    int last_non_zero_offset = 0;
+    for (int level = 0; level < dbfull()->NumberLevels(handles_[cf]); level++) {
+      int f = NumTableFilesAtLevel(level, cf);
+      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;
+  }
+
+  int CountLiveFiles() {
+    std::vector<LiveFileMetaData> metadata;
+    db_->GetLiveFilesMetaData(&metadata);
+    return static_cast<int>(metadata.size());
+  }
+
+  // Do n memtable flushes, each of which produces an sstable
+  // covering the range [small,large].
+  void MakeTables(int cf, int n, const std::string& small,
+                  const std::string& large) {
+    for (int i = 0; i < n; i++) {
+      ASSERT_OK(Put(cf, small, "begin"));
+      ASSERT_OK(Put(cf, large, "end"));
+      ASSERT_OK(db_->Flush(FlushOptions(), handles_[cf]));
+    }
+  }
+
+  int CountLiveLogFiles() {
+    int micros_wait_for_log_deletion = 20000;
+    env_->SleepForMicroseconds(micros_wait_for_log_deletion);
+    int ret = 0;
+    VectorLogPtr wal_files;
+    Status s;
+    // GetSortedWalFiles is a flakey function -- it gets all the wal_dir
+    // children files and then later checks for their existance. if some of the
+    // log files doesn't exist anymore, it reports an error. it does all of this
+    // without DB mutex held, so if a background process deletes the log file
+    // while the function is being executed, it returns an error. We retry the
+    // function 10 times to avoid the error failing the test
+    for (int retries = 0; retries < 10; ++retries) {
+      wal_files.clear();
+      s = db_->GetSortedWalFiles(wal_files);
+      if (s.ok()) {
+        break;
+      }
+    }
+    ASSERT_OK(s);
+    for (const auto& wal : wal_files) {
+      if (wal->Type() == kAliveLogFile) {
+        ++ret;
+      }
+    }
+    return ret;
+  }
+
+  void AssertNumberOfImmutableMemtables(std::vector<int> num_per_cf) {
+    assert(num_per_cf.size() == handles_.size());
+
+    for (size_t i = 0; i < num_per_cf.size(); ++i) {
+      ASSERT_EQ(num_per_cf[i],
+                GetProperty(i, "rocksdb.num-immutable-mem-table"));
+    }
+  }
+
+  void CopyFile(const std::string& source, const std::string& destination,
+                uint64_t size = 0) {
+    const EnvOptions soptions;
+    unique_ptr<SequentialFile> srcfile;
+    ASSERT_OK(env_->NewSequentialFile(source, &srcfile, soptions));
+    unique_ptr<WritableFile> destfile;
+    ASSERT_OK(env_->NewWritableFile(destination, &destfile, soptions));
+
+    if (size == 0) {
+      // default argument means copy everything
+      ASSERT_OK(env_->GetFileSize(source, &size));
+    }
+
+    char buffer[4096];
+    Slice slice;
+    while (size > 0) {
+      uint64_t one = std::min(uint64_t(sizeof(buffer)), size);
+      ASSERT_OK(srcfile->Read(one, &slice, buffer));
+      ASSERT_OK(destfile->Append(slice));
+      size -= slice.size();
+    }
+    ASSERT_OK(destfile->Close());
+  }
+
+  std::vector<ColumnFamilyHandle*> handles_;
+  std::vector<std::string> names_;
+  ColumnFamilyOptions column_family_options_;
+  DBOptions db_options_;
+  std::string dbname_;
+  DB* db_ = nullptr;
+  EnvCounter* env_;
+  Random rnd_;
+};
+
+TEST(ColumnFamilyTest, DontReuseColumnFamilyID) {
+  for (int iter = 0; iter < 3; ++iter) {
+    Open();
+    CreateColumnFamilies({"one", "two", "three"});
+    for (size_t i = 0; i < handles_.size(); ++i) {
+      auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(handles_[i]);
+      ASSERT_EQ(i, cfh->GetID());
+    }
+    if (iter == 1) {
+      Reopen();
+    }
+    DropColumnFamilies({3});
+    Reopen();
+    if (iter == 2) {
+      // this tests if max_column_family is correctly persisted with
+      // WriteSnapshot()
+      Reopen();
+    }
+    CreateColumnFamilies({"three2"});
+    // ID 3 that was used for dropped column family "three" should not be reused
+    auto cfh3 = reinterpret_cast<ColumnFamilyHandleImpl*>(handles_[3]);
+    ASSERT_EQ(4U, cfh3->GetID());
+    Close();
+    Destroy();
+  }
+}
+
+
+TEST(ColumnFamilyTest, AddDrop) {
+  Open();
+  CreateColumnFamilies({"one", "two", "three"});
+  ASSERT_EQ("NOT_FOUND", Get(1, "fodor"));
+  ASSERT_EQ("NOT_FOUND", Get(2, "fodor"));
+  DropColumnFamilies({2});
+  ASSERT_EQ("NOT_FOUND", Get(1, "fodor"));
+  CreateColumnFamilies({"four"});
+  ASSERT_EQ("NOT_FOUND", Get(3, "fodor"));
+  ASSERT_OK(Put(1, "fodor", "mirko"));
+  ASSERT_EQ("mirko", Get(1, "fodor"));
+  ASSERT_EQ("NOT_FOUND", Get(3, "fodor"));
+  Close();
+  ASSERT_TRUE(TryOpen({"default"}).IsInvalidArgument());
+  Open({"default", "one", "three", "four"});
+  DropColumnFamilies({1});
+  Reopen();
+  Close();
+
+  std::vector<std::string> families;
+  ASSERT_OK(DB::ListColumnFamilies(db_options_, dbname_, &families));
+  sort(families.begin(), families.end());
+  ASSERT_TRUE(families ==
+              std::vector<std::string>({"default", "four", "three"}));
+}
+
+TEST(ColumnFamilyTest, DropTest) {
+  // first iteration - dont reopen DB before dropping
+  // second iteration - reopen DB before dropping
+  for (int iter = 0; iter < 2; ++iter) {
+    Open({"default"});
+    CreateColumnFamiliesAndReopen({"pikachu"});
+    for (int i = 0; i < 100; ++i) {
+      ASSERT_OK(Put(1, std::to_string(i), "bar" + std::to_string(i)));
+    }
+    ASSERT_OK(Flush(1));
+
+    if (iter == 1) {
+      Reopen();
+    }
+    ASSERT_EQ("bar1", Get(1, "1"));
+
+    ASSERT_EQ(CountLiveFiles(), 1);
+    DropColumnFamilies({1});
+    // make sure that all files are deleted when we drop the column family
+    ASSERT_EQ(CountLiveFiles(), 0);
+    Destroy();
+  }
+}
+
+TEST(ColumnFamilyTest, WriteBatchFailure) {
+  Open();
+  CreateColumnFamiliesAndReopen({"one", "two"});
+  WriteBatch batch;
+  batch.Put(handles_[1], Slice("non-existing"), Slice("column-family"));
+  ASSERT_OK(db_->Write(WriteOptions(), &batch));
+  DropColumnFamilies({1});
+  Status s = db_->Write(WriteOptions(), &batch);
+  ASSERT_TRUE(s.IsInvalidArgument());
+  Close();
+}
+
+TEST(ColumnFamilyTest, ReadWrite) {
+  Open();
+  CreateColumnFamiliesAndReopen({"one", "two"});
+  ASSERT_OK(Put(0, "foo", "v1"));
+  ASSERT_OK(Put(0, "bar", "v2"));
+  ASSERT_OK(Put(1, "mirko", "v3"));
+  ASSERT_OK(Put(0, "foo", "v2"));
+  ASSERT_OK(Put(2, "fodor", "v5"));
+
+  for (int iter = 0; iter <= 3; ++iter) {
+    ASSERT_EQ("v2", Get(0, "foo"));
+    ASSERT_EQ("v2", Get(0, "bar"));
+    ASSERT_EQ("v3", Get(1, "mirko"));
+    ASSERT_EQ("v5", Get(2, "fodor"));
+    ASSERT_EQ("NOT_FOUND", Get(0, "fodor"));
+    ASSERT_EQ("NOT_FOUND", Get(1, "fodor"));
+    ASSERT_EQ("NOT_FOUND", Get(2, "foo"));
+    if (iter <= 1) {
+      Reopen();
+    }
+  }
+  Close();
+}
+
+TEST(ColumnFamilyTest, IgnoreRecoveredLog) {
+  std::string backup_logs = dbname_ + "/backup_logs";
+
+  // delete old files in backup_logs directory
+  ASSERT_OK(env_->CreateDirIfMissing(dbname_));
+  ASSERT_OK(env_->CreateDirIfMissing(backup_logs));
+  std::vector<std::string> old_files;
+  env_->GetChildren(backup_logs, &old_files);
+  for (auto& file : old_files) {
+    if (file != "." && file != "..") {
+      env_->DeleteFile(backup_logs + "/" + file);
+    }
+  }
+
+  column_family_options_.merge_operator =
+      MergeOperators::CreateUInt64AddOperator();
+  db_options_.wal_dir = dbname_ + "/logs";
+  Destroy();
+  Open();
+  CreateColumnFamilies({"cf1", "cf2"});
+
+  // fill up the DB
+  std::string one, two, three;
+  PutFixed64(&one, 1);
+  PutFixed64(&two, 2);
+  PutFixed64(&three, 3);
+  ASSERT_OK(Merge(0, "foo", one));
+  ASSERT_OK(Merge(1, "mirko", one));
+  ASSERT_OK(Merge(0, "foo", one));
+  ASSERT_OK(Merge(2, "bla", one));
+  ASSERT_OK(Merge(2, "fodor", one));
+  ASSERT_OK(Merge(0, "bar", one));
+  ASSERT_OK(Merge(2, "bla", one));
+  ASSERT_OK(Merge(1, "mirko", two));
+  ASSERT_OK(Merge(1, "franjo", one));
+
+  // copy the logs to backup
+  std::vector<std::string> logs;
+  env_->GetChildren(db_options_.wal_dir, &logs);
+  for (auto& log : logs) {
+    if (log != ".." && log != ".") {
+      CopyFile(db_options_.wal_dir + "/" + log, backup_logs + "/" + log);
+    }
+  }
+
+  // recover the DB
+  Close();
+
+  // 1. check consistency
+  // 2. copy the logs from backup back to WAL dir. if the recovery happens
+  // again on the same log files, this should lead to incorrect results
+  // due to applying merge operator twice
+  // 3. check consistency
+  for (int iter = 0; iter < 2; ++iter) {
+    // assert consistency
+    Open({"default", "cf1", "cf2"});
+    ASSERT_EQ(two, Get(0, "foo"));
+    ASSERT_EQ(one, Get(0, "bar"));
+    ASSERT_EQ(three, Get(1, "mirko"));
+    ASSERT_EQ(one, Get(1, "franjo"));
+    ASSERT_EQ(one, Get(2, "fodor"));
+    ASSERT_EQ(two, Get(2, "bla"));
+    Close();
+
+    if (iter == 0) {
+      // copy the logs from backup back to wal dir
+      for (auto& log : logs) {
+        if (log != ".." && log != ".") {
+          CopyFile(backup_logs + "/" + log, db_options_.wal_dir + "/" + log);
+        }
+      }
+    }
+  }
+}
+
+TEST(ColumnFamilyTest, FlushTest) {
+  Open();
+  CreateColumnFamiliesAndReopen({"one", "two"});
+  ASSERT_OK(Put(0, "foo", "v1"));
+  ASSERT_OK(Put(0, "bar", "v2"));
+  ASSERT_OK(Put(1, "mirko", "v3"));
+  ASSERT_OK(Put(0, "foo", "v2"));
+  ASSERT_OK(Put(2, "fodor", "v5"));
+  for (int i = 0; i < 3; ++i) {
+    Flush(i);
+  }
+  Reopen();
+
+  for (int iter = 0; iter <= 2; ++iter) {
+    ASSERT_EQ("v2", Get(0, "foo"));
+    ASSERT_EQ("v2", Get(0, "bar"));
+    ASSERT_EQ("v3", Get(1, "mirko"));
+    ASSERT_EQ("v5", Get(2, "fodor"));
+    ASSERT_EQ("NOT_FOUND", Get(0, "fodor"));
+    ASSERT_EQ("NOT_FOUND", Get(1, "fodor"));
+    ASSERT_EQ("NOT_FOUND", Get(2, "foo"));
+    if (iter <= 1) {
+      Reopen();
+    }
+  }
+  Close();
+}
+
+// Makes sure that obsolete log files get deleted
+TEST(ColumnFamilyTest, LogDeletionTest) {
+  db_options_.max_total_wal_size = std::numeric_limits<uint64_t>::max();
+  column_family_options_.write_buffer_size = 100000;  // 100KB
+  Open();
+  CreateColumnFamilies({"one", "two", "three", "four"});
+  // Each bracket is one log file. if number is in (), it means
+  // we don't need it anymore (it's been flushed)
+  // []
+  ASSERT_EQ(CountLiveLogFiles(), 0);
+  PutRandomData(0, 1, 100);
+  // [0]
+  PutRandomData(1, 1, 100);
+  // [0, 1]
+  PutRandomData(1, 1000, 100);
+  WaitForFlush(1);
+  // [0, (1)] [1]
+  ASSERT_EQ(CountLiveLogFiles(), 2);
+  PutRandomData(0, 1, 100);
+  // [0, (1)] [0, 1]
+  ASSERT_EQ(CountLiveLogFiles(), 2);
+  PutRandomData(2, 1, 100);
+  // [0, (1)] [0, 1, 2]
+  PutRandomData(2, 1000, 100);
+  WaitForFlush(2);
+  // [0, (1)] [0, 1, (2)] [2]
+  ASSERT_EQ(CountLiveLogFiles(), 3);
+  PutRandomData(2, 1000, 100);
+  WaitForFlush(2);
+  // [0, (1)] [0, 1, (2)] [(2)] [2]
+  ASSERT_EQ(CountLiveLogFiles(), 4);
+  PutRandomData(3, 1, 100);
+  // [0, (1)] [0, 1, (2)] [(2)] [2, 3]
+  PutRandomData(1, 1, 100);
+  // [0, (1)] [0, 1, (2)] [(2)] [1, 2, 3]
+  ASSERT_EQ(CountLiveLogFiles(), 4);
+  PutRandomData(1, 1000, 100);
+  WaitForFlush(1);
+  // [0, (1)] [0, (1), (2)] [(2)] [(1), 2, 3] [1]
+  ASSERT_EQ(CountLiveLogFiles(), 5);
+  PutRandomData(0, 1000, 100);
+  WaitForFlush(0);
+  // [(0), (1)] [(0), (1), (2)] [(2)] [(1), 2, 3] [1, (0)] [0]
+  // delete obsolete logs -->
+  // [(1), 2, 3] [1, (0)] [0]
+  ASSERT_EQ(CountLiveLogFiles(), 3);
+  PutRandomData(0, 1000, 100);
+  WaitForFlush(0);
+  // [(1), 2, 3] [1, (0)], [(0)] [0]
+  ASSERT_EQ(CountLiveLogFiles(), 4);
+  PutRandomData(1, 1000, 100);
+  WaitForFlush(1);
+  // [(1), 2, 3] [(1), (0)] [(0)] [0, (1)] [1]
+  ASSERT_EQ(CountLiveLogFiles(), 5);
+  PutRandomData(2, 1000, 100);
+  WaitForFlush(2);
+  // [(1), (2), 3] [(1), (0)] [(0)] [0, (1)] [1, (2)], [2]
+  ASSERT_EQ(CountLiveLogFiles(), 6);
+  PutRandomData(3, 1000, 100);
+  WaitForFlush(3);
+  // [(1), (2), (3)] [(1), (0)] [(0)] [0, (1)] [1, (2)], [2, (3)] [3]
+  // delete obsolete logs -->
+  // [0, (1)] [1, (2)], [2, (3)] [3]
+  ASSERT_EQ(CountLiveLogFiles(), 4);
+  Close();
+}
+
+// Makes sure that obsolete log files get deleted
+TEST(ColumnFamilyTest, DifferentWriteBufferSizes) {
+  // disable flushing stale column families
+  db_options_.max_total_wal_size = std::numeric_limits<uint64_t>::max();
+  Open();
+  CreateColumnFamilies({"one", "two", "three"});
+  ColumnFamilyOptions default_cf, one, two, three;
+  // setup options. all column families have max_write_buffer_number setup to 10
+  // "default" -> 100KB memtable, start flushing immediatelly
+  // "one" -> 200KB memtable, start flushing with two immutable memtables
+  // "two" -> 1MB memtable, start flushing with three immutable memtables
+  // "three" -> 90KB memtable, start flushing with four immutable memtables
+  default_cf.write_buffer_size = 100000;
+  default_cf.max_write_buffer_number = 10;
+  default_cf.min_write_buffer_number_to_merge = 1;
+  one.write_buffer_size = 200000;
+  one.max_write_buffer_number = 10;
+  one.min_write_buffer_number_to_merge = 2;
+  two.write_buffer_size = 1000000;
+  two.max_write_buffer_number = 10;
+  two.min_write_buffer_number_to_merge = 3;
+  three.write_buffer_size = 90000;
+  three.max_write_buffer_number = 10;
+  three.min_write_buffer_number_to_merge = 4;
+
+  Reopen({default_cf, one, two, three});
+
+  int micros_wait_for_flush = 10000;
+  PutRandomData(0, 100, 1000);
+  WaitForFlush(0);
+  AssertNumberOfImmutableMemtables({0, 0, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 1);
+  PutRandomData(1, 200, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 2);
+  PutRandomData(2, 1000, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 1, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 3);
+  PutRandomData(2, 1000, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 2, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 4);
+  PutRandomData(3, 90, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 2, 1});
+  ASSERT_EQ(CountLiveLogFiles(), 5);
+  PutRandomData(3, 90, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 2, 2});
+  ASSERT_EQ(CountLiveLogFiles(), 6);
+  PutRandomData(3, 90, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 2, 3});
+  ASSERT_EQ(CountLiveLogFiles(), 7);
+  PutRandomData(0, 100, 1000);
+  WaitForFlush(0);
+  AssertNumberOfImmutableMemtables({0, 1, 2, 3});
+  ASSERT_EQ(CountLiveLogFiles(), 8);
+  PutRandomData(2, 100, 10000);
+  WaitForFlush(2);
+  AssertNumberOfImmutableMemtables({0, 1, 0, 3});
+  ASSERT_EQ(CountLiveLogFiles(), 9);
+  PutRandomData(3, 90, 1000);
+  WaitForFlush(3);
+  AssertNumberOfImmutableMemtables({0, 1, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 10);
+  PutRandomData(3, 90, 1000);
+  env_->SleepForMicroseconds(micros_wait_for_flush);
+  AssertNumberOfImmutableMemtables({0, 1, 0, 1});
+  ASSERT_EQ(CountLiveLogFiles(), 11);
+  PutRandomData(1, 200, 1000);
+  WaitForFlush(1);
+  AssertNumberOfImmutableMemtables({0, 0, 0, 1});
+  ASSERT_EQ(CountLiveLogFiles(), 5);
+  PutRandomData(3, 90*6, 1000);
+  WaitForFlush(3);
+  AssertNumberOfImmutableMemtables({0, 0, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 12);
+  PutRandomData(0, 100, 1000);
+  WaitForFlush(0);
+  AssertNumberOfImmutableMemtables({0, 0, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 12);
+  PutRandomData(2, 3*100, 10000);
+  WaitForFlush(2);
+  AssertNumberOfImmutableMemtables({0, 0, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 12);
+  PutRandomData(1, 2*200, 1000);
+  WaitForFlush(1);
+  AssertNumberOfImmutableMemtables({0, 0, 0, 0});
+  ASSERT_EQ(CountLiveLogFiles(), 7);
+  Close();
+}
+
+TEST(ColumnFamilyTest, DifferentMergeOperators) {
+  Open();
+  CreateColumnFamilies({"first", "second"});
+  ColumnFamilyOptions default_cf, first, second;
+  first.merge_operator = MergeOperators::CreateUInt64AddOperator();
+  second.merge_operator = MergeOperators::CreateStringAppendOperator();
+  Reopen({default_cf, first, second});
+
+  std::string one, two, three;
+  PutFixed64(&one, 1);
+  PutFixed64(&two, 2);
+  PutFixed64(&three, 3);
+
+  ASSERT_OK(Put(0, "foo", two));
+  ASSERT_OK(Put(0, "foo", one));
+  ASSERT_TRUE(Merge(0, "foo", two).IsNotSupported());
+  ASSERT_EQ(Get(0, "foo"), one);
+
+  ASSERT_OK(Put(1, "foo", two));
+  ASSERT_OK(Put(1, "foo", one));
+  ASSERT_OK(Merge(1, "foo", two));
+  ASSERT_EQ(Get(1, "foo"), three);
+
+  ASSERT_OK(Put(2, "foo", two));
+  ASSERT_OK(Put(2, "foo", one));
+  ASSERT_OK(Merge(2, "foo", two));
+  ASSERT_EQ(Get(2, "foo"), one + "," + two);
+  Close();
+}
+
+TEST(ColumnFamilyTest, DifferentCompactionStyles) {
+  Open();
+  CreateColumnFamilies({"one", "two"});
+  ColumnFamilyOptions default_cf, one, two;
+  db_options_.max_open_files = 20;  // only 10 files in file cache
+  db_options_.disableDataSync = true;
+
+  default_cf.compaction_style = kCompactionStyleLevel;
+  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;
+  default_cf.disable_seek_compaction = false;
+
+  one.compaction_style = kCompactionStyleUniversal;
+  // trigger compaction if there are >= 4 files
+  one.level0_file_num_compaction_trigger = 4;
+  one.write_buffer_size = 100000;
+
+  two.compaction_style = kCompactionStyleLevel;
+  two.num_levels = 4;
+  two.max_mem_compaction_level = 0;
+  two.level0_file_num_compaction_trigger = 3;
+  two.write_buffer_size = 100000;
+
+  Reopen({default_cf, one, two});
+
+  // SETUP column family "default" - test read compaction
+  ASSERT_EQ("", FilesPerLevel(0));
+  PutRandomData(0, 1, 4096);
+  ASSERT_OK(Flush(0));
+  ASSERT_EQ("0,0,1", FilesPerLevel(0));
+  // write 8MB
+  PutRandomData(0, 2000, 4096);
+  ASSERT_OK(Flush(0));
+  // clear levels 0 and 1
+  dbfull()->TEST_CompactRange(0, nullptr, nullptr, handles_[0]);
+  dbfull()->TEST_CompactRange(1, nullptr, nullptr, handles_[0]);
+  ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0);
+  ASSERT_EQ(NumTableFilesAtLevel(1, 0), 0);
+  // write some new keys into level 0 and 1
+  PutRandomData(0, 1024, 512);
+  ASSERT_OK(Flush(0));
+  WaitForCompaction();
+  PutRandomData(0, 10, 512);
+  ASSERT_OK(Flush(0));
+  // remember number of files in each level
+  int l1 = NumTableFilesAtLevel(0, 0);
+  int l2 = NumTableFilesAtLevel(1, 0);
+  int l3 = NumTableFilesAtLevel(2, 0);
+  ASSERT_NE(l1, 0);
+  ASSERT_NE(l2, 0);
+  ASSERT_NE(l3, 0);
+
+  // SETUP column family "one" -- universal style
+  for (int i = 0; i < one.level0_file_num_compaction_trigger - 1; ++i) {
+    PutRandomData(1, 11, 10000);
+    WaitForFlush(1);
+    ASSERT_EQ(std::to_string(i + 1), FilesPerLevel(1));
+  }
+
+  // SETUP column family "two" -- level style with 4 levels
+  for (int i = 0; i < two.level0_file_num_compaction_trigger - 1; ++i) {
+    PutRandomData(2, 15, 10000);
+    WaitForFlush(2);
+    ASSERT_EQ(std::to_string(i + 1), FilesPerLevel(2));
+  }
+
+  // TRIGGER compaction "default"
+  // read a bunch of times, trigger read compaction
+  for (int i = 0; i < 200000; ++i) {
+    Get(0, std::to_string(i));
+  }
+
+  // TRIGGER compaction "one"
+  PutRandomData(1, 12, 10000);
+
+  // TRIGGER compaction "two"
+  PutRandomData(2, 10, 10000);
+
+  // WAIT for compactions
+  WaitForCompaction();
+
+  // VERIFY compaction "default"
+  // verify that the number of files have decreased
+  // in some level, indicating that there was a compaction
+  ASSERT_TRUE(NumTableFilesAtLevel(0, 0) < l1 ||
+              NumTableFilesAtLevel(1, 0) < l2 ||
+              NumTableFilesAtLevel(2, 0) < l3);
+
+  // VERIFY compaction "one"
+  ASSERT_EQ("1", FilesPerLevel(1));
+
+  // VERIFY compaction "two"
+  ASSERT_EQ("0,1", FilesPerLevel(2));
+  CompactAll(2);
+  ASSERT_EQ("0,1", FilesPerLevel(2));
+
+  Close();
+}
+
+namespace {
+std::string IterStatus(Iterator* iter) {
+  std::string result;
+  if (iter->Valid()) {
+    result = iter->key().ToString() + "->" + iter->value().ToString();
+  } else {
+    result = "(invalid)";
+  }
+  return result;
+}
+}  // anonymous namespace
+
+TEST(ColumnFamilyTest, NewIteratorsTest) {
+  // iter == 0 -- no tailing
+  // iter == 2 -- tailing
+  for (int iter = 0; iter < 2; ++iter) {
+    Open();
+    CreateColumnFamiliesAndReopen({"one", "two"});
+    ASSERT_OK(Put(0, "a", "b"));
+    ASSERT_OK(Put(1, "b", "a"));
+    ASSERT_OK(Put(2, "c", "m"));
+    ASSERT_OK(Put(2, "v", "t"));
+    std::vector<Iterator*> iterators;
+    ReadOptions options;
+    options.tailing = (iter == 1);
+    ASSERT_OK(db_->NewIterators(options, handles_, &iterators));
+
+    for (auto it : iterators) {
+      it->SeekToFirst();
+    }
+    ASSERT_EQ(IterStatus(iterators[0]), "a->b");
+    ASSERT_EQ(IterStatus(iterators[1]), "b->a");
+    ASSERT_EQ(IterStatus(iterators[2]), "c->m");
+
+    ASSERT_OK(Put(1, "x", "x"));
+
+    for (auto it : iterators) {
+      it->Next();
+    }
+
+    ASSERT_EQ(IterStatus(iterators[0]), "(invalid)");
+    if (iter == 0) {
+      // no tailing
+      ASSERT_EQ(IterStatus(iterators[1]), "(invalid)");
+    } else {
+      // tailing
+      ASSERT_EQ(IterStatus(iterators[1]), "x->x");
+    }
+    ASSERT_EQ(IterStatus(iterators[2]), "v->t");
+
+    for (auto it : iterators) {
+      delete it;
+    }
+    Destroy();
+  }
+}
+
+TEST(ColumnFamilyTest, ReadOnlyDBTest) {
+  Open();
+  CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
+  ASSERT_OK(Put(1, "foo", "bla"));
+  ASSERT_OK(Put(2, "foo", "blabla"));
+  ASSERT_OK(Put(3, "foo", "blablabla"));
+  ASSERT_OK(Put(4, "foo", "blablablabla"));
+
+  DropColumnFamilies({2});
+  Close();
+  // open only a subset of column families
+  AssertOpenReadOnly({"default", "one", "four"});
+  ASSERT_EQ("NOT_FOUND", Get(0, "foo"));
+  ASSERT_EQ("bla", Get(1, "foo"));
+  ASSERT_EQ("blablablabla", Get(2, "foo"));
+
+  Close();
+  // can't open dropped column family
+  Status s = OpenReadOnly({"default", "one", "two"});
+  ASSERT_TRUE(!s.ok());
+
+  // Can't open without specifying default column family
+  s = OpenReadOnly({"one", "four"});
+  ASSERT_TRUE(!s.ok());
+}
+
+TEST(ColumnFamilyTest, DontRollEmptyLogs) {
+  Open();
+  CreateColumnFamiliesAndReopen({"one", "two", "three", "four"});
+
+  for (size_t i = 0; i < handles_.size(); ++i) {
+    PutRandomData(i, 10, 100);
+  }
+  int num_writable_file_start = env_->GetNumberOfNewWritableFileCalls();
+  // this will trigger the flushes
+  ASSERT_OK(db_->Write(WriteOptions(), nullptr));
+
+  for (int i = 0; i < 4; ++i) {
+    dbfull()->TEST_WaitForFlushMemTable(handles_[i]);
+  }
+  int total_new_writable_files =
+      env_->GetNumberOfNewWritableFileCalls() - num_writable_file_start;
+  ASSERT_EQ(static_cast<size_t>(total_new_writable_files), handles_.size() + 1);
+  Close();
+}
+
+TEST(ColumnFamilyTest, FlushStaleColumnFamilies) {
+  Open();
+  CreateColumnFamilies({"one", "two"});
+  ColumnFamilyOptions default_cf, one, two;
+  default_cf.write_buffer_size = 100000;  // small write buffer size
+  default_cf.disable_auto_compactions = true;
+  one.disable_auto_compactions = true;
+  two.disable_auto_compactions = true;
+  db_options_.max_total_wal_size = 210000;
+
+  Reopen({default_cf, one, two});
+
+  PutRandomData(2, 1, 10);  // 10 bytes
+  for (int i = 0; i < 2; ++i) {
+    PutRandomData(0, 100, 1000);  // flush
+    WaitForFlush(0);
+    ASSERT_EQ(i + 1, CountLiveFiles());
+  }
+  // third flush. now, CF [two] should be detected as stale and flushed
+  // column family 1 should not be flushed since it's empty
+  PutRandomData(0, 100, 1000);  // flush
+  WaitForFlush(0);
+  WaitForFlush(2);
+  // 3 files for default column families, 1 file for column family [two], zero
+  // files for column family [one], because it's empty
+  ASSERT_EQ(4, CountLiveFiles());
+  Close();
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  return rocksdb::test::RunAllTests();
+}
@@ -8,6 +8,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.

 #include "db/compaction.h"
+#include "db/column_family.h"

 namespace rocksdb {

@@ -29,6 +30,7 @@ Compaction::Compaction(Version* input_version, int level, int out_level,
      max_grandparent_overlap_bytes_(max_grandparent_overlap_bytes),
      input_version_(input_version),
      number_levels_(input_version_->NumberLevels()),
+      cfd_(input_version_->cfd_),
      seek_compaction_(seek_compaction),
      enable_compression_(enable_compression),
      grandparent_index_(0),
@@ -39,10 +41,13 @@ Compaction::Compaction(Version* input_version, int level, int out_level,
      score_(0),
      bottommost_level_(false),
      is_full_compaction_(false),
+      is_manual_compaction_(false),
      level_ptrs_(std::vector<size_t>(number_levels_)) {

+  cfd_->Ref();
  input_version_->Ref();
  edit_ = new VersionEdit();
+  edit_->SetColumnFamily(cfd_->GetID());
  for (int i = 0; i < number_levels_; i++) {
    level_ptrs_[i] = 0;
  }
@@ -53,6 +58,11 @@ Compaction::~Compaction() {
  if (input_version_ != nullptr) {
    input_version_->Unref();
  }
+  if (cfd_ != nullptr) {
+    if (cfd_->Unref()) {
+      delete cfd_;
+    }
+  }
 }

 bool Compaction::IsTrivialMove() const {
@@ -76,12 +86,11 @@ void Compaction::AddInputDeletions(VersionEdit* edit) {
 }

 bool Compaction::IsBaseLevelForKey(const Slice& user_key) {
-  if (input_version_->vset_->options_->compaction_style ==
-      kCompactionStyleUniversal) {
+  if (cfd_->options()->compaction_style == kCompactionStyleUniversal) {
    return bottommost_level_;
  }
  // Maybe use binary search to find right entry instead of linear search?
-  const Comparator* user_cmp = input_version_->vset_->icmp_.user_comparator();
+  const Comparator* user_cmp = cfd_->user_comparator();
  for (int lvl = level_ + 2; lvl < number_levels_; lvl++) {
    const std::vector<FileMetaData*>& files = input_version_->files_[lvl];
    for (; level_ptrs_[lvl] < files.size(); ) {
@@ -102,7 +111,7 @@ bool Compaction::IsBaseLevelForKey(const Slice& user_key) {

 bool Compaction::ShouldStopBefore(const Slice& internal_key) {
  // Scan to find earliest grandparent file that contains key.
-  const InternalKeyComparator* icmp = &input_version_->vset_->icmp_;
+  const InternalKeyComparator* icmp = &cfd_->internal_comparator();
  while (grandparent_index_ < grandparents_.size() &&
      icmp->Compare(internal_key,
                    grandparents_[grandparent_index_]->largest.Encode()) > 0) {
@@ -140,8 +149,7 @@ void Compaction::MarkFilesBeingCompacted(bool value) {

 // Is this compaction producing files at the bottommost level?
 void Compaction::SetupBottomMostLevel(bool isManual) {
-  if (input_version_->vset_->options_->compaction_style  ==
-         kCompactionStyleUniversal) {
+  if (cfd_->options()->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
@@ -154,8 +162,7 @@ void Compaction::SetupBottomMostLevel(bool isManual) {
    return;
  }
  bottommost_level_ = true;
-  int num_levels = input_version_->vset_->NumberLevels();
-  for (int i = output_level() + 1; i < num_levels; i++) {
+  for (int i = output_level() + 1; i < number_levels_; i++) {
    if (input_version_->NumLevelFiles(i) > 0) {
      bottommost_level_ = false;
      break;
@@ -168,47 +175,87 @@ void Compaction::ReleaseInputs() {
    input_version_->Unref();
    input_version_ = nullptr;
  }
+  if (cfd_ != nullptr) {
+    if (cfd_->Unref()) {
+      delete cfd_;
+    }
+    cfd_ = nullptr;
+  }
+}
+
+void Compaction::ReleaseCompactionFiles(Status status) {
+  cfd_->compaction_picker()->ReleaseCompactionFiles(this, status);
 }

 void Compaction::ResetNextCompactionIndex() {
  input_version_->ResetNextCompactionIndex(level_);
 }

-static void InputSummary(std::vector<FileMetaData*>& files, char* output,
+/*
+for sizes >=10TB, print "XXTB"
+for sizes >=10GB, print "XXGB"
+etc.
+*/
+static void FileSizeSummary(unsigned long long sz, char* output, int len) {
+  const unsigned long long ull10 = 10;
+  if (sz >= ull10<<40) {
+    snprintf(output, len, "%lluTB", sz>>40);
+  } else if (sz >= ull10<<30) {
+    snprintf(output, len, "%lluGB", sz>>30);
+  } else if (sz >= ull10<<20) {
+    snprintf(output, len, "%lluMB", sz>>20);
+  } else if (sz >= ull10<<10) {
+    snprintf(output, len, "%lluKB", sz>>10);
+  } else {
+    snprintf(output, len, "%lluB", sz);
+  }
+}
+
+static int InputSummary(std::vector<FileMetaData*>& files, char* output,
                         int len) {
+  *output = '\0';
  int write = 0;
  for (unsigned int i = 0; i < files.size(); i++) {
    int sz = len - write;
-    int ret = snprintf(output + write, sz, "%lu(%lu) ",
-        (unsigned long)files.at(i)->number,
-        (unsigned long)files.at(i)->file_size);
+    int ret;
+    char sztxt[16];
+    FileSizeSummary((unsigned long long)files.at(i)->file_size, sztxt, 16);
+    ret = snprintf(output + write, sz, "%lu(%s) ",
+                   (unsigned long)files.at(i)->number,
+                   sztxt);
    if (ret < 0 || ret >= sz)
      break;
    write += ret;
  }
+  return write;
 }

 void Compaction::Summary(char* output, int len) {
  int write = snprintf(output, len,
-      "Base version %lu Base level %d, seek compaction:%d, inputs:",
+      "Base version %lu Base level %d, seek compaction:%d, inputs: [",
      (unsigned long)input_version_->GetVersionNumber(),
      level_,
      seek_compaction_);
-  if (write < 0 || write > len) {
+  if (write < 0 || write >= len) {
    return;
  }

-  char level_low_summary[100];
-  InputSummary(inputs_[0], level_low_summary, sizeof(level_low_summary));
-  char level_up_summary[100];
-  if (inputs_[1].size()) {
-    InputSummary(inputs_[1], level_up_summary, sizeof(level_up_summary));
-  } else {
-    level_up_summary[0] = '\0';
+  write += InputSummary(inputs_[0], output+write, len-write);
+  if (write < 0 || write >= len) {
+    return;
  }

-  snprintf(output + write, len - write, "[%s],[%s]",
-      level_low_summary, level_up_summary);
+  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;
+  }
+
+  snprintf(output+write, len-write, "]");
 }

 }  // namespace rocksdb
@@ -13,6 +13,7 @@
 namespace rocksdb {

 class Version;
+class ColumnFamilyData;

 // A Compaction encapsulates information about a compaction.
 class Compaction {
@@ -36,6 +37,8 @@ class Compaction {
  // Returns input version of the compaction
  Version* input_version() const { return input_version_; }

+  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]; }

@@ -67,6 +70,10 @@ class Compaction {
  // is successful.
  void ReleaseInputs();

+  // Clear all files to indicate that they are not being compacted
+  // Delete this compaction from the list of running compactions.
+  void ReleaseCompactionFiles(Status status);
+
  void Summary(char* output, int len);

  // Return the score that was used to pick this compaction run.
@@ -78,6 +85,9 @@ class Compaction {
  // Does this compaction include all sst files?
  bool IsFullCompaction() { return is_full_compaction_; }

+  // Was this compaction triggered manually by the client?
+  bool IsManualCompaction() { return is_manual_compaction_; }
+
 private:
  friend class CompactionPicker;
  friend class UniversalCompactionPicker;
@@ -94,6 +104,7 @@ class Compaction {
  Version* input_version_;
  VersionEdit* edit_;
  int number_levels_;
+  ColumnFamilyData* cfd_;

  bool seek_compaction_;
  bool enable_compression_;
@@ -117,6 +128,9 @@ class Compaction {
  // Does this compaction include all sst files?
  bool is_full_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
@@ -10,6 +10,7 @@
 #include "db/compaction_picker.h"

 #include <limits>
+#include "util/log_buffer.h"
 #include "util/statistics.h"

 namespace rocksdb {
@@ -177,7 +178,12 @@ bool CompactionPicker::ExpandWhileOverlapping(Compaction* c) {
  // If, after the expansion, there are files that are already under
  // compaction, then we must drop/cancel this compaction.
  int parent_index = -1;
-  if (FilesInCompaction(c->inputs_[0]) ||
+  if (c->inputs_[0].empty()) {
+    Log(options_->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]) ||
      (c->level() != c->output_level() &&
       ParentRangeInCompaction(c->input_version_, &smallest, &largest, level,
                               &parent_index))) {
@@ -270,16 +276,13 @@ void CompactionPicker::SetupOtherInputs(Compaction* c) {
      if (expanded1.size() == c->inputs_[1].size() &&
          !FilesInCompaction(expanded1)) {
        Log(options_->info_log,
-            "Expanding@%lu %lu+%lu (%lu+%lu bytes) to %lu+%lu (%lu+%lu bytes)"
-            "\n",
-            (unsigned long)level,
+            "[%s] Expanding@%lu %lu+%lu (%lu+%lu bytes) to %lu+%lu (%lu+%lu "
+            "bytes)\n",
+            c->column_family_data()->GetName().c_str(), (unsigned long)level,
            (unsigned long)(c->inputs_[0].size()),
-            (unsigned long)(c->inputs_[1].size()),
-            (unsigned long)inputs0_size,
-            (unsigned long)inputs1_size,
-            (unsigned long)(expanded0.size()),
-            (unsigned long)(expanded1.size()),
-            (unsigned long)expanded0_size,
+            (unsigned long)(c->inputs_[1].size()), (unsigned long)inputs0_size,
+            (unsigned long)inputs1_size, (unsigned long)(expanded0.size()),
+            (unsigned long)(expanded1.size()), (unsigned long)expanded0_size,
            (unsigned long)inputs1_size);
        smallest = new_start;
        largest = new_limit;
@@ -344,7 +347,9 @@ Compaction* CompactionPicker::CompactRange(Version* version, int input_level,
  c->inputs_[0] = inputs;
  if (ExpandWhileOverlapping(c) == false) {
    delete c;
-    Log(options_->info_log, "Could not compact due to expansion failure.\n");
+    Log(options_->info_log,
+        "[%s] Could not compact due to expansion failure.\n",
+        version->cfd_->GetName().c_str());
    return nullptr;
  }

@@ -361,10 +366,14 @@ Compaction* CompactionPicker::CompactRange(Version* version, int input_level,

  // Is this compaction creating a file at the bottommost level
  c->SetupBottomMostLevel(true);
+
+  c->is_manual_compaction_ = true;
+
  return c;
 }

-Compaction* LevelCompactionPicker::PickCompaction(Version* version) {
+Compaction* LevelCompactionPicker::PickCompaction(Version* version,
+                                                  LogBuffer* log_buffer) {
  Compaction* c = nullptr;
  int level = -1;

@@ -372,7 +381,7 @@ 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->Finalize(size_being_compacted);
+  version->ComputeCompactionScore(size_being_compacted);

  // We prefer compactions triggered by too much data in a level over
  // the compactions triggered by seeks.
@@ -510,10 +519,6 @@ Compaction* LevelCompactionPicker::PickCompactionBySize(Version* version,
      nextIndex = i;
    }

-    //if (i > Version::number_of_files_to_sort_) {
-    //  Log(options_->info_log, "XXX Looking at index %d", i);
-    //}
-
    // Do not pick this file if its parents at level+1 are being compacted.
    // Maybe we can avoid redoing this work in SetupOtherInputs
    int parent_index = -1;
@@ -541,37 +546,49 @@ Compaction* LevelCompactionPicker::PickCompactionBySize(Version* version,
 // Universal style of compaction. Pick files that are contiguous in
 // time-range to compact.
 //
-Compaction* UniversalCompactionPicker::PickCompaction(Version* version) {
+Compaction* UniversalCompactionPicker::PickCompaction(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)) {
-    Log(options_->info_log, "Universal: nothing to do\n");
+    LogToBuffer(log_buffer, "[%s] Universal: nothing to do\n",
+                version->cfd_->GetName().c_str());
    return nullptr;
  }
  Version::FileSummaryStorage tmp;
-  Log(options_->info_log, "Universal: candidate files(%lu): %s\n",
-      version->files_[level].size(),
-      version->LevelFileSummary(&tmp, 0));
+  LogToBuffer(log_buffer, "[%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 = PickCompactionUniversalSizeAmp(version, score);
-  if (c == nullptr) {
-
+  Compaction* c;
+  if ((c = PickCompactionUniversalSizeAmp(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;
-    c = PickCompactionUniversalReadAmp(version, score, ratio, UINT_MAX);

-    // Size amplification and file size ratios are within configured limits.
-    // If max read amplification is exceeding configured limits, then force
-    // compaction without looking at filesize ratios and try to reduce
-    // the number of files to fewer than level0_file_num_compaction_trigger.
-    if (c == nullptr) {
+    if ((c = PickCompactionUniversalReadAmp(version, score, ratio, UINT_MAX,
+                                            log_buffer)) != nullptr) {
+      LogToBuffer(log_buffer, "[%s] Universal: compacting for size ratio\n",
+                  version->cfd_->GetName().c_str());
+    } else {
+      // Size amplification and file size ratios are within configured limits.
+      // If max read amplification is exceeding configured limits, then force
+      // compaction without looking at filesize ratios and try to reduce
+      // the number of files to fewer than level0_file_num_compaction_trigger.
      unsigned int num_files = version->files_[level].size() -
                               options_->level0_file_num_compaction_trigger;
-      c = PickCompactionUniversalReadAmp(version, score, UINT_MAX, num_files);
+      if ((c = PickCompactionUniversalReadAmp(
+               version, score, UINT_MAX, num_files, log_buffer)) != nullptr) {
+        LogToBuffer(log_buffer, "[%s] Universal: compacting for file num\n",
+                    version->cfd_->GetName().c_str());
+      }
    }
  }
  if (c == nullptr) {
@@ -623,7 +640,7 @@ Compaction* UniversalCompactionPicker::PickCompaction(Version* version) {
 //
 Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    Version* version, double score, unsigned int ratio,
-    unsigned int max_number_of_files_to_compact) {
+    unsigned int max_number_of_files_to_compact, LogBuffer* log_buffer) {
  int level = 0;

  unsigned int min_merge_width =
@@ -636,7 +653,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
  FileMetaData* f = nullptr;
  bool done = false;
  int start_index = 0;
-  unsigned int candidate_count;
+  unsigned int candidate_count = 0;
  assert(file_by_time.size() == version->files_[level].size());

  unsigned int max_files_to_compact = std::min(max_merge_width,
@@ -658,9 +675,9 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
        candidate_count = 1;
        break;
      }
-      Log(options_->info_log,
-          "Universal: file %lu[%d] being compacted, skipping",
-          (unsigned long)f->number, loop);
+      LogToBuffer(
+          log_buffer, "[%s] Universal: file %lu[%d] being compacted, skipping",
+          version->cfd_->GetName().c_str(), (unsigned long)f->number, loop);
      f = nullptr;
    }

@@ -668,8 +685,9 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    // first candidate to be compacted.
    uint64_t candidate_size =  f != nullptr? f->file_size : 0;
    if (f != nullptr) {
-      Log(options_->info_log, "Universal: Possible candidate file %lu[%d].",
-          (unsigned long)f->number, loop);
+      LogToBuffer(
+          log_buffer, "[%s] Universal: Possible candidate file %lu[%d].",
+          version->cfd_->GetName().c_str(), (unsigned long)f->number, loop);
    }

    // Check if the suceeding files need compaction.
@@ -681,14 +699,32 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
      if (f->being_compacted) {
        break;
      }
-      // pick files if the total candidate file size (increased by the
+      // Pick files if the total/last candidate file size (increased by the
      // specified ratio) is still larger than the next candidate file.
+      // candidate_size is the total size of files picked so far with the
+      // 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->file_size) {
        break;
      }
+      if (options_->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->file_size * (100L + ratio)) / 100;
+        if (sz < 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
+          // by the last-resort read amp strategy which disregards size ratios.
+          break;
+        }
+        candidate_size = f->file_size;
+      } else { // default kCompactionStopStyleTotalSize
+        candidate_size += f->file_size;
+      }
      candidate_count++;
-      candidate_size += f->file_size;
    }

    // Found a series of consecutive files that need compaction.
@@ -701,12 +737,10 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
           i < loop + candidate_count && i < file_by_time.size(); i++) {
       int index = file_by_time[i];
       FileMetaData* f = version->files_[level][index];
-       Log(options_->info_log,
-           "Universal: Skipping file %lu[%d] with size %lu %d\n",
-           (unsigned long)f->number,
-           i,
-           (unsigned long)f->file_size,
-           f->being_compacted);
+       LogToBuffer(log_buffer,
+                   "[%s] Universal: Skipping file %lu[%d] with size %lu %d\n",
+                   version->cfd_->GetName().c_str(), (unsigned long)f->number,
+                   i, (unsigned long)f->file_size, f->being_compacted);
      }
    }
  }
@@ -740,10 +774,10 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
    int index = file_by_time[i];
    FileMetaData* f = c->input_version_->files_[level][index];
    c->inputs_[0].push_back(f);
-    Log(options_->info_log, "Universal: Picking file %lu[%d] with size %lu\n",
-        (unsigned long)f->number,
-        i,
-        (unsigned long)f->file_size);
+    LogToBuffer(log_buffer,
+                "[%s] Universal: Picking file %lu[%d] with size %lu\n",
+                version->cfd_->GetName().c_str(), (unsigned long)f->number, i,
+                (unsigned long)f->file_size);
  }
  return c;
 }
@@ -755,7 +789,7 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalReadAmp(
 // min_merge_width and max_merge_width).
 //
 Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
-    Version* version, double score) {
+    Version* version, double score, LogBuffer* log_buffer) {
  int level = 0;

  // percentage flexibilty while reducing size amplification
@@ -779,20 +813,19 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
      start_index = loop;         // Consider this as the first candidate.
      break;
    }
-    Log(options_->info_log, "Universal: skipping file %lu[%d] compacted %s",
-        (unsigned long)f->number,
-        loop,
-        " cannot be a candidate to reduce size amp.\n");
+    LogToBuffer(log_buffer,
+                "[%s] Universal: skipping file %lu[%d] compacted %s",
+                version->cfd_->GetName().c_str(), (unsigned long)f->number,
+                loop, " cannot be a candidate to reduce size amp.\n");
    f = nullptr;
  }
  if (f == nullptr) {
    return nullptr;             // no candidate files
  }

-  Log(options_->info_log, "Universal: First candidate file %lu[%d] %s",
-      (unsigned long)f->number,
-      start_index,
-      " to reduce size amp.\n");
+  LogToBuffer(log_buffer, "[%s] Universal: First candidate file %lu[%d] %s",
+              version->cfd_->GetName().c_str(), (unsigned long)f->number,
+              start_index, " to reduce size amp.\n");

  // keep adding up all the remaining files
  for (unsigned int loop = start_index; loop < file_by_time.size() - 1;
@@ -800,10 +833,9 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
    int index = file_by_time[loop];
    f = version->files_[level][index];
    if (f->being_compacted) {
-      Log(options_->info_log,
-          "Universal: Possible candidate file %lu[%d] %s.",
-          (unsigned long)f->number,
-          loop,
+      LogToBuffer(
+          log_buffer, "[%s] Universal: Possible candidate file %lu[%d] %s.",
+          version->cfd_->GetName().c_str(), (unsigned long)f->number, loop,
          " is already being compacted. No size amp reduction possible.\n");
      return nullptr;
    }
@@ -820,18 +852,19 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(

  // size amplification = percentage of additional size
  if (candidate_size * 100 < ratio * earliest_file_size) {
-    Log(options_->info_log,
-        "Universal: size amp not needed. newer-files-total-size %lu "
+    LogToBuffer(
+        log_buffer,
+        "[%s] Universal: size amp not needed. newer-files-total-size %lu "
        "earliest-file-size %lu",
-        (unsigned long)candidate_size,
+        version->cfd_->GetName().c_str(), (unsigned long)candidate_size,
        (unsigned long)earliest_file_size);
    return nullptr;
  } else {
-    Log(options_->info_log,
-        "Universal: size amp needed. newer-files-total-size %lu "
-        "earliest-file-size %lu",
-        (unsigned long)candidate_size,
-        (unsigned long)earliest_file_size);
+    LogToBuffer(log_buffer,
+                "[%s] Universal: size amp needed. newer-files-total-size %lu "
+                "earliest-file-size %lu",
+                version->cfd_->GetName().c_str(), (unsigned long)candidate_size,
+                (unsigned long)earliest_file_size);
  }
  assert(start_index >= 0 && start_index < file_by_time.size() - 1);

@@ -845,11 +878,10 @@ Compaction* UniversalCompactionPicker::PickCompactionUniversalSizeAmp(
    int index = file_by_time[loop];
    f = c->input_version_->files_[level][index];
    c->inputs_[0].push_back(f);
-    Log(options_->info_log,
-        "Universal: size amp picking file %lu[%d] with size %lu",
-        (unsigned long)f->number,
-        index,
-        (unsigned long)f->file_size);
+    LogToBuffer(log_buffer,
+                "[%s] Universal: size amp picking file %lu[%d] with size %lu",
+                version->cfd_->GetName().c_str(), (unsigned long)f->number,
+                index, (unsigned long)f->file_size);
  }
  return c;
 }
@@ -12,6 +12,7 @@
 #include "db/compaction.h"
 #include "rocksdb/status.h"
 #include "rocksdb/options.h"
+#include "rocksdb/env.h"

 #include <vector>
 #include <memory>
@@ -19,6 +20,7 @@

 namespace rocksdb {

+class LogBuffer;
 class Compaction;
 class Version;

@@ -31,7 +33,8 @@ class CompactionPicker {
  // 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) = 0;
+  virtual Compaction* PickCompaction(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
@@ -116,6 +119,7 @@ class CompactionPicker {
  std::unique_ptr<uint64_t[]> level_max_bytes_;

  const Options* const options_;
+
 private:
  int num_levels_;

@@ -127,16 +131,19 @@ class UniversalCompactionPicker : public CompactionPicker {
  UniversalCompactionPicker(const Options* options,
                            const InternalKeyComparator* icmp)
      : CompactionPicker(options, icmp) {}
-  virtual Compaction* PickCompaction(Version* version) override;
+  virtual Compaction* PickCompaction(Version* version,
+                                     LogBuffer* log_buffer) override;

 private:
  // Pick Universal compaction to limit read amplification
  Compaction* PickCompactionUniversalReadAmp(Version* version, double score,
                                             unsigned int ratio,
-                                             unsigned int num_files);
+                                             unsigned int num_files,
+                                             LogBuffer* log_buffer);

  // Pick Universal compaction to limit space amplification.
-  Compaction* PickCompactionUniversalSizeAmp(Version* version, double score);
+  Compaction* PickCompactionUniversalSizeAmp(Version* version, double score,
+                                             LogBuffer* log_buffer);
 };

 class LevelCompactionPicker : public CompactionPicker {
@@ -144,7 +151,8 @@ class LevelCompactionPicker : public CompactionPicker {
  LevelCompactionPicker(const Options* options,
                        const InternalKeyComparator* icmp)
      : CompactionPicker(options, icmp) {}
-  virtual Compaction* PickCompaction(Version* version) override;
+  virtual Compaction* PickCompaction(Version* version,
+                                     LogBuffer* log_buffer) override;

 private:
  // For the specfied level, pick a compaction.
@@ -40,7 +40,7 @@ class CorruptionTest {
  CorruptionTest() {
    tiny_cache_ = NewLRUCache(100);
    options_.env = &env_;
-    dbname_ = test::TmpDir() + "/db_test";
+    dbname_ = test::TmpDir() + "/corruption_test";
    DestroyDB(dbname_, options_);

    db_ = nullptr;
@@ -95,7 +95,12 @@ class CorruptionTest {
    int bad_values = 0;
    int correct = 0;
    std::string value_space;
-    Iterator* iter = db_->NewIterator(ReadOptions());
+    // Do not verify checksums. If we verify checksums then the
+    // db itself will raise errors because data is corrupted.
+    // Instead, we want the reads to be successful and this test
+    // will detect whether the appropriate corruptions have
+    // occured.
+    Iterator* iter = db_->NewIterator(ReadOptions(false, true));
    for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
      uint64_t key;
      Slice in(iter->key());
@@ -122,24 +127,7 @@ class CorruptionTest {
    ASSERT_GE(max_expected, correct);
  }

-  void Corrupt(FileType filetype, int offset, int bytes_to_corrupt) {
-    // Pick file to corrupt
-    std::vector<std::string> filenames;
-    ASSERT_OK(env_.GetChildren(dbname_, &filenames));
-    uint64_t number;
-    FileType type;
-    std::string fname;
-    int picked_number = -1;
-    for (unsigned int i = 0; i < filenames.size(); i++) {
-      if (ParseFileName(filenames[i], &number, &type) &&
-          type == filetype &&
-          int(number) > picked_number) {  // Pick latest file
-        fname = dbname_ + "/" + filenames[i];
-        picked_number = number;
-      }
-    }
-    ASSERT_TRUE(!fname.empty()) << filetype;
-
+  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);
@@ -172,6 +160,42 @@ class CorruptionTest {
    ASSERT_TRUE(s.ok()) << s.ToString();
  }

+  void Corrupt(FileType filetype, int offset, int bytes_to_corrupt) {
+    // Pick file to corrupt
+    std::vector<std::string> filenames;
+    ASSERT_OK(env_.GetChildren(dbname_, &filenames));
+    uint64_t number;
+    FileType type;
+    std::string fname;
+    int picked_number = -1;
+    for (unsigned int i = 0; i < filenames.size(); i++) {
+      if (ParseFileName(filenames[i], &number, &type) &&
+          type == filetype &&
+          static_cast<int>(number) > picked_number) {  // Pick latest file
+        fname = dbname_ + "/" + filenames[i];
+        picked_number = number;
+      }
+    }
+    ASSERT_TRUE(!fname.empty()) << filetype;
+
+    CorruptFile(fname, offset, bytes_to_corrupt);
+  }
+
+  // corrupts exactly one file at level `level`. if no file found at level,
+  // asserts
+  void CorruptTableFileAtLevel(int level, int offset, int bytes_to_corrupt) {
+    std::vector<LiveFileMetaData> metadata;
+    db_->GetLiveFilesMetaData(&metadata);
+    for (const auto& m : metadata) {
+      if (m.level == level) {
+        CorruptFile(dbname_ + "/" + m.name, offset, bytes_to_corrupt);
+        return;
+      }
+    }
+    ASSERT_TRUE(false) << "no file found at level";
+  }
+
+
  int Property(const std::string& name) {
    std::string property;
    int result;
@@ -321,30 +345,35 @@ TEST(CorruptionTest, CompactionInputError) {
 TEST(CorruptionTest, CompactionInputErrorParanoid) {
  Options options;
  options.paranoid_checks = true;
-  options.write_buffer_size = 1048576;
+  options.write_buffer_size = 131072;
+  options.max_write_buffer_number = 2;
  Reopen(&options);
  DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);

-  // Fill levels >= 1 so memtable compaction outputs to level 1
+  // Fill levels >= 1 so memtable flush outputs to level 0
  for (int level = 1; level < dbi->NumberLevels(); level++) {
    dbi->Put(WriteOptions(), "", "begin");
    dbi->Put(WriteOptions(), "~", "end");
    dbi->TEST_FlushMemTable();
  }

+  options.max_mem_compaction_level = 0;
+  Reopen(&options);
+
+  dbi = reinterpret_cast<DBImpl*>(db_);
  Build(10);
  dbi->TEST_FlushMemTable();
  dbi->TEST_WaitForCompact();
  ASSERT_EQ(1, Property("rocksdb.num-files-at-level0"));

-  Corrupt(kTableFile, 100, 1);
+  CorruptTableFileAtLevel(0, 100, 1);
  Check(9, 9);

  // Write must eventually fail because of corrupted table
  Status s;
  std::string tmp1, tmp2;
  bool failed = false;
-  for (int i = 0; i < 10000 && s.ok(); i++) {
+  for (int i = 0; i < 10000; i++) {
    s = db_->Put(WriteOptions(), Key(i, &tmp1), Value(i, &tmp2));
    if (!s.ok()) {
      failed = true;
@@ -371,6 +400,39 @@ TEST(CorruptionTest, UnrelatedKeys) {
  ASSERT_EQ(Value(1000, &tmp2).ToString(), v);
 }

+TEST(CorruptionTest, FileSystemStateCorrupted) {
+  for (int iter = 0; iter < 2; ++iter) {
+    Options options;
+    options.paranoid_checks = true;
+    options.create_if_missing = true;
+    Reopen(&options);
+    Build(10);
+    ASSERT_OK(db_->Flush(FlushOptions()));
+    DBImpl* dbi = reinterpret_cast<DBImpl*>(db_);
+    std::vector<LiveFileMetaData> metadata;
+    dbi->GetLiveFilesMetaData(&metadata);
+    ASSERT_GT(metadata.size(), size_t(0));
+    std::string filename = dbname_ + metadata[0].name;
+
+    delete db_;
+    db_ = nullptr;
+
+    if (iter == 0) {  // corrupt file size
+      unique_ptr<WritableFile> file;
+      env_.NewWritableFile(filename, &file, EnvOptions());
+      file->Append(Slice("corrupted sst"));
+      file.reset();
+    } else {  // delete the file
+      env_.DeleteFile(filename);
+    }
+
+    Status x = TryReopen(&options);
+    ASSERT_TRUE(x.IsCorruption());
+    DestroyDB(dbname_, options_);
+    Reopen(&options);
+  }
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -7,6 +7,10 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.

+#ifndef ROCKSDB_LITE
+
+#define __STDC_FORMAT_MACROS
+#include <inttypes.h>
 #include <algorithm>
 #include <string>
 #include <stdint.h>
@@ -17,6 +21,7 @@
 #include "rocksdb/env.h"
 #include "port/port.h"
 #include "util/mutexlock.h"
+#include "util/sync_point.h"

 namespace rocksdb {

@@ -60,21 +65,36 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,

  *manifest_file_size = 0;

+  mutex_.Lock();
+
  if (flush_memtable) {
    // flush all dirty data to disk.
-    Status status =  Flush(FlushOptions());
+    Status status;
+    for (auto cfd : *versions_->GetColumnFamilySet()) {
+      cfd->Ref();
+      mutex_.Unlock();
+      status = FlushMemTable(cfd, FlushOptions());
+      mutex_.Lock();
+      cfd->Unref();
+      if (!status.ok()) {
+        break;
+      }
+    }
+    versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
+
    if (!status.ok()) {
+      mutex_.Unlock();
      Log(options_.info_log, "Cannot Flush data %s\n",
          status.ToString().c_str());
      return status;
    }
  }

-  MutexLock l(&mutex_);
-
  // Make a set of all of the live *.sst files
  std::set<uint64_t> live;
-  versions_->current()->AddLiveFiles(&live);
+  for (auto cfd : *versions_->GetColumnFamilySet()) {
+    cfd->current()->AddLiveFiles(&live);
+  }

  ret.clear();
  ret.reserve(live.size() + 2); //*.sst + CURRENT + MANIFEST
@@ -91,24 +111,62 @@ Status DBImpl::GetLiveFiles(std::vector<std::string>& ret,
  // find length of manifest file while holding the mutex lock
  *manifest_file_size = versions_->ManifestFileSize();

+  mutex_.Unlock();
  return Status::OK();
 }

 Status DBImpl::GetSortedWalFiles(VectorLogPtr& files) {
-  // First get sorted files in archive dir, then append sorted files from main
-  // dir to maintain sorted order
-
-  // list wal files in archive dir.
+  // First get sorted files in db dir, then get sorted files from archived
+  // dir, to avoid a race condition where a log file is moved to archived
+  // dir in between.
  Status s;
+  // list wal files in main db dir.
+  VectorLogPtr logs;
+  s = GetSortedWalsOfType(options_.wal_dir, logs, kAliveLogFile);
+  if (!s.ok()) {
+    return s;
+  }
+
+  // Reproduce the race condition where a log file is moved
+  // to archived dir, between these two sync points, used in
+  // (DBTest,TransactionLogIteratorRace)
+  TEST_SYNC_POINT("DBImpl::GetSortedWalFiles:1");
+  TEST_SYNC_POINT("DBImpl::GetSortedWalFiles:2");
+
+  files.clear();
+  // list wal files in archive dir.
  std::string archivedir = ArchivalDirectory(options_.wal_dir);
  if (env_->FileExists(archivedir)) {
-    s = AppendSortedWalsOfType(archivedir, files, kArchivedLogFile);
+    s = GetSortedWalsOfType(archivedir, files, kArchivedLogFile);
    if (!s.ok()) {
      return s;
    }
  }
-  // list wal files in main db dir.
-  return AppendSortedWalsOfType(options_.wal_dir, files, kAliveLogFile);
+
+  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,
+        latest_archived_log_number);
+  }
+
+  files.reserve(files.size() + logs.size());
+  for (auto& log : logs) {
+    if (log->LogNumber() > latest_archived_log_number) {
+      files.push_back(std::move(log));
+    } else {
+      // When the race condition happens, we could see the
+      // 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->PathName().c_str());
+    }
+  }
+
+  return s;
 }

 }
+
+#endif  // ROCKSDB_LITE
@@ -7,23 +7,29 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 #pragma once
+
 #include <atomic>
 #include <deque>
 #include <set>
 #include <utility>
 #include <vector>
+#include <string>
+
 #include "db/dbformat.h"
 #include "db/log_writer.h"
 #include "db/snapshot.h"
+#include "db/column_family.h"
 #include "db/version_edit.h"
+#include "memtable_list.h"
+#include "port/port.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
 #include "rocksdb/memtablerep.h"
 #include "rocksdb/transaction_log.h"
-#include "port/port.h"
-#include "util/stats_logger.h"
-#include "memtablelist.h"
 #include "util/autovector.h"
+#include "util/stats_logger.h"
+#include "util/thread_local.h"
+#include "db/internal_stats.h"

 namespace rocksdb {

@@ -32,47 +38,87 @@ class TableCache;
 class Version;
 class VersionEdit;
 class VersionSet;
+class CompactionFilterV2;

 class DBImpl : public DB {
 public:
-  DBImpl(const Options& options, const std::string& dbname);
+  DBImpl(const DBOptions& options, const std::string& dbname);
  virtual ~DBImpl();

  // Implementations of the DB interface
-  virtual Status Put(const WriteOptions&, const Slice& key, const Slice& value);
-  virtual Status Merge(const WriteOptions&, const Slice& key,
+  using DB::Put;
+  virtual Status Put(const WriteOptions& options,
+                     ColumnFamilyHandle* column_family, const Slice& key,
+                     const Slice& value);
+  using DB::Merge;
+  virtual Status Merge(const WriteOptions& options,
+                       ColumnFamilyHandle* column_family, const Slice& key,
                       const Slice& value);
-  virtual Status Delete(const WriteOptions&, const Slice& key);
+  using DB::Delete;
+  virtual Status Delete(const WriteOptions& options,
+                        ColumnFamilyHandle* column_family, const Slice& key);
+  using DB::Write;
  virtual Status Write(const WriteOptions& options, WriteBatch* updates);
+  using DB::Get;
  virtual Status Get(const ReadOptions& options,
-                     const Slice& key,
+                     ColumnFamilyHandle* column_family, const Slice& key,
                     std::string* value);
-  virtual std::vector<Status> MultiGet(const ReadOptions& options,
-                                       const std::vector<Slice>& keys,
-                                       std::vector<std::string>* values);
+  using DB::MultiGet;
+  virtual std::vector<Status> MultiGet(
+      const ReadOptions& options,
+      const std::vector<ColumnFamilyHandle*>& column_family,
+      const std::vector<Slice>& keys, std::vector<std::string>* values);
+
+  virtual Status CreateColumnFamily(const ColumnFamilyOptions& options,
+                                    const std::string& column_family,
+                                    ColumnFamilyHandle** handle);
+  virtual Status DropColumnFamily(ColumnFamilyHandle* column_family);

  // Returns false if key doesn't exist in the database and true if it may.
  // If value_found is not passed in as null, then return the value if found in
  // memory. On return, if value was found, then value_found will be set to true
  // , otherwise false.
+  using DB::KeyMayExist;
  virtual bool KeyMayExist(const ReadOptions& options,
-                           const Slice& key,
-                           std::string* value,
-                           bool* value_found = nullptr);
-  virtual Iterator* NewIterator(const ReadOptions&);
+                           ColumnFamilyHandle* column_family, const Slice& key,
+                           std::string* value, bool* value_found = nullptr);
+  using DB::NewIterator;
+  virtual Iterator* NewIterator(const ReadOptions& options,
+                                ColumnFamilyHandle* column_family);
+  virtual Status NewIterators(
+      const ReadOptions& options,
+      const std::vector<ColumnFamilyHandle*>& column_families,
+      std::vector<Iterator*>* iterators);
  virtual const Snapshot* GetSnapshot();
  virtual void ReleaseSnapshot(const Snapshot* snapshot);
-  virtual bool GetProperty(const Slice& property, std::string* value);
-  virtual void GetApproximateSizes(const Range* range, int n, uint64_t* sizes);
-  virtual Status CompactRange(const Slice* begin, const Slice* end,
+  using DB::GetProperty;
+  virtual bool GetProperty(ColumnFamilyHandle* column_family,
+                           const Slice& property, std::string* value);
+  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);
-  virtual int NumberLevels();
-  virtual int MaxMemCompactionLevel();
-  virtual int Level0StopWriteTrigger();
+
+  using DB::NumberLevels;
+  virtual int NumberLevels(ColumnFamilyHandle* column_family);
+  using DB::MaxMemCompactionLevel;
+  virtual int MaxMemCompactionLevel(ColumnFamilyHandle* column_family);
+  using DB::Level0StopWriteTrigger;
+  virtual int Level0StopWriteTrigger(ColumnFamilyHandle* column_family);
  virtual const std::string& GetName() const;
  virtual Env* GetEnv() const;
-  virtual const Options& GetOptions() const;
-  virtual Status Flush(const FlushOptions& options);
+  using DB::GetOptions;
+  virtual const Options& GetOptions(ColumnFamilyHandle* column_family) const;
+  using DB::Flush;
+  virtual Status Flush(const FlushOptions& options,
+                       ColumnFamilyHandle* column_family);
+
+  virtual SequenceNumber GetLatestSequenceNumber() const;
+
+#ifndef ROCKSDB_LITE
  virtual Status DisableFileDeletions();
  virtual Status EnableFileDeletions(bool force);
  // All the returned filenames start with "/"
@@ -80,33 +126,39 @@ class DBImpl : public DB {
                              uint64_t* manifest_file_size,
                              bool flush_memtable = true);
  virtual Status GetSortedWalFiles(VectorLogPtr& files);
-  virtual SequenceNumber GetLatestSequenceNumber() const;
-  virtual Status GetUpdatesSince(SequenceNumber seq_number,
-                                 unique_ptr<TransactionLogIterator>* iter);
+
+  virtual Status GetUpdatesSince(
+      SequenceNumber seq_number, unique_ptr<TransactionLogIterator>* iter,
+      const TransactionLogIterator::ReadOptions&
+          read_options = TransactionLogIterator::ReadOptions());
  virtual Status DeleteFile(std::string name);

-  virtual void GetLiveFilesMetaData(
-    std::vector<LiveFileMetaData> *metadata);
+  virtual void GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata);
+#endif  // ROCKSDB_LITE
+
+  // checks if all live files exist on file system and that their file sizes
+  // match to our in-memory records
+  virtual Status CheckConsistency();

  virtual Status GetDbIdentity(std::string& identity);

-  Status RunManualCompaction(int input_level,
-                             int output_level,
-                             const Slice* begin,
+  Status RunManualCompaction(ColumnFamilyData* cfd, int input_level,
+                             int output_level, const Slice* begin,
                             const Slice* end);

+#ifndef ROCKSDB_LITE
  // Extra methods (for testing) that are not in the public DB interface
+  // Implemented in db_impl_debug.cc

  // Compact any files in the named level that overlap [*begin, *end]
-  Status TEST_CompactRange(int level,
-                           const Slice* begin,
-                           const Slice* end);
+  Status TEST_CompactRange(int level, const Slice* begin, const Slice* end,
+                           ColumnFamilyHandle* column_family = nullptr);

  // Force current memtable contents to be flushed.
-  Status TEST_FlushMemTable();
+  Status TEST_FlushMemTable(bool wait = true);

  // Wait for memtable compaction
-  Status TEST_WaitForFlushMemTable();
+  Status TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family = nullptr);

  // Wait for any compaction
  Status TEST_WaitForCompact();
@@ -114,14 +166,13 @@ 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();
+  Iterator* TEST_NewInternalIterator(ColumnFamilyHandle* column_family =
+                                         nullptr);

  // Return the maximum overlapping data (in bytes) at next level for any
  // file at a level >= 1.
-  int64_t TEST_MaxNextLevelOverlappingBytes();
-
-  // Simulate a db crash, no elegant closing of database.
-  void TEST_Destroy_DBImpl();
+  int64_t TEST_MaxNextLevelOverlappingBytes(ColumnFamilyHandle* column_family =
+                                                nullptr);

  // Return the current manifest file no.
  uint64_t TEST_Current_Manifest_FileNo();
@@ -137,38 +188,19 @@ class DBImpl : public DB {
    default_interval_to_delete_obsolete_WAL_ = default_interval_to_delete_obsolete_WAL;
  }

-  // holds references to memtable, all immutable memtables and version
-  struct SuperVersion {
-    MemTable* mem;
-    MemTableListVersion* imm;
-    Version* current;
-    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
-    // delete all those memtables outside of mutex, during destruction
-    std::vector<MemTable*> to_delete;
+  void TEST_GetFilesMetaData(ColumnFamilyHandle* column_family,
+                             std::vector<std::vector<FileMetaData>>* metadata);

-    // should be called outside the mutex
-    explicit SuperVersion(const int num_memtables = 0);
-    ~SuperVersion();
-    SuperVersion* Ref();
-    // Returns true if this was the last reference and caller should
-    // call Clenaup() and delete the object
-    bool Unref();
+  Status TEST_ReadFirstRecord(const WalFileType type, const uint64_t number,
+                              SequenceNumber* sequence);

-    // call these two methods with db mutex held
-    // Cleanup unrefs mem, imm and current. Also, it stores all memtables
-    // that needs to be deleted in to_delete vector. Unrefing those
-    // objects needs to be done in the mutex
-    void Cleanup();
-    void Init(MemTable* new_mem, MemTableListVersion* new_imm,
-              Version* new_current);
-  };
+  Status TEST_ReadFirstLine(const std::string& fname, SequenceNumber* sequence);
+#endif  // NDEBUG

  // needed for CleanupIteratorState
  struct DeletionState {
    inline bool HaveSomethingToDelete() const {
-      return  all_files.size() ||
+      return  candidate_files.size() ||
        sst_delete_files.size() ||
        log_delete_files.size();
    }
@@ -176,7 +208,7 @@ class DBImpl : public DB {
    // a list of all files that we'll consider deleting
    // (every once in a while this is filled up with all files
    // in the DB directory)
-    std::vector<std::string> all_files;
+    std::vector<std::string> candidate_files;

    // the list of all live sst files that cannot be deleted
    std::vector<uint64_t> sst_live;
@@ -188,25 +220,23 @@ class DBImpl : public DB {
    std::vector<uint64_t> log_delete_files;

    // a list of memtables to be free
-    std::vector<MemTable *> memtables_to_free;
+    autovector<MemTable*> memtables_to_free;

-    SuperVersion* superversion_to_free; // if nullptr nothing to free
+    autovector<SuperVersion*> superversions_to_free;

-    SuperVersion* new_superversion; // if nullptr no new superversion
+    SuperVersion* new_superversion;  // if nullptr no new superversion

    // the current manifest_file_number, log_number and prev_log_number
    // that corresponds to the set of files in 'live'.
-    uint64_t manifest_file_number, log_number, prev_log_number;
+    uint64_t manifest_file_number, pending_manifest_file_number, log_number,
+        prev_log_number;

-    explicit DeletionState(const int num_memtables = 0,
-                           bool create_superversion = false) {
+    explicit DeletionState(bool create_superversion = false) {
      manifest_file_number = 0;
+      pending_manifest_file_number = 0;
      log_number = 0;
      prev_log_number = 0;
-      memtables_to_free.reserve(num_memtables);
-      superversion_to_free = nullptr;
-      new_superversion =
-          create_superversion ? new SuperVersion(num_memtables) : nullptr;
+      new_superversion = create_superversion ? new SuperVersion() : nullptr;
    }

    ~DeletionState() {
@@ -214,8 +244,10 @@ class DBImpl : public DB {
      for (auto m : memtables_to_free) {
        delete m;
      }
-      // free superversion. if nullptr, this will be noop
-      delete superversion_to_free;
+      // free superversions
+      for (auto s : superversions_to_free) {
+        delete s;
+      }
      // if new_superversion was not used, it will be non-nullptr and needs
      // to be freed here
      delete new_superversion;
@@ -223,7 +255,7 @@ class DBImpl : public DB {
  };

  // Returns the list of live files in 'live' and the list
-  // of all files in the filesystem in 'all_files'.
+  // 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,
  // it will not fill up the deletion_state
@@ -237,35 +269,34 @@ class DBImpl : public DB {
  // It is not necessary to hold the mutex when invoking this method.
  void PurgeObsoleteFiles(DeletionState& deletion_state);

+  ColumnFamilyHandle* DefaultColumnFamily() const;
+
 protected:
  Env* const env_;
  const std::string dbname_;
  unique_ptr<VersionSet> versions_;
-  const InternalKeyComparator internal_comparator_;
-  const Options options_;  // options_.comparator == &internal_comparator_
+  const DBOptions options_;

-  const Comparator* user_comparator() const {
-    return internal_comparator_.user_comparator();
-  }
-
-  MemTable* GetMemTable() {
-    return mem_;
-  }
-
-  Iterator* NewInternalIterator(const ReadOptions&,
-                                SequenceNumber* latest_snapshot);
+  Iterator* NewInternalIterator(const ReadOptions&, ColumnFamilyData* cfd,
+                                SuperVersion* super_version);

 private:
  friend class DB;
+  friend class InternalStats;
+#ifndef ROCKSDB_LITE
  friend class TailingIterator;
+#endif
+  friend struct SuperVersion;
  struct CompactionState;
  struct Writer;

  Status NewDB();

  // Recover the descriptor from persistent storage.  May do a significant
-  // amount of work to recover recently logged updates.
-  Status Recover(bool read_only = false, bool error_if_log_file_exist = false);
+  // amount of work to recover recently logged updates.  Any changes to
+  // be made to the descriptor are added to *edit.
+  Status Recover(const std::vector<ColumnFamilyDescriptor>& column_families,
+                 bool read_only = false, bool error_if_log_file_exist = false);

  void MaybeIgnoreError(Status* s) const;

@@ -276,8 +307,9 @@ 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(bool* madeProgress,
-                                   DeletionState& deletion_state);
+  Status FlushMemTableToOutputFile(ColumnFamilyData* cfd, bool* madeProgress,
+                                   DeletionState& deletion_state,
+                                   LogBuffer* log_buffer);

  Status RecoverLogFile(uint64_t log_number, SequenceNumber* max_sequence,
                        bool read_only);
@@ -287,97 +319,129 @@ class DBImpl : public DB {
  // database is opened) and is heavyweight because it holds the mutex
  // for the entire period. The second method WriteLevel0Table supports
  // concurrent flush memtables to storage.
-  Status WriteLevel0TableForRecovery(MemTable* mem, VersionEdit* edit);
-  Status WriteLevel0Table(std::vector<MemTable*> &mems, VersionEdit* edit,
-                                uint64_t* filenumber);
+  Status WriteLevel0TableForRecovery(ColumnFamilyData* cfd, MemTable* mem,
+                                     VersionEdit* edit);
+  Status WriteLevel0Table(ColumnFamilyData* cfd, autovector<MemTable*>& mems,
+                          VersionEdit* edit, uint64_t* filenumber,
+                          LogBuffer* log_buffer);

  uint64_t SlowdownAmount(int n, double bottom, double top);
-  // MakeRoomForWrite will return superversion_to_free through an arugment,
-  // which the caller needs to delete. We do it because caller can delete
-  // the superversion outside of mutex
-  Status MakeRoomForWrite(bool force /* compact even if there is room? */,
-                          SuperVersion** superversion_to_free);
+
+  // 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);
+
  void BuildBatchGroup(Writer** last_writer,
                       autovector<WriteBatch*>* write_batch_group);

  // Force current memtable contents to be flushed.
-  Status FlushMemTable(const FlushOptions& options);
+  Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options);

  // Wait for memtable flushed
-  Status WaitForFlushMemTable();
+  Status WaitForFlushMemTable(ColumnFamilyData* cfd);

  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);
  void BackgroundCallCompaction();
  void BackgroundCallFlush();
-  Status BackgroundCompaction(bool* madeProgress,DeletionState& deletion_state);
-  Status BackgroundFlush(bool* madeProgress, DeletionState& deletion_state);
+  Status BackgroundCompaction(bool* madeProgress, DeletionState& deletion_state,
+                              LogBuffer* log_buffer);
+  Status BackgroundFlush(bool* madeProgress, DeletionState& deletion_state,
+                         LogBuffer* log_buffer);
  void CleanupCompaction(CompactionState* compact, Status status);
  Status DoCompactionWork(CompactionState* compact,
-                          DeletionState& deletion_state);
+                          DeletionState& deletion_state,
+                          LogBuffer* log_buffer);
+
+  // This function is called as part of compaction. It enables Flush process to
+  // preempt compaction, since it's higher prioirty
+  // Returns: micros spent executing
+  uint64_t CallFlushDuringCompaction(ColumnFamilyData* cfd,
+                                     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(
+    SequenceNumber visible_at_tip,
+    SequenceNumber earliest_snapshot,
+    SequenceNumber latest_snapshot,
+    DeletionState& deletion_state,
+    bool bottommost_level,
+    int64_t& imm_micros,
+    Iterator* input,
+    CompactionState* compact,
+    bool is_compaction_v2,
+    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 FinishCompactionOutputFile(CompactionState* compact, Iterator* input);
-  Status InstallCompactionResults(CompactionState* compact);
+  Status InstallCompactionResults(CompactionState* compact,
+                                  LogBuffer* log_buffer);
  void AllocateCompactionOutputFileNumbers(CompactionState* compact);
  void ReleaseCompactionUnusedFileNumbers(CompactionState* compact);

+#ifdef ROCKSDB_LITE
+  void PurgeObsoleteWALFiles() {
+    // this function is used for archiving WAL files. we don't need this in
+    // ROCKSDB_LITE
+  }
+#else
  void PurgeObsoleteWALFiles();

-  Status AppendSortedWalsOfType(const std::string& path,
-                                VectorLogPtr& log_files,
-                                WalFileType type);
+  Status GetSortedWalsOfType(const std::string& path,
+                             VectorLogPtr& log_files,
+                             WalFileType type);

  // Requires: all_logs should be sorted with earliest log file first
  // Retains all log files in all_logs which contain updates with seq no.
  // Greater Than or Equal to the requested SequenceNumber.
  Status RetainProbableWalFiles(VectorLogPtr& all_logs,
                                const SequenceNumber target);
-  //  return true if
-  bool CheckWalFileExistsAndEmpty(const WalFileType type,
-                                  const uint64_t number);

  Status ReadFirstRecord(const WalFileType type, const uint64_t number,
-                         WriteBatch* const result);
+                         SequenceNumber* sequence);

-  Status ReadFirstLine(const std::string& fname, WriteBatch* const batch);
+  Status ReadFirstLine(const std::string& fname, SequenceNumber* sequence);
+#endif  // ROCKSDB_LITE

  void PrintStatistics();

  // dump rocksdb.stats to LOG
  void MaybeDumpStats();

+  // Return true if the current db supports snapshot.  If the current
+  // DB does not support snapshot, then calling GetSnapshot() will always
+  // return nullptr.
+  //
+  // @see GetSnapshot()
+  virtual bool IsSnapshotSupported() const;
+
  // 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(int level);
+  int FindMinimumEmptyLevelFitting(ColumnFamilyData* cfd, int level);

  // Move the files in the input level to the target level.
  // If target_level < 0, automatically calculate the minimum level that could
  // hold the data set.
-  Status ReFitLevel(int level, int target_level = -1);
-
-  // Returns the current SuperVersion number.
-  uint64_t CurrentVersionNumber() const;
-
-  // Returns a pair of iterators (mutable-only and immutable-only) used
-  // internally by TailingIterator and stores CurrentVersionNumber() in
-  // *superversion_number. These iterators are always up-to-date, i.e. can
-  // be used to read new data.
-  std::pair<Iterator*, Iterator*> GetTailingIteratorPair(
-    const ReadOptions& options,
-    uint64_t* superversion_number);
-
-  // Constant after construction
-  const InternalFilterPolicy internal_filter_policy_;
-  bool owns_info_log_;
+  Status ReFitLevel(ColumnFamilyData* cfd, int level, int target_level = -1);

  // table_cache_ provides its own synchronization
-  unique_ptr<TableCache> table_cache_;
+  std::shared_ptr<Cache> table_cache_;

  // Lock over the persistent DB state.  Non-nullptr iff successfully acquired.
  FileLock* db_lock_;
@@ -386,18 +450,24 @@ class DBImpl : public DB {
  port::Mutex mutex_;
  port::AtomicPointer shutting_down_;
  port::CondVar bg_cv_;          // Signalled when background work finishes
-  MemTableRepFactory* mem_rep_factory_;
-  MemTable* mem_;
-  MemTableList imm_;             // Memtable that are not changing
  uint64_t logfile_number_;
  unique_ptr<log::Writer> log_;
-
-  SuperVersion* super_version_;
-
-  // An ordinal representing the current SuperVersion. Updated by
-  // InstallSuperVersion(), i.e. incremented every time super_version_
-  // changes.
-  std::atomic<uint64_t> super_version_number_;
+  bool log_empty_;
+  ColumnFamilyHandleImpl* default_cf_handle_;
+  unique_ptr<ColumnFamilyMemTablesImpl> column_family_memtables_;
+  struct LogFileNumberSize {
+    explicit LogFileNumberSize(uint64_t _number)
+        : number(_number), size(0), getting_flushed(false) {}
+    void AddSize(uint64_t new_size) { size += new_size; }
+    uint64_t number;
+    uint64_t size;
+    bool getting_flushed;
+  };
+  std::deque<LogFileNumberSize> alive_log_files_;
+  uint64_t total_log_size_;
+  // only used for dynamically adjusting max_total_wal_size. it is a sum of
+  // [write_buffer_size * max_write_buffer_number] over all column families
+  uint64_t max_total_in_memory_state_;

  std::string host_name_;

@@ -409,10 +479,18 @@ class DBImpl : public DB {

  SnapshotList snapshots_;

+  // cache for ReadFirstRecord() calls
+  std::unordered_map<uint64_t, SequenceNumber> read_first_record_cache_;
+  port::Mutex read_first_record_cache_mutex_;
+
  // Set of table files to protect from deletion because they are
  // part of ongoing compactions.
  std::set<uint64_t> pending_outputs_;

+  // At least one compaction or flush job is pending but not yet scheduled
+  // because of the max background thread limit.
+  bool bg_schedule_needed_;
+
  // count how many background compactions are running or have been scheduled
  int bg_compaction_scheduled_;

@@ -429,6 +507,7 @@ class DBImpl : public DB {

  // Information for a manual compaction
  struct ManualCompaction {
+    ColumnFamilyData* cfd;
    int input_level;
    int output_level;
    bool done;
@@ -468,89 +547,8 @@ class DBImpl : public DB {
  // enabled and archive size_limit is disabled.
  uint64_t default_interval_to_delete_obsolete_WAL_;

-  // These count the number of microseconds for which MakeRoomForWrite stalls.
-  uint64_t stall_level0_slowdown_;
-  uint64_t stall_memtable_compaction_;
-  uint64_t stall_level0_num_files_;
-  std::vector<uint64_t> stall_leveln_slowdown_;
-  uint64_t stall_level0_slowdown_count_;
-  uint64_t stall_memtable_compaction_count_;
-  uint64_t stall_level0_num_files_count_;
-  std::vector<uint64_t> stall_leveln_slowdown_count_;
-
-  // Time at which this instance was started.
-  const uint64_t started_at_;
-
  bool flush_on_destroy_; // Used when disableWAL is true.

-  // Per level compaction stats.  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;
-
-    // Bytes read from level N+1 during compaction between levels N and N+1
-    int64_t bytes_readnp1;
-
-    // Total bytes written during compaction between levels N and N+1
-    int64_t bytes_written;
-
-    // Files read from level N during compaction between levels N and N+1
-    int     files_in_leveln;
-
-    // Files read from level N+1 during compaction between levels N and N+1
-    int     files_in_levelnp1;
-
-    // Files written during compaction between levels N and N+1
-    int     files_out_levelnp1;
-
-    // Number of compactions done
-    int     count;
-
-    CompactionStats() : micros(0), bytes_readn(0), bytes_readnp1(0),
-                        bytes_written(0), files_in_leveln(0),
-                        files_in_levelnp1(0), files_out_levelnp1(0),
-                        count(0) { }
-
-    void Add(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->count += 1;
-    }
-  };
-
-  std::vector<CompactionStats> stats_;
-
-  // 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
-    // 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_;
-
-    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_;
-
  static const int KEEP_LOG_FILE_NUM = 1000;
  std::string db_absolute_path_;

@@ -566,6 +564,9 @@ class DBImpl : public DB {
  // Guard against multiple concurrent refitting
  bool refitting_level_;

+  // Indicate DB was opened successfully
+  bool opened_successfully_;
+
  // No copying allowed
  DBImpl(const DBImpl&);
  void operator=(const DBImpl&);
@@ -580,23 +581,23 @@ class DBImpl : public DB {
    std::vector<SequenceNumber>& snapshots,
    SequenceNumber* prev_snapshot);

-  // will return a pointer to SuperVersion* if previous SuperVersion
-  // if its reference count is zero and needs deletion or nullptr if not
-  // As argument takes a pointer to allocated SuperVersion
-  // Foreground threads call this function directly (they don't carry
-  // deletion state and have to handle their own creation and deletion
-  // of SuperVersion)
-  SuperVersion* InstallSuperVersion(SuperVersion* new_superversion);
  // Background threads call this function, which is just a wrapper around
-  // the InstallSuperVersion() function above. Background threads carry
+  // the cfd->InstallSuperVersion() function. Background threads carry
  // deletion_state which can have new_superversion already allocated.
-  void InstallSuperVersion(DeletionState& deletion_state);
+  void InstallSuperVersion(ColumnFamilyData* cfd,
+                           DeletionState& deletion_state);
+
+#ifndef ROCKSDB_LITE
+  using DB::GetPropertiesOfAllTables;
+  virtual Status GetPropertiesOfAllTables(ColumnFamilyHandle* column_family,
+                                          TablePropertiesCollection* props)
+      override;
+#endif  // ROCKSDB_LITE

  // Function that Get and KeyMayExist call with no_io true or false
  // Note: 'value_found' from KeyMayExist propagates here
-  Status GetImpl(const ReadOptions& options,
-                 const Slice& key,
-                 std::string* value,
+  Status GetImpl(const ReadOptions& options, ColumnFamilyHandle* column_family,
+                 const Slice& key, std::string* value,
                 bool* value_found = nullptr);
 };

@@ -606,7 +607,7 @@ 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);

 // Determine compression type, based on user options, level of the output
 // file and whether compression is disabled.
@@ -0,0 +1,132 @@
+//  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.
+
+#ifndef ROCKSDB_LITE
+
+#include "db/db_impl.h"
+
+namespace rocksdb {
+
+void DBImpl::TEST_PurgeObsoleteteWAL() { PurgeObsoleteWALFiles(); }
+
+uint64_t DBImpl::TEST_GetLevel0TotalSize() {
+  MutexLock l(&mutex_);
+  return default_cf_handle_->cfd()->current()->NumLevelBytes(0);
+}
+
+Iterator* DBImpl::TEST_NewInternalIterator(ColumnFamilyHandle* column_family) {
+  ColumnFamilyData* cfd;
+  if (column_family == nullptr) {
+    cfd = default_cf_handle_->cfd();
+  } else {
+    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+    cfd = cfh->cfd();
+  }
+
+  mutex_.Lock();
+  SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
+  mutex_.Unlock();
+  ReadOptions roptions;
+  return NewInternalIterator(roptions, cfd, super_version);
+}
+
+int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
+    ColumnFamilyHandle* column_family) {
+  ColumnFamilyData* cfd;
+  if (column_family == nullptr) {
+    cfd = default_cf_handle_->cfd();
+  } else {
+    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+    cfd = cfh->cfd();
+  }
+  MutexLock l(&mutex_);
+  return cfd->current()->MaxNextLevelOverlappingBytes();
+}
+
+void DBImpl::TEST_GetFilesMetaData(
+    ColumnFamilyHandle* column_family,
+    std::vector<std::vector<FileMetaData>>* metadata) {
+  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+  auto cfd = cfh->cfd();
+  MutexLock l(&mutex_);
+  metadata->resize(NumberLevels());
+  for (int level = 0; level < NumberLevels(); level++) {
+    const std::vector<FileMetaData*>& files = cfd->current()->files_[level];
+
+    (*metadata)[level].clear();
+    for (const auto& f : files) {
+      (*metadata)[level].push_back(*f);
+    }
+  }
+}
+
+uint64_t DBImpl::TEST_Current_Manifest_FileNo() {
+  return versions_->ManifestFileNumber();
+}
+
+Status DBImpl::TEST_CompactRange(int level, const Slice* begin,
+                                 const Slice* end,
+                                 ColumnFamilyHandle* column_family) {
+  ColumnFamilyData* cfd;
+  if (column_family == nullptr) {
+    cfd = default_cf_handle_->cfd();
+  } else {
+    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+    cfd = cfh->cfd();
+  }
+  int output_level =
+      (cfd->options()->compaction_style == kCompactionStyleUniversal)
+          ? level
+          : level + 1;
+  return RunManualCompaction(cfd, level, output_level, begin, end);
+}
+
+Status DBImpl::TEST_FlushMemTable(bool wait) {
+  FlushOptions fo;
+  fo.wait = wait;
+  return FlushMemTable(default_cf_handle_->cfd(), fo);
+}
+
+Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
+  ColumnFamilyData* cfd;
+  if (column_family == nullptr) {
+    cfd = default_cf_handle_->cfd();
+  } else {
+    auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+    cfd = cfh->cfd();
+  }
+  return WaitForFlushMemTable(cfd);
+}
+
+Status DBImpl::TEST_WaitForCompact() {
+  // Wait until the compaction completes
+
+  // TODO: a bug here. This function actually does not necessarily
+  // wait for compact. It actually waits for scheduled compaction
+  // OR flush to finish.
+
+  MutexLock l(&mutex_);
+  while ((bg_compaction_scheduled_ || bg_flush_scheduled_) && bg_error_.ok()) {
+    bg_cv_.Wait();
+  }
+  return bg_error_;
+}
+
+Status DBImpl::TEST_ReadFirstRecord(const WalFileType type,
+                                    const uint64_t number,
+                                    SequenceNumber* sequence) {
+  return ReadFirstRecord(type, number, sequence);
+}
+
+Status DBImpl::TEST_ReadFirstLine(const std::string& fname,
+                                  SequenceNumber* sequence) {
+  return ReadFirstLine(fname, sequence);
+}
+}  // namespace rocksdb
+#endif  // ROCKSDB_LITE
@@ -42,8 +42,8 @@

 namespace rocksdb {

-DBImplReadOnly::DBImplReadOnly(const Options& options,
-    const std::string& dbname)
+DBImplReadOnly::DBImplReadOnly(const DBOptions& options,
+                               const std::string& dbname)
    : DBImpl(options, dbname) {
  Log(options_.info_log, "Opening the db in read only mode");
 }
@@ -53,47 +53,102 @@ DBImplReadOnly::~DBImplReadOnly() {

 // Implementations of the DB interface
 Status DBImplReadOnly::Get(const ReadOptions& options,
-                   const Slice& key,
-                   std::string* value) {
+                           ColumnFamilyHandle* column_family, const Slice& key,
+                           std::string* value) {
  Status s;
-  MemTable* mem = GetMemTable();
-  Version* current = versions_->current();
  SequenceNumber snapshot = versions_->LastSequence();
+  auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
+  auto cfd = cfh->cfd();
+  SuperVersion* super_version = cfd->GetSuperVersion();
  MergeContext merge_context;
  LookupKey lkey(key, snapshot);
-  if (mem->Get(lkey, value, &s, merge_context, options_)) {
+  if (super_version->mem->Get(lkey, value, &s, merge_context,
+                              *cfd->options())) {
  } else {
    Version::GetStats stats;
-    current->Get(options, lkey, value, &s, &merge_context, &stats, options_);
+    super_version->current->Get(options, lkey, value, &s, &merge_context,
+                                &stats);
  }
  return s;
 }

-Iterator* DBImplReadOnly::NewIterator(const ReadOptions& options) {
-  SequenceNumber latest_snapshot;
-  Iterator* internal_iter = NewInternalIterator(options, &latest_snapshot);
+Iterator* DBImplReadOnly::NewIterator(const ReadOptions& 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(
-    &dbname_, env_, options_,  user_comparator(),internal_iter,
+      env_, *cfd->options(), cfd->user_comparator(), internal_iter,
      (options.snapshot != nullptr
-      ? reinterpret_cast<const SnapshotImpl*>(options.snapshot)->number_
-      : latest_snapshot));
+           ? reinterpret_cast<const SnapshotImpl*>(options.snapshot)->number_
+           : latest_snapshot));
 }

-
 Status DB::OpenForReadOnly(const Options& options, const std::string& dbname,
-                DB** dbptr, bool error_if_log_file_exist) {
+                           DB** dbptr, bool error_if_log_file_exist) {
  *dbptr = nullptr;

-  DBImplReadOnly* impl = new DBImplReadOnly(options, dbname);
+  DBOptions db_options(options);
+  ColumnFamilyOptions cf_options(options);
+  std::vector<ColumnFamilyDescriptor> column_families;
+  column_families.push_back(
+      ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
+  std::vector<ColumnFamilyHandle*> handles;
+
+  Status 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
+    // reference to default column family
+    delete handles[0];
+  }
+  return s;
+}
+
+Status DB::OpenForReadOnly(
+    const DBOptions& db_options, const std::string& dbname,
+    const std::vector<ColumnFamilyDescriptor>& column_families,
+    std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
+    bool error_if_log_file_exist) {
+  *dbptr = nullptr;
+  handles->clear();
+
+  DBImplReadOnly* impl = new DBImplReadOnly(db_options, dbname);
  impl->mutex_.Lock();
-  Status s = impl->Recover(true /* read only */, error_if_log_file_exist);
+  Status s = impl->Recover(column_families, true /* read only */,
+                           error_if_log_file_exist);
+  if (s.ok()) {
+    // set column family handles
+    for (auto cf : column_families) {
+      auto cfd =
+          impl->versions_->GetColumnFamilySet()->GetColumnFamily(cf.name);
+      if (cfd == nullptr) {
+        s = Status::InvalidArgument("Column family not found: ", cf.name);
+        break;
+      }
+      handles->push_back(new ColumnFamilyHandleImpl(cfd, impl, &impl->mutex_));
+    }
+  }
+  if (s.ok()) {
+    for (auto cfd : *impl->versions_->GetColumnFamilySet()) {
+      delete cfd->InstallSuperVersion(new SuperVersion(), &impl->mutex_);
+    }
+  }
  impl->mutex_.Unlock();
  if (s.ok()) {
    *dbptr = impl;
  } else {
+    for (auto h : *handles) {
+      delete h;
+    }
+    handles->clear();
    delete impl;
  }
  return s;
 }

-}
+
+}   // namespace rocksdb
@@ -12,6 +12,8 @@

 #include <deque>
 #include <set>
+#include <vector>
+#include <string>
 #include "db/dbformat.h"
 #include "db/log_writer.h"
 #include "db/snapshot.h"
@@ -23,57 +25,79 @@
 namespace rocksdb {

 class DBImplReadOnly : public DBImpl {
-public:
-  DBImplReadOnly(const Options& options, const std::string& dbname);
- virtual ~DBImplReadOnly();
+ public:
+  DBImplReadOnly(const DBOptions& options, const std::string& dbname);
+  virtual ~DBImplReadOnly();

- // Implementations of the DB interface
- virtual Status Get(const ReadOptions& options,
-                    const Slice& key,
-                    std::string* value);
+  // Implementations of the DB interface
+  using DB::Get;
+  virtual Status Get(const ReadOptions& options,
+                     ColumnFamilyHandle* column_family, const Slice& key,
+                     std::string* value);

- // TODO: Implement ReadOnly MultiGet?
+  // TODO: Implement ReadOnly MultiGet?

- virtual Iterator* NewIterator(const ReadOptions&);
+  using DBImpl::NewIterator;
+  virtual Iterator* NewIterator(const ReadOptions&,
+                                ColumnFamilyHandle* column_family);

- virtual Status Put(const WriteOptions&, const Slice& key, const Slice& value) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status Merge(const WriteOptions&, const Slice& key,
-                      const Slice& value) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status Delete(const WriteOptions&, const Slice& key) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status Write(const WriteOptions& options, WriteBatch* updates) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status CompactRange(const Slice* begin, const Slice* end,
-                             bool reduce_level = false, int target_level = -1) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status DisableFileDeletions() {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status EnableFileDeletions(bool force) {
-   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) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
- virtual Status Flush(const FlushOptions& options) {
-   return Status::NotSupported("Not supported operation in read only mode.");
- }
+  virtual Status NewIterators(
+      const ReadOptions& options,
+      const std::vector<ColumnFamilyHandle*>& column_family,
+      std::vector<Iterator*>* iterators) {
+   // TODO
+    return Status::NotSupported("Not supported yet.");
+  }

-private:
- friend class DB;
+  using DBImpl::Put;
+  virtual Status Put(const WriteOptions& options,
+                     ColumnFamilyHandle* column_family, const Slice& key,
+                     const Slice& value) {
+    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) {
+    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) {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }
+  virtual Status Write(const WriteOptions& options, WriteBatch* updates) {
+    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) {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }
+  virtual Status DisableFileDeletions() {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }
+  virtual Status EnableFileDeletions(bool force) {
+    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) {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }
+  using DBImpl::Flush;
+  virtual Status Flush(const FlushOptions& options,
+                       ColumnFamilyHandle* column_family) {
+    return Status::NotSupported("Not supported operation in read only mode.");
+  }

- // No copying allowed
- DBImplReadOnly(const DBImplReadOnly&);
- void operator=(const DBImplReadOnly&);
+ private:
+  friend class DB;
+
+  // No copying allowed
+  DBImplReadOnly(const DBImplReadOnly&);
+  void operator=(const DBImplReadOnly&);
 };
-
 }
@@ -57,10 +57,9 @@ class DBIter: public Iterator {
    kReverse
  };

-  DBIter(const std::string* dbname, Env* env, const Options& options,
+  DBIter(Env* env, const Options& options,
         const Comparator* cmp, Iterator* iter, SequenceNumber s)
-      : dbname_(dbname),
-        env_(env),
+      : env_(env),
        logger_(options.info_log.get()),
        user_comparator_(cmp),
        user_merge_operator_(options.merge_operator.get()),
@@ -80,7 +79,7 @@ class DBIter: public Iterator {
  virtual bool Valid() const { return valid_; }
  virtual Slice key() const {
    assert(valid_);
-    return saved_key_;
+    return saved_key_.GetKey();
  }
  virtual Slice value() const {
    assert(valid_);
@@ -102,15 +101,12 @@ class DBIter: public Iterator {
  virtual void SeekToLast();

 private:
-  void FindNextUserEntry(bool skipping);
+  inline void FindNextUserEntry(bool skipping);
+  void FindNextUserEntryInternal(bool skipping);
  void FindPrevUserEntry();
  bool ParseKey(ParsedInternalKey* key);
  void MergeValuesNewToOld();

-  inline void SaveKey(const Slice& k, std::string* dst) {
-    dst->assign(k.data(), k.size());
-  }
-
  inline void ClearSavedValue() {
    if (saved_value_.capacity() > 1048576) {
      std::string empty;
@@ -120,7 +116,6 @@ class DBIter: public Iterator {
    }
  }

-  const std::string* const dbname_;
  Env* const env_;
  Logger* logger_;
  const Comparator* const user_comparator_;
@@ -129,7 +124,7 @@ class DBIter: public Iterator {
  SequenceNumber const sequence_;

  Status status_;
-  std::string saved_key_;     // == current key when direction_==kReverse
+  IterKey saved_key_;   // == current key when direction_==kReverse
  std::string saved_value_;   // == current raw value when direction_==kReverse
  std::string skip_key_;
  Direction direction_;
@@ -169,7 +164,7 @@ void DBIter::Next() {
    }
    if (!iter_->Valid()) {
      valid_ = false;
-      saved_key_.clear();
+      saved_key_.Clear();
      return;
    }
  }
@@ -191,7 +186,14 @@ void DBIter::Next() {
 //
 // NOTE: In between, saved_key_ can point to a user key that has
 //       a delete marker
-void DBIter::FindNextUserEntry(bool skipping) {
+inline void DBIter::FindNextUserEntry(bool skipping) {
+  PERF_TIMER_AUTO(find_next_user_entry_time);
+  FindNextUserEntryInternal(skipping);
+  PERF_TIMER_STOP(find_next_user_entry_time);
+}
+
+// Actual implementation of DBIter::FindNextUserEntry()
+void DBIter::FindNextUserEntryInternal(bool skipping) {
  // Loop until we hit an acceptable entry to yield
  assert(iter_->Valid());
  assert(direction_ == kForward);
@@ -201,32 +203,32 @@ void DBIter::FindNextUserEntry(bool skipping) {
    ParsedInternalKey ikey;
    if (ParseKey(&ikey) && ikey.sequence <= sequence_) {
      if (skipping &&
-          user_comparator_->Compare(ikey.user_key, saved_key_) <= 0) {
+          user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) <= 0) {
        num_skipped++; // skip this entry
-        BumpPerfCount(&perf_context.internal_key_skipped_count);
+        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.
-            SaveKey(ikey.user_key, &saved_key_);
+            saved_key_.SetUserKey(ikey.user_key);
            skipping = true;
            num_skipped = 0;
-            BumpPerfCount(&perf_context.internal_delete_skipped_count);
+            PERF_COUNTER_ADD(internal_delete_skipped_count, 1);
            break;
          case kTypeValue:
            valid_ = true;
-            SaveKey(ikey.user_key, &saved_key_);
+            saved_key_.SetUserKey(ikey.user_key);
            return;
          case kTypeMerge:
            // By now, we are sure the current ikey is going to yield a value
-            SaveKey(ikey.user_key, &saved_key_);
+            saved_key_.SetUserKey(ikey.user_key);
            current_entry_is_merged_ = true;
            valid_ = true;
            MergeValuesNewToOld();  // Go to a different state machine
            return;
-          case kTypeLogData:
+          default:
            assert(false);
            break;
        }
@@ -239,8 +241,8 @@ void DBIter::FindNextUserEntry(bool skipping) {
    if (skipping && num_skipped > max_skip_) {
      num_skipped = 0;
      std::string last_key;
-      AppendInternalKey(&last_key,
-        ParsedInternalKey(Slice(saved_key_), 0, kValueTypeForSeek));
+      AppendInternalKey(&last_key, ParsedInternalKey(saved_key_.GetKey(), 0,
+                                                     kValueTypeForSeek));
      iter_->Seek(last_key);
      RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
    } else {
@@ -275,7 +277,7 @@ void DBIter::MergeValuesNewToOld() {
      continue;
    }

-    if (user_comparator_->Compare(ikey.user_key, saved_key_) != 0) {
+    if (user_comparator_->Compare(ikey.user_key, saved_key_.GetKey()) != 0) {
      // hit the next user key, stop right here
      break;
    }
@@ -304,21 +306,6 @@ void DBIter::MergeValuesNewToOld() {
      // when complete, add result to operands and continue.
      const Slice& value = iter_->value();
      operands.push_front(value.ToString());
-      while(operands.size() >= 2) {
-        // Call user associative-merge until it returns false
-        if (user_merge_operator_->PartialMerge(ikey.user_key,
-                                               Slice(operands[0]),
-                                               Slice(operands[1]),
-                                               &merge_result,
-                                               logger_)) {
-          operands.pop_front();
-          swap(operands.front(), merge_result);
-        } else {
-          // Associative merge returns false ==> stack the operands
-          break;
-        }
-      }
-
    }
  }

@@ -326,7 +313,7 @@ void DBIter::MergeValuesNewToOld() {
  // a deletion marker.
  // feed null as the existing value to the merge operator, such that
  // client can differentiate this scenario and do things accordingly.
-  user_merge_operator_->FullMerge(saved_key_, nullptr, operands,
+  user_merge_operator_->FullMerge(saved_key_.GetKey(), nullptr, operands,
                                  &saved_value_, logger_);
 }

@@ -345,17 +332,17 @@ void DBIter::Prev() {
    // 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
-    SaveKey(ExtractUserKey(iter_->key()), &saved_key_);
+    saved_key_.SetUserKey(ExtractUserKey(iter_->key()));
    while (true) {
      iter_->Prev();
      if (!iter_->Valid()) {
        valid_ = false;
-        saved_key_.clear();
+        saved_key_.Clear();
        ClearSavedValue();
        return;
      }
      if (user_comparator_->Compare(ExtractUserKey(iter_->key()),
-                                    saved_key_) < 0) {
+                                    saved_key_.GetKey()) < 0) {
        break;
      }
    }
@@ -370,42 +357,46 @@ void DBIter::FindPrevUserEntry() {
  uint64_t num_skipped = 0;

  ValueType value_type = kTypeDeletion;
+  bool saved_key_valid = true;
  if (iter_->Valid()) {
    do {
      ParsedInternalKey ikey;
-      bool saved_key_cleared = false;
      if (ParseKey(&ikey) && ikey.sequence <= sequence_) {
        if ((value_type != kTypeDeletion) &&
-            user_comparator_->Compare(ikey.user_key, saved_key_) < 0) {
+            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();
+          saved_key_.Clear();
          ClearSavedValue();
-          saved_key_cleared = true;
+          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_);
          }
-          SaveKey(ExtractUserKey(iter_->key()), &saved_key_);
+          saved_key_.SetUserKey(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_cleared && num_skipped > max_skip_) {
+      if (saved_key_valid && num_skipped > max_skip_) {
        num_skipped = 0;
        std::string last_key;
-        AppendInternalKey(&last_key,
-          ParsedInternalKey(Slice(saved_key_), kMaxSequenceNumber,
-                            kValueTypeForSeek));
+        AppendInternalKey(&last_key, ParsedInternalKey(saved_key_.GetKey(),
+                                                       kMaxSequenceNumber,
+                                                       kValueTypeForSeek));
        iter_->Seek(last_key);
        RecordTick(statistics_, NUMBER_OF_RESEEKS_IN_ITERATION);
      } else {
@@ -417,7 +408,7 @@ void DBIter::FindPrevUserEntry() {
  if (value_type == kTypeDeletion) {
    // End
    valid_ = false;
-    saved_key_.clear();
+    saved_key_.Clear();
    ClearSavedValue();
    direction_ = kForward;
  } else {
@@ -426,13 +417,15 @@ void DBIter::FindPrevUserEntry() {
 }

 void DBIter::Seek(const Slice& target) {
-  direction_ = kForward;
-  ClearSavedValue();
-  saved_key_.clear();
-  AppendInternalKey(
-      &saved_key_, ParsedInternalKey(target, sequence_, kValueTypeForSeek));
-  iter_->Seek(saved_key_);
+  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);
  if (iter_->Valid()) {
+    direction_ = kForward;
+    ClearSavedValue();
    FindNextUserEntry(false /*not skipping */);
  } else {
    valid_ = false;
@@ -442,7 +435,9 @@ void DBIter::Seek(const Slice& target) {
 void DBIter::SeekToFirst() {
  direction_ = kForward;
  ClearSavedValue();
+  PERF_TIMER_AUTO(seek_internal_seek_time);
  iter_->SeekToFirst();
+  PERF_TIMER_STOP(seek_internal_seek_time);
  if (iter_->Valid()) {
    FindNextUserEntry(false /* not skipping */);
  } else {
@@ -461,20 +456,21 @@ void DBIter::SeekToLast() {

  direction_ = kReverse;
  ClearSavedValue();
+  PERF_TIMER_AUTO(seek_internal_seek_time);
  iter_->SeekToLast();
+  PERF_TIMER_STOP(seek_internal_seek_time);
  FindPrevUserEntry();
 }

 }  // anonymous namespace

 Iterator* NewDBIterator(
-    const std::string* dbname,
    Env* env,
    const Options& options,
    const Comparator *user_key_comparator,
    Iterator* internal_iter,
    const SequenceNumber& sequence) {
-  return new DBIter(dbname, env, options, user_key_comparator,
+  return new DBIter(env, options, user_key_comparator,
                    internal_iter, sequence);
 }

@@ -18,7 +18,6 @@ namespace rocksdb {
 // "*internal_iter") that were live at the specified "sequence" number
 // into appropriate user keys.
 extern Iterator* NewDBIterator(
-    const std::string* dbname,
    Env* env,
    const Options& options,
    const Comparator *user_key_comparator,
@@ -20,7 +20,8 @@
 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()) {
@@ -65,7 +66,7 @@ void DBImpl::LogDBDeployStats() {

  uint64_t file_total_size = 0;
  uint32_t file_total_num = 0;
-  Version* current = versions_->current();
+  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);
@@ -89,6 +90,6 @@ void DBImpl::LogDBDeployStats() {
  bg_logstats_scheduled_ = false;
  bg_cv_.SignalAll();
  mutex_.Unlock();
+#endif
 }
-
 }
@@ -6,16 +6,16 @@
 // 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/dbformat.h"

 #include <stdio.h>
-#include "db/dbformat.h"
 #include "port/port.h"
 #include "util/coding.h"
 #include "util/perf_context_imp.h"

 namespace rocksdb {

-static uint64_t PackSequenceAndType(uint64_t seq, ValueType t) {
+uint64_t PackSequenceAndType(uint64_t seq, ValueType t) {
  assert(seq <= kMaxSequenceNumber);
  assert(t <= kValueTypeForSeek);
  return (seq << 8) | t;
@@ -59,7 +59,7 @@ int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {
  //    decreasing sequence number
  //    decreasing type (though sequence# should be enough to disambiguate)
  int r = user_comparator_->Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
-  BumpPerfCount(&perf_context.user_key_comparison_count);
+  PERF_COUNTER_ADD(user_key_comparison_count, 1);
  if (r == 0) {
    const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8);
    const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8);
@@ -72,6 +72,28 @@ int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {
  return r;
 }

+int InternalKeyComparator::Compare(const ParsedInternalKey& a,
+                                   const ParsedInternalKey& b) const {
+  // Order by:
+  //    increasing user key (according to user-supplied comparator)
+  //    decreasing sequence number
+  //    decreasing type (though sequence# should be enough to disambiguate)
+  int r = user_comparator_->Compare(a.user_key, b.user_key);
+  PERF_COUNTER_ADD(user_key_comparison_count, 1);
+  if (r == 0) {
+    if (a.sequence > b.sequence) {
+      r = -1;
+    } else if (a.sequence < b.sequence) {
+      r = +1;
+    } else if (a.type > b.type) {
+      r = -1;
+    } else if (a.type < b.type) {
+      r = +1;
+    }
+  }
+  return r;
+}
+
 void InternalKeyComparator::FindShortestSeparator(
      std::string* start,
      const Slice& limit) const {
@@ -13,6 +13,7 @@
 #include "rocksdb/db.h"
 #include "rocksdb/filter_policy.h"
 #include "rocksdb/slice.h"
+#include "rocksdb/slice_transform.h"
 #include "rocksdb/table.h"
 #include "rocksdb/types.h"
 #include "util/coding.h"
@@ -25,12 +26,21 @@ class InternalKey;
 // Value types encoded as the last component of internal keys.
 // DO NOT CHANGE THESE ENUM VALUES: they are embedded in the on-disk
 // data structures.
-enum ValueType {
+// The highest bit of the value type needs to be reserved to SST tables
+// for them to do more flexible encoding.
+enum ValueType : unsigned char {
  kTypeDeletion = 0x0,
  kTypeValue = 0x1,
  kTypeMerge = 0x2,
-  kTypeLogData = 0x3
+  // Following types are used only in write ahead logs. They are not used in
+  // memtables or sst files:
+  kTypeLogData = 0x3,
+  kTypeColumnFamilyDeletion = 0x4,
+  kTypeColumnFamilyValue = 0x5,
+  kTypeColumnFamilyMerge = 0x6,
+  kMaxValue = 0x7F
 };
+
 // kValueTypeForSeek defines the ValueType that should be passed when
 // constructing a ParsedInternalKey object for seeking to a particular
 // sequence number (since we sort sequence numbers in decreasing order
@@ -60,6 +70,8 @@ inline size_t InternalKeyEncodingLength(const ParsedInternalKey& key) {
  return key.user_key.size() + 8;
 }

+extern uint64_t PackSequenceAndType(uint64_t seq, ValueType t);
+
 // Append the serialization of "key" to *result.
 extern void AppendInternalKey(std::string* result,
                              const ParsedInternalKey& key);
@@ -96,6 +108,7 @@ class InternalKeyComparator : public Comparator {
    name_("rocksdb.InternalKeyComparator:" +
          std::string(user_comparator_->Name())) {
  }
+  virtual ~InternalKeyComparator() {}

  virtual const char* Name() const;
  virtual int Compare(const Slice& a, const Slice& b) const;
@@ -107,6 +120,7 @@ class InternalKeyComparator : public Comparator {
  const Comparator* user_comparator() const { return user_comparator_; }

  int Compare(const InternalKey& a, const InternalKey& b) const;
+  int Compare(const ParsedInternalKey& a, const ParsedInternalKey& b) const;
 };

 // Filter policy wrapper that converts from internal keys to user keys
@@ -163,6 +177,7 @@ inline bool ParseInternalKey(const Slice& internal_key,
  unsigned char c = num & 0xff;
  result->sequence = num >> 8;
  result->type = static_cast<ValueType>(c);
+  assert(result->type <= ValueType::kMaxValue);
  result->user_key = Slice(internal_key.data(), n - 8);
  return (c <= static_cast<unsigned char>(kValueTypeForSeek));
 }
@@ -226,4 +241,100 @@ inline LookupKey::~LookupKey() {
  if (start_ != space_) delete[] start_;
 }

+class IterKey {
+ public:
+  IterKey() : key_(space_), buf_size_(sizeof(space_)), key_size_(0) {}
+
+  ~IterKey() { ResetBuffer(); }
+
+  Slice GetKey() const { return Slice(key_, key_size_); }
+
+  void Clear() { key_size_ = 0; }
+
+  void SetUserKey(const Slice& user_key) {
+    size_t size = user_key.size();
+    EnlargeBufferIfNeeded(size);
+    memcpy(key_, user_key.data(), size);
+    key_size_ = size;
+  }
+
+  void SetInternalKey(const Slice& user_key, SequenceNumber s,
+                      ValueType value_type = kValueTypeForSeek) {
+    size_t usize = user_key.size();
+    EnlargeBufferIfNeeded(usize + sizeof(uint64_t));
+    memcpy(key_, user_key.data(), usize);
+    EncodeFixed64(key_ + usize, PackSequenceAndType(s, value_type));
+    key_size_ = usize + sizeof(uint64_t);
+  }
+
+  void SetInternalKey(const ParsedInternalKey& parsed_key) {
+    SetInternalKey(parsed_key.user_key, parsed_key.sequence, parsed_key.type);
+  }
+
+ private:
+  char* key_;
+  size_t buf_size_;
+  size_t key_size_;
+  char space_[32];  // Avoid allocation for short keys
+
+  void ResetBuffer() {
+    if (key_ != nullptr && key_ != space_) {
+      delete[] key_;
+    }
+    key_ = space_;
+    buf_size_ = sizeof(space_);
+    key_size_ = 0;
+  }
+
+  // Enlarge the buffer size if needed based on key_size.
+  // By default, static allocated buffer is used. Once there is a key
+  // larger than the static allocated buffer, another buffer is dynamically
+  // allocated, until a larger key buffer is requested. In that case, we
+  // reallocate buffer and delete the old one.
+  void EnlargeBufferIfNeeded(size_t key_size) {
+    // If size is smaller than buffer size, continue using current buffer,
+    // or the static allocated one, as default
+    if (key_size > buf_size_) {
+      // Need to enlarge the buffer.
+      ResetBuffer();
+      key_ = new char[key_size];
+      buf_size_ = key_size;
+    }
+  }
+
+  // No copying allowed
+  IterKey(const IterKey&) = delete;
+  void operator=(const IterKey&) = delete;
+};
+
+class InternalKeySliceTransform : public SliceTransform {
+ public:
+  explicit InternalKeySliceTransform(const SliceTransform* transform)
+      : transform_(transform) {}
+
+  virtual const char* Name() const { return transform_->Name(); }
+
+  virtual Slice Transform(const Slice& src) const {
+    auto user_key = ExtractUserKey(src);
+    return transform_->Transform(user_key);
+  }
+
+  virtual bool InDomain(const Slice& src) const {
+    auto user_key = ExtractUserKey(src);
+    return transform_->InDomain(user_key);
+  }
+
+  virtual bool InRange(const Slice& dst) const {
+    auto user_key = ExtractUserKey(dst);
+    return transform_->InRange(user_key);
+  }
+
+  const SliceTransform* user_prefix_extractor() const { return transform_; }
+
+ private:
+  // Like comparator, InternalKeySliceTransform will not take care of the
+  // deletion of transform_
+  const SliceTransform* const transform_;
+};
+
 }  // namespace rocksdb
@@ -0,0 +1,202 @@
+//  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/file_indexer.h"
+#include <algorithm>
+#include "rocksdb/comparator.h"
+#include "db/version_edit.h"
+
+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) {
+}
+
+
+uint32_t FileIndexer::NumLevelIndex() {
+  return next_level_index_.size();
+}
+
+uint32_t FileIndexer::LevelIndexSize(uint32_t level) {
+  return next_level_index_[level].size();
+}
+
+void FileIndexer::GetNextLevelIndex(
+    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) {
+  assert(level > 0);
+
+  // Last level, no hint
+  if (level == num_levels_ - 1) {
+    *left_bound = 0;
+    *right_bound = -1;
+    return;
+  }
+
+  assert(level < num_levels_ - 1);
+  assert(static_cast<int32_t>(file_index) <= level_rb_[level]);
+
+  const auto& index = next_level_index_[level][file_index];
+
+  if (cmp_smallest < 0) {
+    *left_bound = (level > 0 && file_index > 0) ?
+      next_level_index_[level][file_index - 1].largest_lb : 0;
+    *right_bound = index.smallest_rb;
+  } else if (cmp_smallest == 0) {
+    *left_bound = index.smallest_lb;
+    *right_bound = index.smallest_rb;
+  } else if (cmp_smallest > 0 && cmp_largest < 0) {
+    *left_bound = index.smallest_lb;
+    *right_bound = index.largest_rb;
+  } else if (cmp_largest == 0) {
+    *left_bound = index.largest_lb;
+    *right_bound = index.largest_rb;
+  } else if (cmp_largest > 0) {
+    *left_bound = index.largest_lb;
+    *right_bound = level_rb_[level + 1];
+  } else {
+    assert(false);
+  }
+
+  assert(*left_bound >= 0);
+  assert(*left_bound <= *right_bound + 1);
+  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) {
+  if (files == nullptr) {
+    return;
+  }
+
+  // L1 - Ln-1
+  for (uint32_t level = 1; level < num_levels_ - 1; ++level) {
+    const auto& upper_files = files[level];
+    const int32_t upper_size = upper_files.size();
+    const auto& lower_files = files[level + 1];
+    level_rb_[level] = upper_files.size() - 1;
+    if (upper_size == 0) {
+      continue;
+    }
+    auto& index = next_level_index_[level];
+    index.resize(upper_size);
+
+    CalculateLB(upper_files, lower_files, &index,
+        [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 {
+          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 {
+          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 {
+          return ucmp_->Compare(a->largest.user_key(), b->smallest.user_key());
+        },
+        [](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,
+    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;
+  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);
+      ++upper_idx;
+      ++lower_idx;
+    } else if (cmp > 0) {
+      // Lower level's file (largest) is smaller, a key won't hit in that
+      // file. Move to next lower file
+      ++lower_idx;
+    } else {
+      // Lower level's file becomes larger, update the index, and
+      // move to the next upper file
+      set_index(&(*index)[upper_idx], lower_idx);
+      ++upper_idx;
+    }
+  }
+
+  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);
+    ++upper_idx;
+  }
+}
+
+void FileIndexer::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) {
+  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;
+  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);
+      --upper_idx;
+      --lower_idx;
+    } else if (cmp < 0) {
+      // Lower level's file (smallest) is larger, a key won't hit in that
+      // file. Move to next lower file.
+      --lower_idx;
+    } 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);
+      --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);
+    --upper_idx;
+  }
+}
+
+}  // namespace rocksdb
@@ -0,0 +1,129 @@
+//  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.
+
+#pragma once
+#include <cstdint>
+#include <functional>
+#include <limits>
+#include <vector>
+
+namespace rocksdb {
+
+class Comparator;
+struct FileMetaData;
+
+// 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
+// and then check if a target key can be found in the file by comparing the key
+// to each file's smallest and largest key. The results of these comparisions
+// can be reused beyond checking if a key falls into a file's range.
+// With some pre-calculated knowledge, each key comparision that has been done
+// can serve as a hint to narrow down further searches: if a key compared to
+// be smaller than a file's smallest or largest, that comparison can be used
+// to find out the right bound of next binary search. Similarly, if a key
+// compared to be larger than a file's smallest or largest, it can be utilized
+// to find out the left bound of next binary search.
+// With these hints: it can greatly reduce the range of binary search,
+// especially for bottom levels, given that one file most likely overlaps with
+// only N files from level below (where N is max_bytes_for_level_multiplier).
+// So on level L, we will only look at ~N files instead of N^L files on the
+// naive approach.
+class FileIndexer {
+ public:
+  FileIndexer(const uint32_t num_levels, const Comparator* ucmp);
+
+  uint32_t NumLevelIndex();
+
+  uint32_t LevelIndexSize(uint32_t level);
+
+  // Return a file index range in the next level to search for a key based on
+  // smallest and largest key comparision for the current file specified by
+  // level and file_index. When *left_index < *right_index, both index should
+  // be valid and fit in the vector size.
+  void GetNextLevelIndex(
+    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);
+
+  enum {
+    kLevelMaxIndex = std::numeric_limits<int32_t>::max()
+  };
+
+ private:
+  const uint32_t num_levels_;
+  const Comparator* ucmp_;
+
+  struct IndexUnit {
+    IndexUnit()
+      : smallest_lb(0), largest_lb(0), smallest_rb(-1), largest_rb(-1) {}
+    // During file search, a key is compared against smallest and largest
+    // from a FileMetaData. It can have 3 possible outcomes:
+    // (1) key is smaller than smallest, implying it is also smaller than
+    //     larger. Precalculated index based on "smallest < smallest" can
+    //     be used to provide right bound.
+    // (2) key is in between smallest and largest.
+    //     Precalculated index based on "smallest > greatest" can be used to
+    //     provide left bound.
+    //     Precalculated index based on "largest < smallest" can be used to
+    //     provide right bound.
+    // (3) key is larger than largest, implying it is also larger than smallest.
+    //     Precalculated index based on "largest > largest" can be used to
+    //     provide left bound.
+    //
+    // As a result, we will need to do:
+    // Compare smallest (<=) and largest keys from upper level file with
+    // smallest key from lower level to get a right bound.
+    // Compare smallest (>=) and largest keys from upper level file with
+    // largest key from lower level to get a left bound.
+    //
+    // Example:
+    //    level 1:              [50 - 60]
+    //    level 2:        [1 - 40], [45 - 55], [58 - 80]
+    // A key 35, compared to be less than 50, 3rd file on level 2 can be
+    // skipped according to rule (1). LB = 0, RB = 1.
+    // A key 53, sits in the middle 50 and 60. 1st file on level 2 can be
+    // skipped according to rule (2)-a, but the 3rd file cannot be skipped
+    // because 60 is greater than 58. LB = 1, RB = 2.
+    // A key 70, compared to be larger than 60. 1st and 2nd file can be skipped
+    // according to rule (3). LB = 2, RB = 2.
+    //
+    // Point to a left most file in a lower level that may contain a key,
+    // which compares greater than smallest of a FileMetaData (upper level)
+    int32_t smallest_lb;
+    // Point to a left most file in a lower level that may contain a key,
+    // which compares greater than largest of a FileMetaData (upper level)
+    int32_t largest_lb;
+    // Point to a right most file in a lower level that may contain a key,
+    // which compares smaller than smallest of a FileMetaData (upper level)
+    int32_t smallest_rb;
+    // Point to a right most file in a lower level that may contain a key,
+    // which compares smaller than largest of a FileMetaData (upper level)
+    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);
+
+  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);
+
+  std::vector<std::vector<IndexUnit>> next_level_index_;
+  std::vector<int32_t> level_rb_;
+};
+
+}  // namespace rocksdb
@@ -0,0 +1,330 @@
+//  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 <string>
+#include "db/file_indexer.h"
+#include "db/dbformat.h"
+#include "db/version_edit.h"
+#include "rocksdb/comparator.h"
+#include "util/testharness.h"
+#include "util/testutil.h"
+
+namespace rocksdb {
+
+class IntComparator : public Comparator {
+ public:
+  int Compare(const Slice& a, const Slice& b) const {
+    assert(a.size() == 8);
+    assert(b.size() == 8);
+    return *reinterpret_cast<const int64_t*>(a.data()) -
+      *reinterpret_cast<const int64_t*>(b.data());
+  }
+
+  const char* Name() const {
+    return "IntComparator";
+  }
+
+  void FindShortestSeparator(std::string* start, const Slice& limit) const {}
+
+  void FindShortSuccessor(std::string* key) const {}
+};
+
+
+struct FileIndexerTest {
+ public:
+  FileIndexerTest() :
+    kNumLevels(4), indexer(kNumLevels, &ucmp),
+    files(new std::vector<FileMetaData*>[kNumLevels]) {
+  }
+
+  ~FileIndexerTest() {
+    Reset();
+    delete[] files;
+  }
+
+  void AddFile(int level, int64_t smallest, int64_t largest) {
+    auto* f = new FileMetaData();
+    f->smallest = IntKey(smallest);
+    f->largest = IntKey(largest);
+    files[level].push_back(f);
+  }
+
+  InternalKey IntKey(int64_t v) {
+    return InternalKey(Slice(reinterpret_cast<char*>(&v), 8), 0, kTypeValue);
+  }
+
+  void Reset() {
+    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,
+      const int cmp_smallest, const int cmp_largest, int32_t* left_index,
+      int32_t* right_index) {
+    *left_index = 100;
+    *right_index = 100;
+    indexer.GetNextLevelIndex(level, file_index, cmp_smallest, cmp_largest,
+        left_index, right_index);
+  }
+
+  const uint32_t kNumLevels;
+  IntComparator ucmp;
+  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 1: no overlap, files are on the left of next level files
+  // level 1
+  AddFile(1, 100, 200);
+  AddFile(1, 300, 400);
+  AddFile(1, 500, 600);
+  // level 2
+  AddFile(2, 1500, 1600);
+  AddFile(2, 1601, 1699);
+  AddFile(2, 1700, 1800);
+  // level 3
+  AddFile(3, 2500, 2600);
+  AddFile(3, 2601, 2699);
+  AddFile(3, 2700, 2800);
+  indexer.UpdateIndex(files);
+  int32_t left = 100;
+  int32_t right = 100;
+  for (uint32_t level = 1; level < 3; ++level) {
+    for (uint32_t f = 0; f < 3; ++f) {
+      GetNextLevelIndex(level, f, -1, -1, &left, &right);
+      ASSERT_EQ(0, left);
+      ASSERT_EQ(-1, right);
+      GetNextLevelIndex(level, f, 0, -1, &left, &right);
+      ASSERT_EQ(0, left);
+      ASSERT_EQ(-1, right);
+      GetNextLevelIndex(level, f, 1, -1, &left, &right);
+      ASSERT_EQ(0, left);
+      ASSERT_EQ(-1, right);
+      GetNextLevelIndex(level, f, 1, 0, &left, &right);
+      ASSERT_EQ(0, left);
+      ASSERT_EQ(-1, right);
+      GetNextLevelIndex(level, f, 1, 1, &left, &right);
+      ASSERT_EQ(0, left);
+      ASSERT_EQ(2, right);
+    }
+  }
+
+  // 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));
+  }
+  // level 1
+  AddFile(1, 2100, 2200);
+  AddFile(1, 2300, 2400);
+  AddFile(1, 2500, 2600);
+  // level 2
+  AddFile(2, 1500, 1600);
+  AddFile(2, 1501, 1699);
+  AddFile(2, 1700, 1800);
+  // level 3
+  AddFile(3, 500, 600);
+  AddFile(3, 501, 699);
+  AddFile(3, 700, 800);
+  indexer.UpdateIndex(files);
+  for (uint32_t level = 1; level < 3; ++level) {
+    for (uint32_t f = 0; f < 3; ++f) {
+      GetNextLevelIndex(level, f, -1, -1, &left, &right);
+      ASSERT_EQ(f == 0 ? 0 : 3, left);
+      ASSERT_EQ(2, right);
+      GetNextLevelIndex(level, f, 0, -1, &left, &right);
+      ASSERT_EQ(3, left);
+      ASSERT_EQ(2, right);
+      GetNextLevelIndex(level, f, 1, -1, &left, &right);
+      ASSERT_EQ(3, left);
+      ASSERT_EQ(2, right);
+      GetNextLevelIndex(level, f, 1, -1, &left, &right);
+      ASSERT_EQ(3, left);
+      ASSERT_EQ(2, right);
+      GetNextLevelIndex(level, f, 1, 0, &left, &right);
+      ASSERT_EQ(3, left);
+      ASSERT_EQ(2, right);
+      GetNextLevelIndex(level, f, 1, 1, &left, &right);
+      ASSERT_EQ(3, left);
+      ASSERT_EQ(2, right);
+    }
+  }
+
+  // Case 3: empty L2
+  Reset();
+  for (uint32_t i = 1; i < kNumLevels; ++i) {
+    ASSERT_EQ(0U, indexer.LevelIndexSize(i));
+  }
+  // level 1
+  AddFile(1, 2100, 2200);
+  AddFile(1, 2300, 2400);
+  AddFile(1, 2500, 2600);
+  // level 3
+  AddFile(3, 500, 600);
+  AddFile(3, 501, 699);
+  AddFile(3, 700, 800);
+  indexer.UpdateIndex(files);
+  for (uint32_t f = 0; f < 3; ++f) {
+    GetNextLevelIndex(1, f, -1, -1, &left, &right);
+    ASSERT_EQ(0, left);
+    ASSERT_EQ(-1, right);
+    GetNextLevelIndex(1, f, 0, -1, &left, &right);
+    ASSERT_EQ(0, left);
+    ASSERT_EQ(-1, right);
+    GetNextLevelIndex(1, f, 1, -1, &left, &right);
+    ASSERT_EQ(0, left);
+    ASSERT_EQ(-1, right);
+    GetNextLevelIndex(1, f, 1, -1, &left, &right);
+    ASSERT_EQ(0, left);
+    ASSERT_EQ(-1, right);
+    GetNextLevelIndex(1, f, 1, 0, &left, &right);
+    ASSERT_EQ(0, left);
+    ASSERT_EQ(-1, right);
+    GetNextLevelIndex(1, f, 1, 1, &left, &right);
+    ASSERT_EQ(0, left);
+    ASSERT_EQ(-1, right);
+  }
+
+
+  // Case 4: mixed
+  Reset();
+  for (uint32_t i = 1; i < kNumLevels; ++i) {
+    ASSERT_EQ(0U, indexer.LevelIndexSize(i));
+  }
+  // level 1
+  AddFile(1, 100, 200);
+  AddFile(1, 250, 400);
+  AddFile(1, 450, 500);
+  // level 2
+  AddFile(2, 100, 150);  // 0
+  AddFile(2, 200, 250);  // 1
+  AddFile(2, 251, 300);  // 2
+  AddFile(2, 301, 350);  // 3
+  AddFile(2, 500, 600);  // 4
+  // level 3
+  AddFile(3, 0, 50);
+  AddFile(3, 100, 200);
+  AddFile(3, 201, 250);
+  indexer.UpdateIndex(files);
+  // level 1, 0
+  GetNextLevelIndex(1, 0, -1, -1, &left, &right);
+  ASSERT_EQ(0, left);
+  ASSERT_EQ(0, right);
+  GetNextLevelIndex(1, 0, 0, -1, &left, &right);
+  ASSERT_EQ(0, left);
+  ASSERT_EQ(0, right);
+  GetNextLevelIndex(1, 0, 1, -1, &left, &right);
+  ASSERT_EQ(0, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(1, 0, 1, 0, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(1, 0, 1, 1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(4, right);
+  // level 1, 1
+  GetNextLevelIndex(1, 1, -1, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(1, 1, 0, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(1, 1, 1, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(3, right);
+  GetNextLevelIndex(1, 1, 1, 0, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(3, right);
+  GetNextLevelIndex(1, 1, 1, 1, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(4, right);
+  // level 1, 2
+  GetNextLevelIndex(1, 2, -1, -1, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(3, right);
+  GetNextLevelIndex(1, 2, 0, -1, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(3, right);
+  GetNextLevelIndex(1, 2, 1, -1, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(4, right);
+  GetNextLevelIndex(1, 2, 1, 0, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(4, right);
+  GetNextLevelIndex(1, 2, 1, 1, &left, &right);
+  ASSERT_EQ(4, left);
+  ASSERT_EQ(4, right);
+  // level 2, 0
+  GetNextLevelIndex(2, 0, -1, -1, &left, &right);
+  ASSERT_EQ(0, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(2, 0, 0, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(2, 0, 1, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(2, 0, 1, 0, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(2, 0, 1, 1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(2, right);
+  // level 2, 1
+  GetNextLevelIndex(2, 1, -1, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(2, 1, 0, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(1, right);
+  GetNextLevelIndex(2, 1, 1, -1, &left, &right);
+  ASSERT_EQ(1, left);
+  ASSERT_EQ(2, right);
+  GetNextLevelIndex(2, 1, 1, 0, &left, &right);
+  ASSERT_EQ(2, left);
+  ASSERT_EQ(2, right);
+  GetNextLevelIndex(2, 1, 1, 1, &left, &right);
+  ASSERT_EQ(2, left);
+  ASSERT_EQ(2, right);
+  // level 2, [2 - 4], no overlap
+  for (uint32_t f = 2; f <= 4; ++f) {
+    GetNextLevelIndex(2, f, -1, -1, &left, &right);
+    ASSERT_EQ(f == 2 ? 2 : 3, left);
+    ASSERT_EQ(2, right);
+    GetNextLevelIndex(2, f, 0, -1, &left, &right);
+    ASSERT_EQ(3, left);
+    ASSERT_EQ(2, right);
+    GetNextLevelIndex(2, f, 1, -1, &left, &right);
+    ASSERT_EQ(3, left);
+    ASSERT_EQ(2, right);
+    GetNextLevelIndex(2, f, 1, 0, &left, &right);
+    ASSERT_EQ(3, left);
+    ASSERT_EQ(2, right);
+    GetNextLevelIndex(2, f, 1, 1, &left, &right);
+    ASSERT_EQ(3, left);
+    ASSERT_EQ(2, right);
+  }
+}
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  return rocksdb::test::RunAllTests();
+}
@@ -44,10 +44,6 @@ static int FlattenPath(const std::string& path, char* dest, int len) {
  return write_idx;
 }

-// A utility routine: write "data" to the named file and Sync() it.
-extern Status WriteStringToFileSync(Env* env, const Slice& data,
-                                    const std::string& fname);
-
 static std::string MakeFileName(const std::string& name, uint64_t number,
                                const char* suffix) {
  char buf[100];
@@ -92,7 +88,6 @@ std::string LockFileName(const std::string& dbname) {
 }

 std::string TempFileName(const std::string& dbname, uint64_t number) {
-  assert(number >= 0);
  return MakeFileName(dbname, number, "dbtmp");
 }

@@ -238,7 +233,7 @@ Status SetCurrentFile(Env* env, const std::string& dbname,
  assert(contents.starts_with(dbname + "/"));
  contents.remove_prefix(dbname.size() + 1);
  std::string tmp = TempFileName(dbname, descriptor_number);
-  Status s = WriteStringToFileSync(env, contents.ToString() + "\n", tmp);
+  Status s = WriteStringToFile(env, contents.ToString() + "\n", tmp, true);
  if (s.ok()) {
    s = env->RenameFile(tmp, CurrentFileName(dbname));
  }
@@ -253,7 +248,7 @@ Status SetIdentityFile(Env* env, const std::string& dbname) {
  assert(!id.empty());
  // Reserve the filename dbname/000000.dbtmp for the temporary identity file
  std::string tmp = TempFileName(dbname, 0);
-  Status s = WriteStringToFileSync(env, id, tmp);
+  Status s = WriteStringToFile(env, id, tmp, true);
  if (s.ok()) {
    s = env->RenameFile(tmp, IdentityFileName(dbname));
  }
@@ -0,0 +1,369 @@
+//  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/internal_stats.h"
+#include "db/column_family.h"
+
+#include <vector>
+
+namespace rocksdb {
+
+DBPropertyType GetPropertyType(const Slice& property) {
+  Slice in = property;
+  Slice prefix("rocksdb.");
+  if (!in.starts_with(prefix)) return kUnknown;
+  in.remove_prefix(prefix.size());
+
+  if (in.starts_with("num-files-at-level")) {
+    return kNumFilesAtLevel;
+  } else if (in == "levelstats") {
+    return kLevelStats;
+  } else if (in == "stats") {
+    return kStats;
+  } else if (in == "sstables") {
+    return kSsTables;
+  } else if (in == "num-immutable-mem-table") {
+    return kNumImmutableMemTable;
+  } else if (in == "mem-table-flush-pending") {
+    return kMemtableFlushPending;
+  } else if (in == "compaction-pending") {
+    return kCompactionPending;
+  } else if (in == "background-errors") {
+    return kBackgroundErrors;
+  } else if (in == "cur-size-active-mem-table") {
+    return kCurSizeActiveMemTable;
+  } else if (in == "num-entries-active-mem-table") {
+    return kNumEntriesInMutableMemtable;
+  } else if (in == "num-entries-imm-mem-tables") {
+    return kNumEntriesInImmutableMemtable;
+  }
+  return kUnknown;
+}
+
+bool InternalStats::GetProperty(DBPropertyType property_type,
+                                const Slice& property, std::string* value,
+                                ColumnFamilyData* cfd) {
+  Version* current = cfd->current();
+  Slice in = property;
+
+  switch (property_type) {
+    case kNumFilesAtLevel: {
+      in.remove_prefix(strlen("rocksdb.num-files-at-level"));
+      uint64_t level;
+      bool ok = ConsumeDecimalNumber(&in, &level) && in.empty();
+      if (!ok || (int)level >= number_levels_) {
+        return false;
+      } else {
+        char buf[100];
+        snprintf(buf, sizeof(buf), "%d",
+                 current->NumLevelFiles(static_cast<int>(level)));
+        *value = buf;
+        return true;
+      }
+    }
+    case kLevelStats: {
+      char buf[1000];
+      snprintf(buf, sizeof(buf),
+               "Level Files Size(MB)\n"
+               "--------------------\n");
+      value->append(buf);
+
+      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);
+        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);
+      }
+
+      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);
+        }
+      }
+
+      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 + wal_bytes) /
+              (interval_bytes_new + 1),
+          (double)(interval_bytes_written + interval_bytes_read + 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 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;
+  }
+}
+
+}  // namespace rocksdb
@@ -0,0 +1,187 @@
+//  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.
+//
+
+#pragma once
+#include "rocksdb/statistics.h"
+#include "util/statistics.h"
+#include "db/version_set.h"
+
+#include <vector>
+#include <string>
+
+class ColumnFamilyData;
+
+namespace rocksdb {
+
+class MemTableList;
+class DBImpl;
+
+enum DBPropertyType {
+  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
+  kSsTables,         // Return a human readable string of current SST files
+  kNumImmutableMemTable,   // Return number of immutable mem tables
+  kMemtableFlushPending,   // Return 1 if mem table flushing is pending,
+                           // otherwise 0.
+  kCompactionPending,      // Return 1 if a compaction is pending. Otherwise 0.
+  kBackgroundErrors,       // Return accumulated background errors encountered.
+  kCurSizeActiveMemTable,  // Return current size of the active memtable
+  kNumEntriesInMutableMemtable,    // Return number of entries in the mutable
+                                   // memtable.
+  kNumEntriesInImmutableMemtable,  // Return sum of number of entries in all
+                                   // the immutable mem tables.
+  kUnknown,
+};
+
+extern DBPropertyType GetPropertyType(const Slice& property);
+
+class InternalStats {
+ public:
+  enum WriteStallType {
+    LEVEL0_SLOWDOWN,
+    MEMTABLE_COMPACTION,
+    LEVEL0_NUM_FILES,
+    WRITE_STALLS_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),
+        bg_error_count_(0),
+        number_levels_(num_levels),
+        statistics_(statistics),
+        env_(env),
+        started_at_(env->NowMicros()) {}
+
+  // Per level compaction stats.  compaction_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;
+
+    // Bytes read from level N+1 during compaction between levels N and N+1
+    int64_t bytes_readnp1;
+
+    // Total bytes written during compaction between levels N and N+1
+    int64_t bytes_written;
+
+    // Files read from level N during compaction between levels N and N+1
+    int files_in_leveln;
+
+    // Files read from level N+1 during compaction between levels N and N+1
+    int files_in_levelnp1;
+
+    // Files written during compaction between levels N and N+1
+    int files_out_levelnp1;
+
+    // Number of compactions done
+    int count;
+
+    CompactionStats()
+        : micros(0),
+          bytes_readn(0),
+          bytes_readnp1(0),
+          bytes_written(0),
+          files_in_leveln(0),
+          files_in_levelnp1(0),
+          files_out_levelnp1(0),
+          count(0) {}
+
+    void Add(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->count += 1;
+    }
+  };
+
+  void AddCompactionStats(int level, const CompactionStats& stats) {
+    compaction_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) {
+    stall_leveln_slowdown_[level] += micros;
+    stall_leveln_slowdown_count_[level] += micros;
+  }
+
+  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);
+
+ private:
+  std::vector<CompactionStats> compaction_stats_;
+
+  // 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
+    // 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_;
+
+    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_;
+
+  // Total number of background errors encountered. Every time a flush task
+  // or compaction task fails, this counter is incremented. The failure can
+  // be caused by any possible reason, including file system errors, out of
+  // resources, or input file corruption. Failing when retrying the same flush
+  // or compaction will cause the counter to increase too.
+  uint64_t bg_error_count_;
+
+  int number_levels_;
+  Statistics* statistics_;
+  Env* env_;
+  uint64_t started_at_;
+};
+
+}  // namespace rocksdb
@@ -0,0 +1,79 @@
+//  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 <vector>
+
+#include "util/testharness.h"
+#include "util/benchharness.h"
+#include "db/version_set.h"
+#include "util/mutexlock.h"
+
+namespace rocksdb {
+
+std::string MakeKey(unsigned int num) {
+  char buf[30];
+  snprintf(buf, sizeof(buf), "%016u", num);
+  return std::string(buf);
+}
+
+void BM_LogAndApply(int iters, int num_base_files) {
+  VersionSet* vset;
+  ColumnFamilyData* default_cfd;
+  uint64_t fnum = 1;
+  port::Mutex mu;
+  MutexLock l(&mu);
+
+  BENCHMARK_SUSPEND {
+    std::string dbname = test::TmpDir() + "/rocksdb_test_benchmark";
+    ASSERT_OK(DestroyDB(dbname, Options()));
+
+    DB* db = nullptr;
+    Options opts;
+    opts.create_if_missing = true;
+    Status s = DB::Open(opts, dbname, &db);
+    ASSERT_OK(s);
+    ASSERT_TRUE(db != nullptr);
+
+    delete db;
+    db = nullptr;
+
+    Options options;
+    EnvOptions sopt;
+    vset = new VersionSet(dbname, &options, sopt, nullptr);
+    std::vector<ColumnFamilyDescriptor> dummy;
+    dummy.push_back(ColumnFamilyDescriptor());
+    ASSERT_OK(vset->Recover(dummy));
+    default_cfd = vset->GetColumnFamilySet()->GetDefault();
+    VersionEdit vbase;
+    for (int i = 0; i < num_base_files; i++) {
+      InternalKey start(MakeKey(2 * fnum), 1, kTypeValue);
+      InternalKey limit(MakeKey(2 * fnum + 1), 1, kTypeDeletion);
+      vbase.AddFile(2, ++fnum, 1 /* file size */, start, limit, 1, 1);
+    }
+    ASSERT_OK(vset->LogAndApply(default_cfd, &vbase, &mu));
+  }
+
+  for (int i = 0; i < iters; i++) {
+    VersionEdit vedit;
+    vedit.DeleteFile(2, fnum);
+    InternalKey start(MakeKey(2 * fnum), 1, kTypeValue);
+    InternalKey limit(MakeKey(2 * fnum + 1), 1, kTypeDeletion);
+    vedit.AddFile(2, ++fnum, 1 /* file size */, start, limit, 1, 1);
+    vset->LogAndApply(default_cfd, &vedit, &mu);
+  }
+}
+
+BENCHMARK_NAMED_PARAM(BM_LogAndApply, 1000_iters_1_file, 1000, 1)
+BENCHMARK_NAMED_PARAM(BM_LogAndApply, 1000_iters_100_files, 1000, 100)
+BENCHMARK_NAMED_PARAM(BM_LogAndApply, 1000_iters_10000_files, 1000, 10000)
+BENCHMARK_NAMED_PARAM(BM_LogAndApply, 100_iters_100000_files, 100, 100000)
+
+}  // namespace rocksdb
+
+int main(int argc, char** argv) {
+  rocksdb::benchmark::RunBenchmarks();
+  return 0;
+}
@@ -17,7 +17,6 @@ namespace log {
 enum RecordType {
  // Zero is reserved for preallocated files
  kZeroType = 0,
-
  kFullType = 1,

  // For fragments
@@ -140,7 +140,9 @@ bool Reader::ReadRecord(Slice* record, std::string* scratch) {

      case kEof:
        if (in_fragmented_record) {
-          ReportCorruption(scratch->size(), "partial record without end(3)");
+          // This can be caused by the writer dying immediately after
+          //  writing a physical record but before completing the next; don't
+          //  treat it as a corruption, just ignore the entire logical record.
          scratch->clear();
        }
        return false;
@@ -264,13 +266,12 @@ unsigned int Reader::ReadPhysicalRecord(Slice* result) {
          eof_offset_ = buffer_.size();
        }
        continue;
-      } else if (buffer_.size() == 0) {
-        // End of file
-        return kEof;
      } else {
-        size_t drop_size = buffer_.size();
+        // Note that if buffer_ is non-empty, we have a truncated header at the
+        //  end of the file, which can be caused by the writer crashing in the
+        //  middle of writing the header. Instead of considering this an error,
+        //  just report EOF.
        buffer_.clear();
-        ReportCorruption(drop_size, "truncated record at end of file");
        return kEof;
      }
    }
@@ -284,14 +285,22 @@ unsigned int Reader::ReadPhysicalRecord(Slice* result) {
    if (kHeaderSize + length > buffer_.size()) {
      size_t drop_size = buffer_.size();
      buffer_.clear();
-      ReportCorruption(drop_size, "bad record length");
-      return kBadRecord;
+      if (!eof_) {
+        ReportCorruption(drop_size, "bad record length");
+        return kBadRecord;
+      }
+      // If the end of the file has been reached without reading |length| bytes
+      // of payload, assume the writer died in the middle of writing the record.
+      // Don't report a corruption.
+      return kEof;
    }

    if (type == kZeroType && length == 0) {
      // Skip zero length record without reporting any drops since
      // such records are produced by the mmap based writing code in
      // env_posix.cc that preallocates file regions.
+      // NOTE: this should never happen in DB written by new RocksDB versions,
+      // since we turn off mmap writes to manifest and log files
      buffer_.clear();
      return kBadRecord;
    }
@@ -446,20 +446,32 @@ TEST(LogTest, BadRecordType) {
  ASSERT_EQ("OK", MatchError("unknown record type"));
 }

-TEST(LogTest, TruncatedTrailingRecord) {
+TEST(LogTest, TruncatedTrailingRecordIsIgnored) {
  Write("foo");
  ShrinkSize(4);   // Drop all payload as well as a header byte
  ASSERT_EQ("EOF", Read());
-  ASSERT_EQ((unsigned int)(kHeaderSize - 1), DroppedBytes());
-  ASSERT_EQ("OK", MatchError("truncated record at end of file"));
+  // Truncated last record is ignored, not treated as an error
+  ASSERT_EQ(0U, DroppedBytes());
+  ASSERT_EQ("", ReportMessage());
 }

 TEST(LogTest, BadLength) {
+  const int kPayloadSize = kBlockSize - kHeaderSize;
+  Write(BigString("bar", kPayloadSize));
+  Write("foo");
+  // Least significant size byte is stored in header[4].
+  IncrementByte(4, 1);
+  ASSERT_EQ("foo", Read());
+  ASSERT_EQ(kBlockSize, DroppedBytes());
+  ASSERT_EQ("OK", MatchError("bad record length"));
+}
+
+TEST(LogTest, BadLengthAtEndIsIgnored) {
  Write("foo");
  ShrinkSize(1);
  ASSERT_EQ("EOF", Read());
-  ASSERT_EQ((unsigned int)(kHeaderSize + 2), DroppedBytes());
-  ASSERT_EQ("OK", MatchError("bad record length"));
+  ASSERT_EQ(0U, DroppedBytes());
+  ASSERT_EQ("", ReportMessage());
 }

 TEST(LogTest, ChecksumMismatch) {
@@ -510,6 +522,24 @@ TEST(LogTest, UnexpectedFirstType) {
  ASSERT_EQ("OK", MatchError("partial record without end"));
 }

+TEST(LogTest, MissingLastIsIgnored) {
+  Write(BigString("bar", kBlockSize));
+  // Remove the LAST block, including header.
+  ShrinkSize(14);
+  ASSERT_EQ("EOF", Read());
+  ASSERT_EQ("", ReportMessage());
+  ASSERT_EQ(0U, DroppedBytes());
+}
+
+TEST(LogTest, PartialLastIsIgnored) {
+  Write(BigString("bar", kBlockSize));
+  // Cause a bad record length in the LAST block.
+  ShrinkSize(1);
+  ASSERT_EQ("EOF", Read());
+  ASSERT_EQ("", ReportMessage());
+  ASSERT_EQ(0U, DroppedBytes());
+}
+
 TEST(LogTest, ErrorJoinsRecords) {
  // Consider two fragmented records:
  //    first(R1) last(R1) first(R2) last(R2)
@@ -10,6 +10,8 @@
 #include "db/memtable.h"

 #include <memory>
+#include <algorithm>
+#include <limits>

 #include "db/dbformat.h"
 #include "db/merge_context.h"
@@ -17,52 +19,133 @@
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
 #include "rocksdb/merge_operator.h"
+#include "rocksdb/slice_transform.h"
+#include "util/arena.h"
 #include "util/coding.h"
-#include "util/mutexlock.h"
 #include "util/murmurhash.h"
+#include "util/mutexlock.h"
+#include "util/perf_context_imp.h"
 #include "util/statistics.h"
-
-namespace std {
-template <>
-struct hash<rocksdb::Slice> {
-  size_t operator()(const rocksdb::Slice& slice) const {
-    return MurmurHash(slice.data(), slice.size(), 0);
-  }
-};
-}
+#include "util/stop_watch.h"

 namespace rocksdb {

 MemTable::MemTable(const InternalKeyComparator& cmp, const Options& options)
    : comparator_(cmp),
      refs_(0),
-      arena_impl_(options.arena_block_size),
-      table_(options.memtable_factory->CreateMemTableRep(comparator_,
-                                                         &arena_impl_)),
+      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())),
+      num_entries_(0),
      flush_in_progress_(false),
      flush_completed_(false),
      file_number_(0),
      first_seqno_(0),
      mem_next_logfile_number_(0),
-      mem_logfile_number_(0),
      locks_(options.inplace_update_support ? options.inplace_update_num_locks
-                                            : 0) {}
+                                            : 0),
+      prefix_extractor_(options.prefix_extractor.get()),
+      should_flush_(ShouldFlushNow()) {
+  // 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) {
+    prefix_bloom_.reset(new DynamicBloom(
+        options.memtable_prefix_bloom_bits, options.bloom_locality,
+        options.memtable_prefix_bloom_probes, nullptr,
+        options.memtable_prefix_bloom_huge_page_tlb_size));
+  }
+}

 MemTable::~MemTable() {
  assert(refs_ == 0);
 }

 size_t MemTable::ApproximateMemoryUsage() {
-  return arena_impl_.ApproximateMemoryUsage() +
-         table_->ApproximateMemoryUsage();
+  size_t arena_usage = arena_.ApproximateMemoryUsage();
+  size_t table_usage = table_->ApproximateMemoryUsage();
+  // let MAX_USAGE =  std::numeric_limits<size_t>::max()
+  // then if arena_usage + total_usage >= MAX_USAGE, return MAX_USAGE.
+  // the following variation is to avoid numeric overflow.
+  if (arena_usage >= std::numeric_limits<size_t>::max() - table_usage) {
+    return std::numeric_limits<size_t>::max();
+  }
+  // otherwise, return the actual usage
+  return arena_usage + table_usage;
 }

-int MemTable::KeyComparator::operator()(const char* aptr, const char* bptr)
+bool MemTable::ShouldFlushNow() const {
+  // In a lot of times, we cannot allocate arena blocks that exactly matches the
+  // buffer size. Thus we have to decide if we should over-allocate or
+  // under-allocate.
+  // This constant avariable can be interpreted as: if we still have more than
+  // "kAllowOverAllocationRatio * kArenaBlockSize" space left, we'd try to over
+  // allocate one more block.
+  const double kAllowOverAllocationRatio = 0.6;
+
+  // If arena still have room for new block allocation, we can safely say it
+  // shouldn't flush.
+  auto allocated_memory =
+      table_->ApproximateMemoryUsage() + arena_.MemoryAllocatedBytes();
+
+  // 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) {
+    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) {
+    return true;
+  }
+
+  // In this code path, Arena has already allocated its "last block", which
+  // means the total allocatedmemory size is either:
+  //  (1) "moderately" over allocated the memory (no more than `0.6 * arena
+  // block size`. Or,
+  //  (2) the allocated memory is less than write buffer size, but we'll stop
+  // here since if we allocate a new arena block, we'll over allocate too much
+  // more (half of the arena block size) memory.
+  //
+  // In either case, to avoid over-allocate, the last block will stop allocation
+  // when its usage reaches a certain ratio, which we carefully choose "0.75
+  // full" as the stop condition because it addresses the following issue with
+  // great simplicity: What if the next inserted entry's size is
+  // bigger than AllocatedAndUnused()?
+  //
+  // The answer is: if the entry size is also bigger than 0.25 *
+  // kArenaBlockSize, a dedicated block will be allocated for it; otherwise
+  // arena will anyway skip the AllocatedAndUnused() and allocate a new, empty
+  // and regular block. In either case, we *overly* over-allocated.
+  //
+  // Therefore, setting the last block to be at most "0.75 full" avoids both
+  // cases.
+  //
+  // NOTE: the average percentage of waste space of this approach can be counted
+  // as: "arena block size * 0.25 / write buffer size". User who specify a small
+  // write buffer size and/or big arena block size may suffer.
+  return arena_.AllocatedAndUnused() < kArenaBlockSize / 4;
+}
+
+int MemTable::KeyComparator::operator()(const char* prefix_len_key1,
+                                        const char* prefix_len_key2) const {
+  // Internal keys are encoded as length-prefixed strings.
+  Slice k1 = GetLengthPrefixedSlice(prefix_len_key1);
+  Slice k2 = GetLengthPrefixedSlice(prefix_len_key2);
+  return comparator.Compare(k1, k2);
+}
+
+int MemTable::KeyComparator::operator()(const char* prefix_len_key,
+                                        const Slice& key)
    const {
  // Internal keys are encoded as length-prefixed strings.
-  Slice a = GetLengthPrefixedSlice(aptr);
-  Slice b = GetLengthPrefixedSlice(bptr);
-  return comparator.Compare(a, b);
+  Slice a = GetLengthPrefixedSlice(prefix_len_key);
+  return comparator.Compare(a, key);
 }

 Slice MemTableRep::UserKey(const char* key) const {
@@ -70,10 +153,15 @@ Slice MemTableRep::UserKey(const char* key) const {
  return Slice(slice.data(), slice.size() - 8);
 }

+KeyHandle MemTableRep::Allocate(const size_t len, char** buf) {
+  *buf = arena_->Allocate(len);
+  return static_cast<KeyHandle>(*buf);
+}
+
 // Encode a suitable internal key target for "target" and return it.
 // Uses *scratch as scratch space, and the returned pointer will point
 // into this scratch space.
-static const char* EncodeKey(std::string* scratch, const Slice& target) {
+const char* EncodeKey(std::string* scratch, const Slice& target) {
  scratch->clear();
  PutVarint32(scratch, target.size());
  scratch->append(target.data(), target.size());
@@ -82,27 +170,53 @@ static const char* EncodeKey(std::string* scratch, const Slice& target) {

 class MemTableIterator: public Iterator {
 public:
-  MemTableIterator(MemTableRep* table, const ReadOptions& options)
-    : iter_() {
-    if (options.prefix) {
-      iter_.reset(table->GetPrefixIterator(*options.prefix));
-    } else if (options.prefix_seek) {
-      iter_.reset(table->GetDynamicPrefixIterator());
+  MemTableIterator(const MemTable& mem, const ReadOptions& options,
+                   bool enforce_total_order)
+      : bloom_(nullptr),
+        prefix_extractor_(mem.prefix_extractor_),
+        valid_(false) {
+    if (prefix_extractor_ != nullptr && !enforce_total_order) {
+      bloom_ = mem.prefix_bloom_.get();
+      iter_.reset(mem.table_->GetDynamicPrefixIterator());
    } else {
-      iter_.reset(table->GetIterator());
+      iter_.reset(mem.table_->GetIterator());
    }
  }

-  virtual bool Valid() const { return iter_->Valid(); }
-  virtual void Seek(const Slice& k) { iter_->Seek(EncodeKey(&tmp_, k)); }
-  virtual void SeekToFirst() { iter_->SeekToFirst(); }
-  virtual void SeekToLast() { iter_->SeekToLast(); }
-  virtual void Next() { iter_->Next(); }
-  virtual void Prev() { iter_->Prev(); }
+  virtual bool Valid() const { return valid_; }
+  virtual void Seek(const Slice& k) {
+    if (bloom_ != nullptr &&
+        !bloom_->MayContain(prefix_extractor_->Transform(ExtractUserKey(k)))) {
+      valid_ = false;
+      return;
+    }
+    iter_->Seek(k, nullptr);
+    valid_ = iter_->Valid();
+  }
+  virtual void SeekToFirst() {
+    iter_->SeekToFirst();
+    valid_ = iter_->Valid();
+  }
+  virtual void SeekToLast() {
+    iter_->SeekToLast();
+    valid_ = iter_->Valid();
+  }
+  virtual void Next() {
+    assert(Valid());
+    iter_->Next();
+    valid_ = iter_->Valid();
+  }
+  virtual void Prev() {
+    assert(Valid());
+    iter_->Prev();
+    valid_ = iter_->Valid();
+  }
  virtual Slice key() const {
+    assert(Valid());
    return GetLengthPrefixedSlice(iter_->key());
  }
  virtual Slice value() const {
+    assert(Valid());
    Slice key_slice = GetLengthPrefixedSlice(iter_->key());
    return GetLengthPrefixedSlice(key_slice.data() + key_slice.size());
  }
@@ -110,24 +224,28 @@ class MemTableIterator: public Iterator {
  virtual Status status() const { return Status::OK(); }

 private:
+  DynamicBloom* bloom_;
+  const SliceTransform* const prefix_extractor_;
  std::unique_ptr<MemTableRep::Iterator> iter_;
-  std::string tmp_;       // For passing to EncodeKey
+  bool valid_;

  // No copying allowed
  MemTableIterator(const MemTableIterator&);
  void operator=(const MemTableIterator&);
 };

-Iterator* MemTable::NewIterator(const ReadOptions& options) {
-  return new MemTableIterator(table_.get(), options);
+Iterator* MemTable::NewIterator(const ReadOptions& options,
+    bool enforce_total_order) {
+  return new MemTableIterator(*this, options, enforce_total_order);
 }

 port::RWMutex* MemTable::GetLock(const Slice& key) {
-  return &locks_[std::hash<Slice>()(key) % locks_.size()];
+  static murmur_hash hash;
+  return &locks_[hash(key) % locks_.size()];
 }

 void MemTable::Add(SequenceNumber s, ValueType type,
-                   const Slice& key,
+                   const Slice& key, /* user key */
                   const Slice& value) {
  // Format of an entry is concatenation of:
  //  key_size     : varint32 of internal_key.size()
@@ -140,7 +258,9 @@ void MemTable::Add(SequenceNumber s, ValueType type,
  const size_t encoded_len =
      VarintLength(internal_key_size) + internal_key_size +
      VarintLength(val_size) + val_size;
-  char* buf = arena_impl_.Allocate(encoded_len);
+  char* buf = nullptr;
+  KeyHandle handle = table_->Allocate(encoded_len, &buf);
+  assert(buf != nullptr);
  char* p = EncodeVarint32(buf, internal_key_size);
  memcpy(p, key.data(), key_size);
  p += key_size;
@@ -148,132 +268,174 @@ void MemTable::Add(SequenceNumber s, ValueType type,
  p += 8;
  p = EncodeVarint32(p, val_size);
  memcpy(p, value.data(), val_size);
-  assert((p + val_size) - buf == (unsigned)encoded_len);
-  table_->Insert(buf);
+  assert((unsigned)(p + val_size - buf) == (unsigned)encoded_len);
+  table_->Insert(handle);
+  num_entries_++;
+
+  if (prefix_bloom_) {
+    assert(prefix_extractor_);
+    prefix_bloom_->Add(prefix_extractor_->Transform(key));
+  }

  // The first sequence number inserted into the memtable
  assert(first_seqno_ == 0 || s > first_seqno_);
  if (first_seqno_ == 0) {
    first_seqno_ = s;
  }
+
+  should_flush_ = ShouldFlushNow();
+}
+
+// Callback from MemTable::Get()
+namespace {
+
+struct Saver {
+  Status* status;
+  const LookupKey* key;
+  bool* found_final_value;  // Is value set correctly? Used by KeyMayExist
+  bool* merge_in_progress;
+  std::string* value;
+  const MergeOperator* merge_operator;
+  // the merge operations encountered;
+  MergeContext* merge_context;
+  MemTable* mem;
+  Logger* logger;
+  Statistics* statistics;
+  bool inplace_update_support;
+};
+}  // namespace
+
+static bool SaveValue(void* arg, const char* entry) {
+  Saver* s = reinterpret_cast<Saver*>(arg);
+  MergeContext* merge_context = s->merge_context;
+  const MergeOperator* merge_operator = s->merge_operator;
+
+  assert(s != nullptr && merge_context != nullptr);
+
+  // entry format is:
+  //    klength  varint32
+  //    userkey  char[klength-8]
+  //    tag      uint64
+  //    vlength  varint32
+  //    value    char[vlength]
+  // Check that it belongs to same user key.  We do not check the
+  // sequence number since the Seek() call above should have skipped
+  // all entries with overly large sequence numbers.
+  uint32_t key_length;
+  const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
+  if (s->mem->GetInternalKeyComparator().user_comparator()->Compare(
+          Slice(key_ptr, key_length - 8), s->key->user_key()) == 0) {
+    // Correct user key
+    const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
+    switch (static_cast<ValueType>(tag & 0xff)) {
+      case kTypeValue: {
+        if (s->inplace_update_support) {
+          s->mem->GetLock(s->key->user_key())->ReadLock();
+        }
+        Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
+        *(s->status) = Status::OK();
+        if (*(s->merge_in_progress)) {
+          assert(merge_operator);
+          if (!merge_operator->FullMerge(s->key->user_key(), &v,
+                                         merge_context->GetOperands(), s->value,
+                                         s->logger)) {
+            RecordTick(s->statistics, NUMBER_MERGE_FAILURES);
+            *(s->status) =
+                Status::Corruption("Error: Could not perform merge.");
+          }
+        } else {
+          s->value->assign(v.data(), v.size());
+        }
+        if (s->inplace_update_support) {
+          s->mem->GetLock(s->key->user_key())->Unlock();
+        }
+        *(s->found_final_value) = true;
+        return false;
+      }
+      case kTypeDeletion: {
+        if (*(s->merge_in_progress)) {
+          assert(merge_operator);
+          *(s->status) = Status::OK();
+          if (!merge_operator->FullMerge(s->key->user_key(), nullptr,
+                                         merge_context->GetOperands(), s->value,
+                                         s->logger)) {
+            RecordTick(s->statistics, NUMBER_MERGE_FAILURES);
+            *(s->status) =
+                Status::Corruption("Error: Could not perform merge.");
+          }
+        } else {
+          *(s->status) = Status::NotFound();
+        }
+        *(s->found_final_value) = true;
+        return false;
+      }
+      case kTypeMerge: {
+        std::string merge_result;  // temporary area for merge results later
+        Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
+        *(s->merge_in_progress) = true;
+        merge_context->PushOperand(v);
+        return true;
+      }
+      default:
+        assert(false);
+        return true;
+    }
+  }
+
+  // s->state could be Corrupt, merge or notfound
+  return false;
 }

 bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
                   MergeContext& merge_context, const Options& options) {
-  Slice memkey = key.memtable_key();
-  std::unique_ptr<MemTableRep::Iterator> iter(
-      table_->GetIterator(key.user_key()));
-  iter->Seek(memkey.data());
+  PERF_TIMER_AUTO(get_from_memtable_time);

+  Slice user_key = key.user_key();
+  bool found_final_value = false;
  bool merge_in_progress = s->IsMergeInProgress();
-  auto merge_operator = options.merge_operator.get();
-  auto logger = options.info_log;
-  std::string merge_result;

-  for (; iter->Valid(); iter->Next()) {
-    // entry format is:
-    //    klength  varint32
-    //    userkey  char[klength-8]
-    //    tag      uint64
-    //    vlength  varint32
-    //    value    char[vlength]
-    // Check that it belongs to same user key.  We do not check the
-    // sequence number since the Seek() call above should have skipped
-    // all entries with overly large sequence numbers.
-    const char* entry = iter->key();
-    uint32_t key_length;
-    const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
-    if (comparator_.comparator.user_comparator()->Compare(
-        Slice(key_ptr, key_length - 8), key.user_key()) == 0) {
-      // Correct user key
-      const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
-      switch (static_cast<ValueType>(tag & 0xff)) {
-        case kTypeValue: {
-          if (options.inplace_update_support) {
-            GetLock(key.user_key())->ReadLock();
-          }
-          Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
-          *s = Status::OK();
-          if (merge_in_progress) {
-            assert(merge_operator);
-          if (!merge_operator->FullMerge(key.user_key(), &v,
-                                         merge_context.GetOperands(), value,
-                                         logger.get())) {
-              RecordTick(options.statistics.get(), NUMBER_MERGE_FAILURES);
-              *s = Status::Corruption("Error: Could not perform merge.");
-            }
-          } else {
-            value->assign(v.data(), v.size());
-          }
-          if (options.inplace_update_support) {
-            GetLock(key.user_key())->Unlock();
-          }
-          return true;
-        }
-        case kTypeDeletion: {
-          if (merge_in_progress) {
-            assert(merge_operator);
-            *s = Status::OK();
-          if (!merge_operator->FullMerge(key.user_key(), nullptr,
-                                         merge_context.GetOperands(), value,
-                                         logger.get())) {
-              RecordTick(options.statistics.get(), NUMBER_MERGE_FAILURES);
-              *s = Status::Corruption("Error: Could not perform merge.");
-            }
-          } else {
-            *s = Status::NotFound();
-          }
-          return true;
-        }
-        case kTypeMerge: {
-          Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
-          merge_in_progress = true;
-          merge_context.PushOperand(v);
-          while(merge_context.GetNumOperands() >= 2) {
-            // Attempt to associative merge. (Returns true if successful)
-          if (merge_operator->PartialMerge(key.user_key(),
-                                           merge_context.GetOperand(0),
-                                           merge_context.GetOperand(1),
-                                           &merge_result, logger.get())) {
-              merge_context.PushPartialMergeResult(merge_result);
-            } else {
-              // Stack them because user can't associative merge
-              break;
-            }
-          }
-          break;
-        }
-        case kTypeLogData:
-          assert(false);
-          break;
-      }
-    } else {
-      // exit loop if user key does not match
-      break;
-    }
+  if (prefix_bloom_ &&
+      !prefix_bloom_->MayContain(prefix_extractor_->Transform(user_key))) {
+    // iter is null if prefix bloom says the key does not exist
+  } else {
+    Saver saver;
+    saver.status = s;
+    saver.found_final_value = &found_final_value;
+    saver.merge_in_progress = &merge_in_progress;
+    saver.key = &key;
+    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();
+    table_->Get(key, &saver, SaveValue);
  }

  // No change to value, since we have not yet found a Put/Delete
-
-  if (merge_in_progress) {
+  if (!found_final_value && merge_in_progress) {
    *s = Status::MergeInProgress("");
  }
-  return false;
+  PERF_TIMER_STOP(get_from_memtable_time);
+  PERF_COUNTER_ADD(get_from_memtable_count, 1);
+  return found_final_value;
 }

-bool MemTable::Update(SequenceNumber seq, ValueType type,
+void MemTable::Update(SequenceNumber seq,
                      const Slice& key,
                      const Slice& value) {
  LookupKey lkey(key, seq);
-  Slice memkey = lkey.memtable_key();
+  Slice mem_key = lkey.memtable_key();

  std::unique_ptr<MemTableRep::Iterator> iter(
-      table_->GetIterator(lkey.user_key()));
-  iter->Seek(memkey.data());
+    table_->GetIterator(lkey.user_key()));
+  iter->Seek(lkey.internal_key(), mem_key.data());

  if (iter->Valid()) {
    // entry format is:
-    //    klength  varint32
+    //    key_length  varint32
    //    userkey  char[klength-8]
    //    tag      uint64
    //    vlength  varint32
@@ -282,7 +444,7 @@ bool MemTable::Update(SequenceNumber seq, ValueType type,
    // sequence number since the Seek() call above should have skipped
    // all entries with overly large sequence numbers.
    const char* entry = iter->key();
-    uint32_t key_length;
+    uint32_t key_length = 0;
    const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
    if (comparator_.comparator.user_comparator()->Compare(
        Slice(key_ptr, key_length - 8), lkey.user_key()) == 0) {
@@ -290,36 +452,108 @@ bool MemTable::Update(SequenceNumber seq, ValueType type,
      const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
      switch (static_cast<ValueType>(tag & 0xff)) {
        case kTypeValue: {
-          uint32_t vlength;
-          GetVarint32Ptr(key_ptr + key_length,
-                         key_ptr + key_length+5, &vlength);
-          // Update value, if newValue size  <= curValue size
-          if (value.size() <= vlength) {
+          Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
+          uint32_t prev_size = prev_value.size();
+          uint32_t new_size = value.size();
+
+          // Update value, if new value size  <= previous value size
+          if (new_size <= prev_size ) {
            char* p = EncodeVarint32(const_cast<char*>(key_ptr) + key_length,
-                                     value.size());
+                                     new_size);
            WriteLock wl(GetLock(lkey.user_key()));
            memcpy(p, value.data(), value.size());
-            assert(
-              (p + value.size()) - entry ==
-              (unsigned) (VarintLength(key_length) +
-                          key_length +
-                          VarintLength(value.size()) +
-                          value.size())
-            );
-            return true;
+            assert((unsigned)((p + value.size()) - entry) ==
+                   (unsigned)(VarintLength(key_length) + key_length +
+                              VarintLength(value.size()) + value.size()));
+            return;
          }
        }
        default:
          // If the latest value is kTypeDeletion, kTypeMerge or kTypeLogData
-          // then we probably don't have enough space to update in-place
-          // Maybe do something later
-          // Return false, and do normal Add()
-          return false;
+          // we don't have enough space for update inplace
+            Add(seq, kTypeValue, key, value);
+            return;
      }
    }
  }

-  // Key doesn't exist
+  // key doesn't exist
+  Add(seq, kTypeValue, key, value);
+}
+
+bool MemTable::UpdateCallback(SequenceNumber seq,
+                              const Slice& key,
+                              const Slice& delta,
+                              const Options& options) {
+  LookupKey lkey(key, seq);
+  Slice memkey = lkey.memtable_key();
+
+  std::unique_ptr<MemTableRep::Iterator> iter(
+    table_->GetIterator(lkey.user_key()));
+  iter->Seek(lkey.internal_key(), memkey.data());
+
+  if (iter->Valid()) {
+    // entry format is:
+    //    key_length  varint32
+    //    userkey  char[klength-8]
+    //    tag      uint64
+    //    vlength  varint32
+    //    value    char[vlength]
+    // Check that it belongs to same user key.  We do not check the
+    // sequence number since the Seek() call above should have skipped
+    // all entries with overly large sequence numbers.
+    const char* entry = iter->key();
+    uint32_t key_length = 0;
+    const char* key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
+    if (comparator_.comparator.user_comparator()->Compare(
+        Slice(key_ptr, key_length - 8), lkey.user_key()) == 0) {
+      // Correct user key
+      const uint64_t tag = DecodeFixed64(key_ptr + key_length - 8);
+      switch (static_cast<ValueType>(tag & 0xff)) {
+        case kTypeValue: {
+          Slice prev_value = GetLengthPrefixedSlice(key_ptr + key_length);
+          uint32_t  prev_size = prev_value.size();
+
+          char* prev_buffer = const_cast<char*>(prev_value.data());
+          uint32_t  new_prev_size = prev_size;
+
+          std::string str_value;
+          WriteLock wl(GetLock(lkey.user_key()));
+          auto status = options.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);
+            if (new_prev_size < prev_size) {
+              // overwrite the new prev_size
+              char* p = EncodeVarint32(const_cast<char*>(key_ptr) + key_length,
+                                       new_prev_size);
+              if (VarintLength(new_prev_size) < VarintLength(prev_size)) {
+                // shift the value buffer as well.
+                memcpy(p, prev_buffer, new_prev_size);
+              }
+            }
+            RecordTick(options.statistics.get(), 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);
+            should_flush_ = ShouldFlushNow();
+            return true;
+          } else if (status == UpdateStatus::UPDATE_FAILED) {
+            // No action required. Return.
+            should_flush_ = ShouldFlushNow();
+            return true;
+          }
+        }
+        default:
+          break;
+      }
+    }
+  }
+  // If the latest value is not kTypeValue
+  // or key doesn't exist
  return false;
 }

@@ -331,16 +565,16 @@ size_t MemTable::CountSuccessiveMergeEntries(const LookupKey& key) {
  // The iterator only needs to be ordered within the same user key.
  std::unique_ptr<MemTableRep::Iterator> iter(
      table_->GetIterator(key.user_key()));
-  iter->Seek(memkey.data());
+  iter->Seek(key.internal_key(), memkey.data());

  size_t num_successive_merges = 0;

  for (; iter->Valid(); iter->Next()) {
    const char* entry = iter->key();
-    uint32_t key_length;
+    uint32_t key_length = 0;
    const char* iter_key_ptr = GetVarint32Ptr(entry, entry + 5, &key_length);
-    if (!comparator_.comparator.user_comparator()->Compare(
-        Slice(iter_key_ptr, key_length - 8), key.user_key()) == 0) {
+    if (comparator_.comparator.user_comparator()->Compare(
+            Slice(iter_key_ptr, key_length - 8), key.user_key()) != 0) {
      break;
    }

@@ -355,4 +589,13 @@ size_t MemTable::CountSuccessiveMergeEntries(const LookupKey& key) {
  return num_successive_merges;
 }

+void MemTableRep::Get(const LookupKey& k, void* callback_args,
+                      bool (*callback_func)(void* arg, const char* entry)) {
+  auto iter = GetIterator(k.user_key());
+  for (iter->Seek(k.internal_key(), k.memtable_key().data());
+       iter->Valid() && callback_func(callback_args, iter->key());
+       iter->Next()) {
+  }
+}
+
 }  // namespace rocksdb
@@ -13,10 +13,11 @@
 #include <deque>
 #include "db/dbformat.h"
 #include "db/skiplist.h"
-#include "db/version_set.h"
+#include "db/version_edit.h"
 #include "rocksdb/db.h"
 #include "rocksdb/memtablerep.h"
-#include "util/arena_impl.h"
+#include "util/arena.h"
+#include "util/dynamic_bloom.h"

 namespace rocksdb {

@@ -29,13 +30,16 @@ class MemTable {
  struct KeyComparator : public MemTableRep::KeyComparator {
    const InternalKeyComparator comparator;
    explicit KeyComparator(const InternalKeyComparator& c) : comparator(c) { }
-    virtual int operator()(const char* a, const char* b) const;
+    virtual int operator()(const char* prefix_len_key1,
+                           const char* prefix_len_key2) const;
+    virtual int operator()(const char* prefix_len_key,
+                           const Slice& key) const override;
  };

  // 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 = Options());
+                    const Options& options);

  ~MemTable();

@@ -60,6 +64,10 @@ class MemTable {
  // operations on the same MemTable.
  size_t ApproximateMemoryUsage();

+  // This method heuristically determines if the memtable should continue to
+  // host more data.
+  bool ShouldFlush() const { return should_flush_; }
+
  // Return an iterator that yields the contents of the memtable.
  //
  // The caller must ensure that the underlying MemTable remains live
@@ -67,14 +75,10 @@ class MemTable {
  // iterator are internal keys encoded by AppendInternalKey in the
  // db/dbformat.{h,cc} module.
  //
-  // If options.prefix is supplied, it is passed to the underlying MemTableRep
-  // as a hint that the iterator only need to support access to keys with that
-  // specific prefix.
-  // If options.prefix is not supplied and options.prefix_seek is set, the
-  // iterator is not bound to a specific prefix. However, the semantics of
-  // Seek is changed - the result might only include keys with the same prefix
-  // as the seek-key.
-  Iterator* NewIterator(const ReadOptions& options = ReadOptions());
+  // By default, it returns an iterator for prefix seek if prefix_extractor
+  // is configured in Options.
+  Iterator* NewIterator(const ReadOptions& options,
+                        bool enforce_total_order = false);

  // Add an entry into memtable that maps key to value at the
  // specified sequence number and with the specified type.
@@ -94,21 +98,39 @@ class MemTable {
  bool Get(const LookupKey& key, std::string* value, Status* s,
           MergeContext& merge_context, const Options& options);

-  // Update the value and return status ok,
-  //   if key exists in current memtable
-  //     if new sizeof(new_value) <= sizeof(old_value) &&
-  //       old_value for that key is a put i.e. kTypeValue
-  //     else return false, and status - NotUpdatable()
-  //   else return false, and status - NotFound()
-  bool Update(SequenceNumber seq, ValueType type,
+  // Attempts to update the new_value inplace, else does normal Add
+  // Pseudocode
+  //   if key exists in current memtable && prev_value is of type kTypeValue
+  //     if new sizeof(new_value) <= sizeof(prev_value)
+  //       update inplace
+  //     else add(key, new_value)
+  //   else add(key, new_value)
+  void Update(SequenceNumber seq,
              const Slice& key,
              const Slice& value);

+  // If prev_value for key exits, attempts to update it inplace.
+  // else returns false
+  // Pseudocode
+  //   if key exists in current memtable && prev_value is of type kTypeValue
+  //     new_value = delta(prev_value)
+  //     if sizeof(new_value) <= sizeof(prev_value)
+  //       update inplace
+  //     else add(key, new_value)
+  //   else return false
+  bool UpdateCallback(SequenceNumber seq,
+                      const Slice& key,
+                      const Slice& delta,
+                      const Options& options);
+
  // 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
  // key in the memtable.
  size_t CountSuccessiveMergeEntries(const LookupKey& key);

+  // Get total number of entries in the mem table.
+  uint64_t GetNumEntries() const { return num_entries_; }
+
  // Returns the edits area that is needed for flushing the memtable
  VersionEdit* GetEdits() { return &edit_; }

@@ -124,33 +146,49 @@ class MemTable {
  // be flushed to storage
  void SetNextLogNumber(uint64_t num) { mem_next_logfile_number_ = num; }

-  // Returns the logfile number that can be safely deleted when this
-  // memstore is flushed to storage
-  uint64_t GetLogNumber() { return mem_logfile_number_; }
-
-  // Sets the logfile number that can be safely deleted when this
-  // memstore is flushed to storage
-  void SetLogNumber(uint64_t num) { mem_logfile_number_ = num; }
-
  // Notify the underlying storage that no more items will be added
  void MarkImmutable() { table_->MarkReadOnly(); }

+  // return true if the current MemTableRep supports merge operator.
+  bool IsMergeOperatorSupported() const {
+    return table_->IsMergeOperatorSupported();
+  }
+
+  // return true if the current MemTableRep supports snapshots.
+  bool IsSnapshotSupported() const { return table_->IsSnapshotSupported(); }
+
+  // Get the lock associated for the key
+  port::RWMutex* GetLock(const Slice& key);
+
+  const InternalKeyComparator& GetInternalKeyComparator() const {
+    return comparator_.comparator;
+  }
+
+  const Arena& TEST_GetArena() const { return arena_; }
+
 private:
+  // Dynamically check if we can add more incoming entries.
+  bool ShouldFlushNow() const;
+
  friend class MemTableIterator;
  friend class MemTableBackwardIterator;
  friend class MemTableList;

  KeyComparator comparator_;
  int refs_;
-  ArenaImpl arena_impl_;
+  const size_t kArenaBlockSize;
+  const size_t kWriteBufferSize;
+  Arena arena_;
  unique_ptr<MemTableRep> table_;

+  uint64_t num_entries_;
+
  // These are used to manage memtable flushes to storage
  bool flush_in_progress_; // started the flush
  bool flush_completed_;   // finished the flush
  uint64_t file_number_;    // filled up after flush is complete

-  // The udpates to be applied to the transaction log when this
+  // The updates to be applied to the transaction log when this
  // memtable is flushed to storage.
  VersionEdit edit_;

@@ -160,10 +198,6 @@ class MemTable {
  // The log files earlier than this number can be deleted.
  uint64_t mem_next_logfile_number_;

-  // The log file that backs this memtable (to be deleted when
-  // memtable flush is done)
-  uint64_t mem_logfile_number_;
-
  // rw locks for inplace updates
  std::vector<port::RWMutex> locks_;

@@ -171,8 +205,13 @@ class MemTable {
  MemTable(const MemTable&);
  void operator=(const MemTable&);

-  // Get the lock associated for the key
-  port::RWMutex* GetLock(const Slice& key);
+  const SliceTransform* const prefix_extractor_;
+  std::unique_ptr<DynamicBloom> prefix_bloom_;
+
+  // a flag indicating if a memtable has met the criteria to flush
+  bool should_flush_;
 };

+extern const char* EncodeKey(std::string* scratch, const Slice& target);
+
 }  // namespace rocksdb
@@ -3,14 +3,16 @@
 //  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/memtablelist.h"
+#include "db/memtable_list.h"

 #include <string>
 #include "rocksdb/db.h"
 #include "db/memtable.h"
+#include "db/version_set.h"
 #include "rocksdb/env.h"
 #include "rocksdb/iterator.h"
 #include "util/coding.h"
+#include "util/log_buffer.h"

 namespace rocksdb {

@@ -30,7 +32,7 @@ MemTableListVersion::MemTableListVersion(MemTableListVersion* old) {

 void MemTableListVersion::Ref() { ++refs_; }

-void MemTableListVersion::Unref(std::vector<MemTable*>* to_delete) {
+void MemTableListVersion::Unref(autovector<MemTable*>* to_delete) {
  assert(refs_ >= 1);
  --refs_;
  if (refs_ == 0) {
@@ -76,6 +78,14 @@ void MemTableListVersion::AddIterators(const ReadOptions& options,
  }
 }

+uint64_t MemTableListVersion::GetTotalNumEntries() const {
+  uint64_t total_num = 0;
+  for (auto& m : memlist_) {
+    total_num += m->GetNumEntries();
+  }
+  return total_num;
+}
+
 // caller is responsible for referencing m
 void MemTableListVersion::Add(MemTable* m) {
  assert(refs_ == 1);  // only when refs_ == 1 is MemTableListVersion mutable
@@ -92,7 +102,7 @@ void MemTableListVersion::Remove(MemTable* m) {

 // Returns true if there is at least one memtable on which flush has
 // not yet started.
-bool MemTableList::IsFlushPending() {
+bool MemTableList::IsFlushPending() const {
  if ((flush_requested_ && num_flush_not_started_ >= 1) ||
      (num_flush_not_started_ >= min_write_buffer_number_to_merge_)) {
    assert(imm_flush_needed.NoBarrier_Load() != nullptr);
@@ -102,7 +112,7 @@ bool MemTableList::IsFlushPending() {
 }

 // Returns the memtables that need to be flushed.
-void MemTableList::PickMemtablesToFlush(std::vector<MemTable*>* ret) {
+void MemTableList::PickMemtablesToFlush(autovector<MemTable*>* ret) {
  const auto& memlist = current_->memlist_;
  for (auto it = memlist.rbegin(); it != memlist.rend(); ++it) {
    MemTable* m = *it;
@@ -112,38 +122,41 @@ void MemTableList::PickMemtablesToFlush(std::vector<MemTable*>* ret) {
      if (num_flush_not_started_ == 0) {
        imm_flush_needed.Release_Store(nullptr);
      }
-      m->flush_in_progress_ = true; // flushing will start very soon
+      m->flush_in_progress_ = true;  // flushing will start very soon
      ret->push_back(m);
    }
  }
-  flush_requested_ = false; // start-flush request is complete
+  flush_requested_ = false;  // start-flush request is complete
+}
+
+void MemTableList::RollbackMemtableFlush(const autovector<MemTable*>& mems,
+                                         uint64_t file_number,
+                                         std::set<uint64_t>* pending_outputs) {
+  assert(!mems.empty());
+
+  // If the flush was not successful, then just reset state.
+  // Maybe a suceeding attempt to flush will be successful.
+  for (MemTable* m : mems) {
+    assert(m->flush_in_progress_);
+    assert(m->file_number_ == 0);
+
+    m->flush_in_progress_ = false;
+    m->flush_completed_ = false;
+    m->edit_.Clear();
+    num_flush_not_started_++;
+  }
+  pending_outputs->erase(file_number);
+  imm_flush_needed.Release_Store(reinterpret_cast<void *>(1));
 }

 // Record a successful flush in the manifest file
 Status MemTableList::InstallMemtableFlushResults(
-    const std::vector<MemTable*>& mems, VersionSet* vset, Status flushStatus,
+    ColumnFamilyData* cfd, const autovector<MemTable*>& mems, VersionSet* vset,
    port::Mutex* mu, Logger* info_log, uint64_t file_number,
-    std::set<uint64_t>& pending_outputs, std::vector<MemTable*>* to_delete,
-    Directory* db_directory) {
+    std::set<uint64_t>& pending_outputs, autovector<MemTable*>* to_delete,
+    Directory* db_directory, LogBuffer* log_buffer) {
  mu->AssertHeld();

-  // If the flush was not successful, then just reset state.
-  // Maybe a suceeding attempt to flush will be successful.
-  if (!flushStatus.ok()) {
-    for (MemTable* m : mems) {
-      assert(m->flush_in_progress_);
-      assert(m->file_number_ == 0);
-
-      m->flush_in_progress_ = false;
-      m->flush_completed_ = false;
-      m->edit_.Clear();
-      num_flush_not_started_++;
-      imm_flush_needed.Release_Store((void *)1);
-      pending_outputs.erase(file_number);
-    }
-    return flushStatus;
-  }
-
  // flush was sucessful
  for (size_t i = 0; i < mems.size(); ++i) {
    // All the edits are associated with the first memtable of this batch.
@@ -171,12 +184,11 @@ Status MemTableList::InstallMemtableFlushResults(
      break;
    }

-    Log(info_log,
-        "Level-0 commit table #%lu started",
-        (unsigned long)m->file_number_);
+    LogToBuffer(log_buffer, "[%s] Level-0 commit table #%lu started",
+                cfd->GetName().c_str(), (unsigned long)m->file_number_);

    // this can release and reacquire the mutex.
-    s = vset->LogAndApply(&m->edit_, mu, db_directory);
+    s = vset->LogAndApply(cfd, &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
@@ -187,10 +199,10 @@ Status MemTableList::InstallMemtableFlushResults(
    uint64_t mem_id = 1;  // how many memtables has been flushed.
    do {
      if (s.ok()) { // commit new state
-        Log(info_log,
-            "Level-0 commit table #%lu: memtable #%lu done",
-            (unsigned long)m->file_number_,
-            (unsigned long)mem_id);
+        LogToBuffer(log_buffer,
+                    "[%s] Level-0 commit table #%lu: memtable #%lu done",
+                    cfd->GetName().c_str(), (unsigned long)m->file_number_,
+                    (unsigned long)mem_id);
        current_->Remove(m);
        assert(m->file_number_ > 0);

@@ -215,7 +227,6 @@ Status MemTableList::InstallMemtableFlushResults(
        pending_outputs.erase(m->file_number_);
        m->file_number_ = 0;
        imm_flush_needed.Release_Store((void *)1);
-        s = Status::IOError("Unable to commit flushed memtable");
      }
      ++mem_id;
    } while (!current_->memlist_.empty() && (m = current_->memlist_.back()) &&
@@ -3,20 +3,29 @@
 //  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 <string>
 #include <list>
 #include <vector>
+#include <set>
+#include <deque>
 #include "rocksdb/db.h"
 #include "rocksdb/options.h"
 #include "rocksdb/iterator.h"
+
 #include "db/dbformat.h"
 #include "db/skiplist.h"
-#include "memtable.h"
+#include "db/memtable.h"
+#include "rocksdb/db.h"
+#include "rocksdb/iterator.h"
+#include "rocksdb/options.h"
+#include "util/autovector.h"
+#include "util/log_buffer.h"

 namespace rocksdb {

+class ColumnFamilyData;
 class InternalKeyComparator;
 class Mutex;

@@ -28,7 +37,7 @@ class MemTableListVersion {
  explicit MemTableListVersion(MemTableListVersion* old = nullptr);

  void Ref();
-  void Unref(std::vector<MemTable*>* to_delete = nullptr);
+  void Unref(autovector<MemTable*>* to_delete = nullptr);

  int size() const;

@@ -40,6 +49,8 @@ class MemTableListVersion {
  void AddIterators(const ReadOptions& options,
                    std::vector<Iterator*>* iterator_list);

+  uint64_t GetTotalNumEntries() const;
+
 private:
  // REQUIRE: m is mutable memtable
  void Add(MemTable* m);
@@ -74,8 +85,8 @@ class MemTableList {

  MemTableListVersion* current() { return current_; }

-  // so that backgrund threads can detect non-nullptr pointer to
-  // determine whether this is anything more to start flushing.
+  // so that background threads can detect non-nullptr pointer to
+  // determine whether there is anything more to start flushing.
  port::AtomicPointer imm_flush_needed;

  // Returns the total number of memtables in the list
@@ -83,20 +94,27 @@ class MemTableList {

  // Returns true if there is at least one memtable on which flush has
  // not yet started.
-  bool IsFlushPending();
+  bool IsFlushPending() const;

  // Returns the earliest memtables that needs to be flushed. The returned
  // memtables are guaranteed to be in the ascending order of created time.
-  void PickMemtablesToFlush(std::vector<MemTable*>* mems);
+  void PickMemtablesToFlush(autovector<MemTable*>* mems);
+
+  // Reset status of the given memtable list back to pending state so that
+  // they can get picked up again on the next round of flush.
+  void RollbackMemtableFlush(const autovector<MemTable*>& mems,
+                             uint64_t file_number,
+                             std::set<uint64_t>* pending_outputs);

  // Commit a successful flush in the manifest file
-  Status InstallMemtableFlushResults(const std::vector<MemTable*>& m,
-                                     VersionSet* vset, Status flushStatus,
-                                     port::Mutex* mu, Logger* info_log,
-                                     uint64_t file_number,
+  Status InstallMemtableFlushResults(ColumnFamilyData* cfd,
+                                     const autovector<MemTable*>& m,
+                                     VersionSet* vset, port::Mutex* mu,
+                                     Logger* info_log, uint64_t file_number,
                                     std::set<uint64_t>& pending_outputs,
-                                     std::vector<MemTable*>* to_delete,
-                                     Directory* db_directory);
+                                     autovector<MemTable*>* to_delete,
+                                     Directory* db_directory,
+                                     LogBuffer* log_buffer);

  // New memtables are inserted at the front of the list.
  // Takes ownership of the referenced held on *m by the caller of Add().
@@ -25,12 +25,12 @@ public:
      operand_list->clear();
    }
  }
-  // Replace the first two operands of merge_result, which are expected be the
-  // merge results of them.
+  // Replace all operands with merge_result, which are expected to be the
+  // merge result of them.
  void PushPartialMergeResult(std::string& merge_result) {
    assert (operand_list);
-    operand_list->pop_front();
-    swap(operand_list->front(), merge_result);
+    operand_list->clear();
+    operand_list->push_front(std::move(merge_result));
  }
  // Push a merge operand
  void PushOperand(const Slice& operand_slice) {
@@ -21,7 +21,7 @@ namespace rocksdb {
 //       operands_ stores the list of merge operands encountered while merging.
 //       keys_[i] corresponds to operands_[i] for each i.
 void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,
-                             bool at_bottom, Statistics* stats) {
+                             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.
  keys_.clear();
@@ -40,9 +40,12 @@ void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,
  ParseInternalKey(keys_.back(), &orig_ikey);

  bool hit_the_next_user_key = false;
-  ParsedInternalKey ikey;
  std::string merge_result;  // Temporary value for merge results
+  if (steps) {
+    ++(*steps);
+  }
  for (iter->Next(); iter->Valid(); iter->Next()) {
+    ParsedInternalKey ikey;
    assert(operands_.size() >= 1);        // Should be invariants!
    assert(keys_.size() == operands_.size());

@@ -91,6 +94,9 @@ void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,

      // move iter to the next entry (before doing anything else)
      iter->Next();
+      if (steps) {
+        ++(*steps);
+      }
      return;
    }

@@ -119,6 +125,9 @@ void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,

      // move iter to the next entry
      iter->Next();
+      if (steps) {
+        ++(*steps);
+      }
      return;
    }

@@ -129,28 +138,13 @@ void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,
      //   => then continue because we haven't yet seen a Put/Delete.
      assert(!operands_.empty()); // Should have at least one element in it

+      // keep queuing keys and operands until we either meet a put / delete
+      // request or later did a partial merge.
      keys_.push_front(iter->key().ToString());
      operands_.push_front(iter->value().ToString());
-      while (operands_.size() >= 2) {
-        // Returns false when the merge_operator can no longer process it
-        if (user_merge_operator_->PartialMerge(ikey.user_key,
-                                               Slice(operands_[0]),
-                                               Slice(operands_[1]),
-                                               &merge_result,
-                                               logger_)) {
-          // Merging of operands (associative merge) was successful.
-          // Replace these frontmost two operands with the merge result
-          keys_.pop_front();
-          operands_.pop_front();
-          swap(operands_.front(), merge_result);
-        } else {
-          // Merging of operands (associative merge) returned false.
-          // The user merge_operator does not know how to merge these operands.
-          // So we just stack them up until we find a Put/Delete or end of key.
-          break;
-        }
+      if (steps) {
+        ++(*steps);
      }
-      continue;
    }
  }

@@ -192,6 +186,23 @@ void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,
      RecordTick(stats, NUMBER_MERGE_FAILURES);
      // Do nothing if not success_. Leave keys() and operands() as they are.
    }
+  } else {
+    // We haven't seen the beginning of the key nor a Put/Delete.
+    // Attempt to use the user's associative merge function to
+    // merge the stacked merge operands into a single operand.
+
+    if (operands_.size() >= 2 &&
+        operands_.size() >= min_partial_merge_operands_ &&
+        user_merge_operator_->PartialMergeMulti(
+            orig_ikey.user_key,
+            std::deque<Slice>(operands_.begin(), operands_.end()),
+            &merge_result, logger_)) {
+      // Merging of operands (associative merge) was successful.
+      // Replace operands with the merge result
+      operands_.clear();
+      operands_.push_front(std::move(merge_result));
+      keys_.erase(keys_.begin(), keys_.end() - 1);
+    }
  }
 }

@@ -22,12 +22,13 @@ class Statistics;
 class MergeHelper {
 public:
  MergeHelper(const Comparator* user_comparator,
-              const MergeOperator* user_merge_operator,
-              Logger* logger,
+              const MergeOperator* user_merge_operator, Logger* logger,
+              unsigned min_partial_merge_operands,
              bool assert_valid_internal_key)
      : user_comparator_(user_comparator),
        user_merge_operator_(user_merge_operator),
        logger_(logger),
+        min_partial_merge_operands_(min_partial_merge_operands),
        assert_valid_internal_key_(assert_valid_internal_key),
        keys_(),
        operands_(),
@@ -46,7 +47,8 @@ class MergeHelper {
  // at_bottom:   (IN) true if the iterator covers the bottem level, which means
  //                   we could reach the start of the history of this user key.
  void MergeUntil(Iterator* iter, SequenceNumber stop_before = 0,
-                  bool at_bottom = false, Statistics* stats = nullptr);
+                  bool at_bottom = false, Statistics* stats = nullptr,
+                  int* steps = nullptr);

  // Query the merge result
  // These are valid until the next MergeUntil call
@@ -88,6 +90,7 @@ class MergeHelper {
  const Comparator* user_comparator_;
  const MergeOperator* user_merge_operator_;
  Logger* logger_;
+  unsigned min_partial_merge_operands_;
  bool assert_valid_internal_key_; // enforce no internal key corruption?

  // the scratch area that holds the result of MergeUntil
@@ -11,6 +11,31 @@

 namespace rocksdb {

+// The default implementation of PartialMergeMulti, which invokes
+// PartialMerge multiple times internally and merges two operands at
+// a time.
+bool MergeOperator::PartialMergeMulti(const Slice& key,
+                                      const std::deque<Slice>& operand_list,
+                                      std::string* new_value,
+                                      Logger* logger) const {
+  assert(operand_list.size() >= 2);
+  // Simply loop through the operands
+  std::string temp_value;
+  Slice temp_slice(operand_list[0]);
+
+  for (size_t i = 1; i < operand_list.size(); ++i) {
+    auto& operand = operand_list[i];
+    if (!PartialMerge(key, temp_slice, operand, &temp_value, logger)) {
+      return false;
+    }
+    swap(temp_value, *new_value);
+    temp_slice = Slice(*new_value);
+  }
+
+  // The result will be in *new_value. All merges succeeded.
+  return true;
+}
+
 // Given a "real" merge from the library, call the user's
 // associative merge function one-by-one on each of the operands.
 // NOTE: It is assumed that the client's merge-operator will handle any errors.
@@ -46,7 +71,6 @@ bool AssociativeMergeOperator::PartialMerge(
    const Slice& right_operand,
    std::string* new_value,
    Logger* logger) const {
-
  return Merge(key, &left_operand, right_operand, new_value, logger);
 }

@@ -17,17 +17,21 @@
 #include "db/write_batch_internal.h"
 #include "utilities/merge_operators.h"
 #include "util/testharness.h"
-#include "utilities/utility_db.h"
+#include "utilities/db_ttl.h"

 using namespace std;
 using namespace rocksdb;

 namespace {
  int numMergeOperatorCalls;
-
  void resetNumMergeOperatorCalls() {
    numMergeOperatorCalls = 0;
  }
+
+  int num_partial_merge_calls;
+  void resetNumPartialMergeCalls() {
+    num_partial_merge_calls = 0;
+  }
 }

 class CountMergeOperator : public AssociativeMergeOperator {
@@ -42,6 +46,11 @@ class CountMergeOperator : public AssociativeMergeOperator {
                     std::string* new_value,
                     Logger* logger) const override {
    ++numMergeOperatorCalls;
+    if (existing_value == nullptr) {
+      new_value->assign(value.data(), value.size());
+      return true;
+    }
+
    return mergeOperator_->PartialMerge(
        key,
        *existing_value,
@@ -50,6 +59,14 @@ class CountMergeOperator : public AssociativeMergeOperator {
        logger);
  }

+  virtual bool PartialMergeMulti(const Slice& key,
+                                 const std::deque<Slice>& operand_list,
+                                 std::string* new_value, Logger* logger) const {
+    ++num_partial_merge_calls;
+    return mergeOperator_->PartialMergeMulti(key, operand_list, new_value,
+                                             logger);
+  }
+
  virtual const char* Name() const override {
    return "UInt64AddOperator";
  }
@@ -58,22 +75,23 @@ class CountMergeOperator : public AssociativeMergeOperator {
  std::shared_ptr<MergeOperator> mergeOperator_;
 };

-std::shared_ptr<DB> OpenDb(
-    const string& dbname,
-    const bool ttl = false,
-    const unsigned max_successive_merges = 0) {
+namespace {
+std::shared_ptr<DB> OpenDb(const string& dbname, const bool ttl = false,
+                           const size_t max_successive_merges = 0,
+                           const uint32_t min_partial_merge_operands = 2) {
  DB* db;
-  StackableDB* sdb;
  Options options;
  options.create_if_missing = true;
  options.merge_operator = std::make_shared<CountMergeOperator>();
  options.max_successive_merges = max_successive_merges;
+  options.min_partial_merge_operands = min_partial_merge_operands;
  Status s;
  DestroyDB(dbname, Options());
  if (ttl) {
    cout << "Opening database with TTL\n";
-    s = UtilityDB::OpenTtlDB(options, dbname, &sdb);
-    db = sdb;
+    DBWithTTL* db_with_ttl;
+    s = DBWithTTL::Open(options, dbname, &db_with_ttl);
+    db = db_with_ttl;
  } else {
    s = DB::Open(options, dbname, &db);
  }
@@ -83,6 +101,7 @@ std::shared_ptr<DB> OpenDb(
  }
  return std::shared_ptr<DB>(db);
 }
+}  // namespace

 // Imagine we are maintaining a set of uint64 counters.
 // Each counter has a distinct name. And we would like
@@ -184,12 +203,16 @@ class Counters {

  uint64_t assert_get(const string& key) {
    uint64_t value = default_;
-    assert(get(key, &value));
+    int result = get(key, &value);
+    assert(result);
+    if (result == 0) exit(1); // Disable unused variable warning.
    return value;
  }

  void assert_add(const string& key, uint64_t value) {
-    assert(add(key, value));
+    int result = add(key, value);
+    assert(result);
+    if (result == 0) exit(1); // Disable unused variable warning. 
  }
 };

@@ -220,6 +243,7 @@ class MergeBasedCounters : public Counters {
  }
 };

+namespace {
 void dumpDb(DB* db) {
  auto it = unique_ptr<Iterator>(db->NewIterator(ReadOptions()));
  for (it->SeekToFirst(); it->Valid(); it->Next()) {
@@ -306,6 +330,44 @@ void testSuccessiveMerge(
  }
 }

+void testPartialMerge(Counters* counters, DB* db, int max_merge, int min_merge,
+                      int count) {
+  FlushOptions o;
+  o.wait = true;
+
+  // Test case 1: partial merge should be called when the number of merge
+  //              operands exceeds the threshold.
+  uint64_t tmp_sum = 0;
+  resetNumPartialMergeCalls();
+  for (int i = 1; i <= count; i++) {
+    counters->assert_add("b", i);
+    tmp_sum += i;
+  }
+  db->Flush(o);
+  db->CompactRange(nullptr, nullptr);
+  ASSERT_EQ(tmp_sum, counters->assert_get("b"));
+  if (count > max_merge) {
+    // in this case, FullMerge should be called instead.
+    ASSERT_EQ(num_partial_merge_calls, 0);
+  } else {
+    // if count >= min_merge, then partial merge should be called once.
+    ASSERT_EQ((count >= min_merge), (num_partial_merge_calls == 1));
+  }
+
+  // Test case 2: partial merge should not be called when a put is found.
+  resetNumPartialMergeCalls();
+  tmp_sum = 0;
+  db->Put(rocksdb::WriteOptions(), "c", "10");
+  for (int i = 1; i <= count; i++) {
+    counters->assert_add("c", i);
+    tmp_sum += i;
+  }
+  db->Flush(o);
+  db->CompactRange(nullptr, nullptr);
+  ASSERT_EQ(tmp_sum, counters->assert_get("c"));
+  ASSERT_EQ(num_partial_merge_calls, 0);
+}
+
 void testSingleBatchSuccessiveMerge(
    DB* db,
    int max_num_merges,
@@ -370,20 +432,41 @@ void runTest(int argc, const string& dbname, const bool use_ttl = false) {

  {
    cout << "Test merge in memtable... \n";
-    unsigned maxMerge = 5;
-    auto db = OpenDb(dbname, use_ttl, maxMerge);
+    size_t max_merge = 5;
+    auto db = OpenDb(dbname, use_ttl, max_merge);
    MergeBasedCounters counters(db, 0);
    testCounters(counters, db.get(), compact);
-    testSuccessiveMerge(counters, maxMerge, maxMerge * 2);
+    testSuccessiveMerge(counters, max_merge, max_merge * 2);
    testSingleBatchSuccessiveMerge(db.get(), 5, 7);
    DestroyDB(dbname, Options());
  }

+  {
+    cout << "Test Partial-Merge\n";
+    size_t max_merge = 100;
+    for (uint32_t min_merge = 5; min_merge < 25; min_merge += 5) {
+      for (uint32_t count = min_merge - 1; count <= min_merge + 1; count++) {
+        auto db = OpenDb(dbname, use_ttl, max_merge, min_merge);
+        MergeBasedCounters counters(db, 0);
+        testPartialMerge(&counters, db.get(), max_merge, min_merge, count);
+        DestroyDB(dbname, Options());
+      }
+      {
+        auto db = OpenDb(dbname, use_ttl, max_merge, min_merge);
+        MergeBasedCounters counters(db, 0);
+        testPartialMerge(&counters, db.get(), max_merge, min_merge,
+                         min_merge * 10);
+        DestroyDB(dbname, Options());
+      }
+    }
+  }
 }
+}  // namespace

 int main(int argc, char *argv[]) {
  //TODO: Make this test like a general rocksdb unit-test
  runTest(argc, test::TmpDir() + "/merge_testdb");
  runTest(argc, test::TmpDir() + "/merge_testdbttl", true); // Run test on TTL database
+  printf("Passed all tests!\n");
  return 0;
 }
@@ -10,6 +10,8 @@

 #include "rocksdb/db.h"
 #include "rocksdb/perf_context.h"
+#include "rocksdb/slice_transform.h"
+#include "rocksdb/memtablerep.h"
 #include "util/histogram.h"
 #include "util/stop_watch.h"
 #include "util/testharness.h"
@@ -174,6 +176,13 @@ void ProfileKeyComparison() {

  HistogramImpl hist_put;
  HistogramImpl hist_get;
+  HistogramImpl hist_get_snapshot;
+  HistogramImpl hist_get_memtable;
+  HistogramImpl hist_get_post_process;
+  HistogramImpl hist_num_memtable_checked;
+  HistogramImpl hist_write_pre_post;
+  HistogramImpl hist_write_wal_time;
+  HistogramImpl hist_write_memtable_time;

  std::cout << "Inserting " << FLAGS_total_keys << " key/value pairs\n...\n";

@@ -192,16 +201,37 @@ void ProfileKeyComparison() {

    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);

    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_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);
  }

  std::cout << "Put uesr key comparison: \n" << hist_put.ToString()
            << "Get 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"
+            << hist_write_wal_time.ToString() << "\n"
+            << " Writing Mem Table time: \n"
+            << hist_write_memtable_time.ToString() << "\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"
+            << " Number of memtables checked: \n"
+            << hist_num_memtable_checked.ToString() << "\n"
+            << " Time to post process: \n"
+            << hist_get_post_process.ToString() << "\n";
 }

 TEST(PerfContextTest, KeyComparisonCount) {
@@ -259,8 +289,8 @@ TEST(PerfContextTest, SeekKeyComparison) {
    db->Put(write_options, key, value);
    auto put_time = timer.ElapsedNanos();
    hist_put_time.Add(put_time);
-    hist_wal_time.Add(perf_context.wal_write_time);
-    hist_time_diff.Add(put_time - perf_context.wal_write_time);
+    hist_wal_time.Add(perf_context.write_wal_time);
+    hist_time_diff.Add(put_time - perf_context.write_wal_time);
  }

  std::cout << "Put time:\n" << hist_put_time.ToString()
@@ -0,0 +1,853 @@
+// 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 "db/db_impl.h"
+#include "db/filename.h"
+#include "db/version_set.h"
+#include "db/write_batch_internal.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/compaction_filter.h"
+#include "rocksdb/db.h"
+#include "rocksdb/env.h"
+#include "rocksdb/filter_policy.h"
+#include "rocksdb/slice_transform.h"
+#include "rocksdb/table.h"
+#include "table/meta_blocks.h"
+#include "table/plain_table_factory.h"
+#include "table/plain_table_reader.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;
+
+namespace rocksdb {
+
+class PlainTableDBTest {
+ protected:
+ private:
+  std::string dbname_;
+  Env* env_;
+  DB* db_;
+
+  Options last_options_;
+
+ public:
+  PlainTableDBTest() : env_(Env::Default()) {
+    dbname_ = test::TmpDir() + "/plain_table_db_test";
+    ASSERT_OK(DestroyDB(dbname_, Options()));
+    db_ = nullptr;
+    Reopen();
+  }
+
+  ~PlainTableDBTest() {
+    delete db_;
+    ASSERT_OK(DestroyDB(dbname_, Options()));
+  }
+
+  // Return the current option configuration.
+  Options CurrentOptions() {
+    Options options;
+    options.table_factory.reset(NewPlainTableFactory(16, 2, 0.8, 3));
+    options.prefix_extractor.reset(NewFixedPrefixTransform(8));
+    options.allow_mmap_reads = true;
+    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(PlainTableDBTest, Empty) {
+  ASSERT_TRUE(dbfull() != nullptr);
+  ASSERT_EQ("NOT_FOUND", Get("0000000000000foo"));
+}
+
+class TestPlainTableReader : public PlainTableReader {
+ public:
+  TestPlainTableReader(const EnvOptions& storage_options,
+                       const InternalKeyComparator& icomparator,
+                       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,
+                         file_size, bloom_bits_per_key, hash_table_ratio,
+                         index_sparseness, table_properties, 2 * 1024 * 1024),
+        expect_bloom_not_match_(expect_bloom_not_match) {
+    Status s = PopulateIndex(const_cast<TableProperties*>(table_properties));
+    ASSERT_TRUE(s.ok());
+  }
+
+  virtual ~TestPlainTableReader() {}
+
+ private:
+  virtual bool MatchBloom(uint32_t hash) const override {
+    bool ret = PlainTableReader::MatchBloom(hash);
+    ASSERT_TRUE(!*expect_bloom_not_match_ || !ret);
+    return ret;
+  }
+  bool* expect_bloom_not_match_;
+};
+
+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)
+      : PlainTableFactory(user_key_len, user_key_len, hash_table_ratio,
+                          index_sparseness, huge_page_tlb_size),
+        bloom_bits_per_key_(bloom_bits_per_key),
+        hash_table_ratio_(hash_table_ratio),
+        index_sparseness_(index_sparseness),
+        expect_bloom_not_match_(expect_bloom_not_match) {}
+
+  Status NewTableReader(const Options& options, const EnvOptions& soptions,
+                        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);
+    ASSERT_TRUE(s.ok());
+
+    std::unique_ptr<PlainTableReader> new_reader(new TestPlainTableReader(
+        soptions, internal_comparator, file_size, bloom_bits_per_key_,
+        hash_table_ratio_, index_sparseness_, props, std::move(file), options,
+        expect_bloom_not_match_));
+
+    *table = std::move(new_reader);
+    return s;
+  }
+
+ private:
+  int bloom_bits_per_key_;
+  double hash_table_ratio_;
+  size_t index_sparseness_;
+  bool* expect_bloom_not_match_;
+};
+
+TEST(PlainTableDBTest, Flush) {
+  for (size_t huge_page_tlb_size = 0; huge_page_tlb_size <= 2 * 1024 * 1024;
+       huge_page_tlb_size += 2 * 1024 * 1024) {
+    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.table_factory.reset(NewTotalOrderPlainTableFactory(
+              16, bloom_bits, 2, huge_page_tlb_size));
+        } else {
+          options.table_factory.reset(NewPlainTableFactory(
+              16, bloom_bits, 0.75, 16, huge_page_tlb_size));
+        }
+        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(total_order ? "9" : "0", (tp->user_collected_properties).at(
+                                               "plain_table_sub_index_size"));
+
+        ASSERT_EQ("v3", Get("1000000000000foo"));
+        ASSERT_EQ("v2", Get("0000000000000bar"));
+      }
+    }
+  }
+}
+
+TEST(PlainTableDBTest, Flush2) {
+  for (size_t huge_page_tlb_size = 0; huge_page_tlb_size <= 2 * 1024 * 1024;
+       huge_page_tlb_size += 2 * 1024 * 1024) {
+    for (int bloom_bits = 0; bloom_bits <= 117; bloom_bits += 117) {
+      for (int total_order = 0; total_order <= 1; total_order++) {
+        bool expect_bloom_not_match = false;
+        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 = nullptr;
+          options.table_factory.reset(
+              new TestPlainTableFactory(&expect_bloom_not_match, 16, bloom_bits,
+                                        0, 2, huge_page_tlb_size));
+        } else {
+          options.table_factory.reset(
+              new TestPlainTableFactory(&expect_bloom_not_match, 16, bloom_bits,
+                                        0.75, 16, huge_page_tlb_size));
+        }
+        DestroyAndReopen(&options);
+        ASSERT_OK(Put("0000000000000bar", "b"));
+        ASSERT_OK(Put("1000000000000foo", "v1"));
+        dbfull()->TEST_FlushMemTable();
+
+        ASSERT_OK(Put("1000000000000foo", "v2"));
+        dbfull()->TEST_FlushMemTable();
+        ASSERT_EQ("v2", Get("1000000000000foo"));
+
+        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"));
+        ASSERT_OK(Put("9000000000000eee", "v5"));
+        dbfull()->TEST_FlushMemTable();
+        ASSERT_EQ("v5", Get("0000000000000eee"));
+
+        // Test Bloom Filter
+        if (bloom_bits > 0) {
+          // 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"));
+            ASSERT_EQ("NOT_FOUND", Get("0000000000000not"));
+          }
+          expect_bloom_not_match = false;
+        }
+      }
+    }
+  }
+}
+
+TEST(PlainTableDBTest, Iterator) {
+  for (size_t huge_page_tlb_size = 0; huge_page_tlb_size <= 2 * 1024 * 1024;
+       huge_page_tlb_size += 2 * 1024 * 1024) {
+    for (int bloom_bits = 0; bloom_bits <= 117; bloom_bits += 117) {
+      for (int total_order = 0; total_order <= 1; total_order++) {
+        bool expect_bloom_not_match = false;
+        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 = nullptr;
+          options.table_factory.reset(
+              new TestPlainTableFactory(&expect_bloom_not_match, 16, bloom_bits,
+                                        0, 2, huge_page_tlb_size));
+        } else {
+          options.table_factory.reset(
+              new TestPlainTableFactory(&expect_bloom_not_match, 16, bloom_bits,
+                                        0.75, 16, huge_page_tlb_size));
+        }
+        DestroyAndReopen(&options);
+
+        ASSERT_OK(Put("1000000000foo002", "v_2"));
+        ASSERT_OK(Put("0000000000000bar", "random"));
+        ASSERT_OK(Put("1000000000foo001", "v1"));
+        ASSERT_OK(Put("3000000000000bar", "bar_v"));
+        ASSERT_OK(Put("1000000000foo003", "v__3"));
+        ASSERT_OK(Put("1000000000foo004", "v__4"));
+        ASSERT_OK(Put("1000000000foo005", "v__5"));
+        ASSERT_OK(Put("1000000000foo007", "v__7"));
+        ASSERT_OK(Put("1000000000foo008", "v__8"));
+        dbfull()->TEST_FlushMemTable();
+        ASSERT_EQ("v1", Get("1000000000foo001"));
+        ASSERT_EQ("v__3", Get("1000000000foo003"));
+        Iterator* iter = dbfull()->NewIterator(ReadOptions());
+        iter->Seek("1000000000foo000");
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo001", iter->key().ToString());
+        ASSERT_EQ("v1", iter->value().ToString());
+
+        iter->Next();
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo002", iter->key().ToString());
+        ASSERT_EQ("v_2", iter->value().ToString());
+
+        iter->Next();
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo003", iter->key().ToString());
+        ASSERT_EQ("v__3", iter->value().ToString());
+
+        iter->Next();
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo004", iter->key().ToString());
+        ASSERT_EQ("v__4", iter->value().ToString());
+
+        iter->Seek("3000000000000bar");
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("3000000000000bar", iter->key().ToString());
+        ASSERT_EQ("bar_v", iter->value().ToString());
+
+        iter->Seek("1000000000foo000");
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo001", iter->key().ToString());
+        ASSERT_EQ("v1", iter->value().ToString());
+
+        iter->Seek("1000000000foo005");
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo005", iter->key().ToString());
+        ASSERT_EQ("v__5", iter->value().ToString());
+
+        iter->Seek("1000000000foo006");
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo007", iter->key().ToString());
+        ASSERT_EQ("v__7", iter->value().ToString());
+
+        iter->Seek("1000000000foo008");
+        ASSERT_TRUE(iter->Valid());
+        ASSERT_EQ("1000000000foo008", iter->key().ToString());
+        ASSERT_EQ("v__8", iter->value().ToString());
+
+        if (total_order == 0) {
+          iter->Seek("1000000000foo009");
+          ASSERT_TRUE(iter->Valid());
+          ASSERT_EQ("3000000000000bar", iter->key().ToString());
+        }
+
+        // Test Bloom Filter
+        if (bloom_bits > 0) {
+          if (!total_order) {
+            // Neither key nor value should exist.
+            expect_bloom_not_match = true;
+            iter->Seek("2not000000000bar");
+            ASSERT_TRUE(!iter->Valid());
+            ASSERT_EQ("NOT_FOUND", Get("2not000000000bar"));
+            expect_bloom_not_match = false;
+          } else {
+            expect_bloom_not_match = true;
+            ASSERT_EQ("NOT_FOUND", Get("2not000000000bar"));
+            expect_bloom_not_match = false;
+          }
+        }
+
+        delete iter;
+      }
+    }
+  }
+}
+
+namespace {
+std::string MakeLongKey(size_t length, char c) {
+  return std::string(length, c);
+}
+}  // namespace
+
+TEST(PlainTableDBTest, IteratorLargeKeys) {
+  Options options = CurrentOptions();
+  options.table_factory.reset(NewTotalOrderPlainTableFactory(0, 0, 16));
+  options.create_if_missing = true;
+  options.prefix_extractor.reset();
+  DestroyAndReopen(&options);
+
+  std::string key_list[] = {
+      MakeLongKey(30, '0'),
+      MakeLongKey(16, '1'),
+      MakeLongKey(32, '2'),
+      MakeLongKey(60, '3'),
+      MakeLongKey(90, '4'),
+      MakeLongKey(50, '5'),
+      MakeLongKey(26, '6')
+  };
+
+  for (size_t i = 0; i < 7; i++) {
+    ASSERT_OK(Put(key_list[i], std::to_string(i)));
+  }
+
+  dbfull()->TEST_FlushMemTable();
+
+  Iterator* iter = dbfull()->NewIterator(ReadOptions());
+  iter->Seek(key_list[0]);
+
+  for (size_t i = 0; i < 7; i++) {
+    ASSERT_TRUE(iter->Valid());
+    ASSERT_EQ(key_list[i], iter->key().ToString());
+    ASSERT_EQ(std::to_string(i), iter->value().ToString());
+    iter->Next();
+  }
+
+  ASSERT_TRUE(!iter->Valid());
+
+  delete iter;
+}
+
+// A test comparator which compare two strings in this way:
+// (1) first compare prefix of 8 bytes in alphabet order,
+// (2) if two strings share the same prefix, sort the other part of the string
+//     in the reverse alphabet order.
+class SimpleSuffixReverseComparator : public Comparator {
+ public:
+  SimpleSuffixReverseComparator() {}
+
+  virtual const char* Name() const { return "SimpleSuffixReverseComparator"; }
+
+  virtual int Compare(const Slice& a, const Slice& b) const {
+    Slice prefix_a = Slice(a.data(), 8);
+    Slice prefix_b = Slice(b.data(), 8);
+    int prefix_comp = prefix_a.compare(prefix_b);
+    if (prefix_comp != 0) {
+      return prefix_comp;
+    } else {
+      Slice suffix_a = Slice(a.data() + 8, a.size() - 8);
+      Slice suffix_b = Slice(b.data() + 8, b.size() - 8);
+      return -(suffix_a.compare(suffix_b));
+    }
+  }
+  virtual void FindShortestSeparator(std::string* start,
+                                     const Slice& limit) const {}
+
+  virtual void FindShortSuccessor(std::string* key) const {}
+};
+
+TEST(PlainTableDBTest, IteratorReverseSuffixComparator) {
+  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
+  SimpleSuffixReverseComparator comp;
+  options.comparator = &comp;
+  DestroyAndReopen(&options);
+
+  ASSERT_OK(Put("1000000000foo002", "v_2"));
+  ASSERT_OK(Put("0000000000000bar", "random"));
+  ASSERT_OK(Put("1000000000foo001", "v1"));
+  ASSERT_OK(Put("3000000000000bar", "bar_v"));
+  ASSERT_OK(Put("1000000000foo003", "v__3"));
+  ASSERT_OK(Put("1000000000foo004", "v__4"));
+  ASSERT_OK(Put("1000000000foo005", "v__5"));
+  ASSERT_OK(Put("1000000000foo007", "v__7"));
+  ASSERT_OK(Put("1000000000foo008", "v__8"));
+  dbfull()->TEST_FlushMemTable();
+  ASSERT_EQ("v1", Get("1000000000foo001"));
+  ASSERT_EQ("v__3", Get("1000000000foo003"));
+  Iterator* iter = dbfull()->NewIterator(ReadOptions());
+  iter->Seek("1000000000foo009");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo008", iter->key().ToString());
+  ASSERT_EQ("v__8", iter->value().ToString());
+
+  iter->Next();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo007", iter->key().ToString());
+  ASSERT_EQ("v__7", iter->value().ToString());
+
+  iter->Next();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo005", iter->key().ToString());
+  ASSERT_EQ("v__5", iter->value().ToString());
+
+  iter->Next();
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo004", iter->key().ToString());
+  ASSERT_EQ("v__4", iter->value().ToString());
+
+  iter->Seek("3000000000000bar");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("3000000000000bar", iter->key().ToString());
+  ASSERT_EQ("bar_v", iter->value().ToString());
+
+  iter->Seek("1000000000foo005");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo005", iter->key().ToString());
+  ASSERT_EQ("v__5", iter->value().ToString());
+
+  iter->Seek("1000000000foo006");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo005", iter->key().ToString());
+  ASSERT_EQ("v__5", iter->value().ToString());
+
+  iter->Seek("1000000000foo008");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("1000000000foo008", iter->key().ToString());
+  ASSERT_EQ("v__8", iter->value().ToString());
+
+  iter->Seek("1000000000foo000");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("3000000000000bar", iter->key().ToString());
+
+  delete iter;
+}
+
+TEST(PlainTableDBTest, HashBucketConflict) {
+  for (size_t huge_page_tlb_size = 0; huge_page_tlb_size <= 2 * 1024 * 1024;
+       huge_page_tlb_size += 2 * 1024 * 1024) {
+    for (unsigned char i = 1; i <= 3; i++) {
+      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
+      options.table_factory.reset(
+          NewTotalOrderPlainTableFactory(16, 0, 2 ^ i, huge_page_tlb_size));
+      DestroyAndReopen(&options);
+      ASSERT_OK(Put("5000000000000fo0", "v1"));
+      ASSERT_OK(Put("5000000000000fo1", "v2"));
+      ASSERT_OK(Put("5000000000000fo2", "v"));
+      ASSERT_OK(Put("2000000000000fo0", "v3"));
+      ASSERT_OK(Put("2000000000000fo1", "v4"));
+      ASSERT_OK(Put("2000000000000fo2", "v"));
+      ASSERT_OK(Put("2000000000000fo3", "v"));
+
+      dbfull()->TEST_FlushMemTable();
+
+      ASSERT_EQ("v1", Get("5000000000000fo0"));
+      ASSERT_EQ("v2", Get("5000000000000fo1"));
+      ASSERT_EQ("v3", Get("2000000000000fo0"));
+      ASSERT_EQ("v4", Get("2000000000000fo1"));
+
+      ASSERT_EQ("NOT_FOUND", Get("5000000000000bar"));
+      ASSERT_EQ("NOT_FOUND", Get("2000000000000bar"));
+      ASSERT_EQ("NOT_FOUND", Get("5000000000000fo8"));
+      ASSERT_EQ("NOT_FOUND", Get("2000000000000fo8"));
+
+      ReadOptions ro;
+      Iterator* iter = dbfull()->NewIterator(ro);
+
+      iter->Seek("5000000000000fo0");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo0", iter->key().ToString());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo1", iter->key().ToString());
+
+      iter->Seek("5000000000000fo1");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo1", iter->key().ToString());
+
+      iter->Seek("2000000000000fo0");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo0", iter->key().ToString());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo1", iter->key().ToString());
+
+      iter->Seek("2000000000000fo1");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo1", iter->key().ToString());
+
+      iter->Seek("2000000000000bar");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo0", iter->key().ToString());
+
+      iter->Seek("5000000000000bar");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo0", iter->key().ToString());
+
+      iter->Seek("2000000000000fo8");
+      ASSERT_TRUE(!iter->Valid() ||
+                  options.comparator->Compare(iter->key(), "20000001") > 0);
+
+      iter->Seek("5000000000000fo8");
+      ASSERT_TRUE(!iter->Valid());
+
+      iter->Seek("1000000000000fo2");
+      ASSERT_TRUE(!iter->Valid());
+
+      iter->Seek("3000000000000fo2");
+      ASSERT_TRUE(!iter->Valid());
+
+      iter->Seek("8000000000000fo2");
+      ASSERT_TRUE(!iter->Valid());
+
+      delete iter;
+    }
+  }
+}
+
+TEST(PlainTableDBTest, HashBucketConflictReverseSuffixComparator) {
+  for (size_t huge_page_tlb_size = 0; huge_page_tlb_size <= 2 * 1024 * 1024;
+       huge_page_tlb_size += 2 * 1024 * 1024) {
+    for (unsigned char i = 1; i <= 3; i++) {
+      Options options = CurrentOptions();
+      options.create_if_missing = true;
+      SimpleSuffixReverseComparator comp;
+      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(
+          NewTotalOrderPlainTableFactory(16, 0, 2 ^ i, huge_page_tlb_size));
+      DestroyAndReopen(&options);
+      ASSERT_OK(Put("5000000000000fo0", "v1"));
+      ASSERT_OK(Put("5000000000000fo1", "v2"));
+      ASSERT_OK(Put("5000000000000fo2", "v"));
+      ASSERT_OK(Put("2000000000000fo0", "v3"));
+      ASSERT_OK(Put("2000000000000fo1", "v4"));
+      ASSERT_OK(Put("2000000000000fo2", "v"));
+      ASSERT_OK(Put("2000000000000fo3", "v"));
+
+      dbfull()->TEST_FlushMemTable();
+
+      ASSERT_EQ("v1", Get("5000000000000fo0"));
+      ASSERT_EQ("v2", Get("5000000000000fo1"));
+      ASSERT_EQ("v3", Get("2000000000000fo0"));
+      ASSERT_EQ("v4", Get("2000000000000fo1"));
+
+      ASSERT_EQ("NOT_FOUND", Get("5000000000000bar"));
+      ASSERT_EQ("NOT_FOUND", Get("2000000000000bar"));
+      ASSERT_EQ("NOT_FOUND", Get("5000000000000fo8"));
+      ASSERT_EQ("NOT_FOUND", Get("2000000000000fo8"));
+
+      ReadOptions ro;
+      Iterator* iter = dbfull()->NewIterator(ro);
+
+      iter->Seek("5000000000000fo1");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo1", iter->key().ToString());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo0", iter->key().ToString());
+
+      iter->Seek("5000000000000fo1");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo1", iter->key().ToString());
+
+      iter->Seek("2000000000000fo1");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo1", iter->key().ToString());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo0", iter->key().ToString());
+
+      iter->Seek("2000000000000fo1");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo1", iter->key().ToString());
+
+      iter->Seek("2000000000000var");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("2000000000000fo3", iter->key().ToString());
+
+      iter->Seek("5000000000000var");
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_EQ("5000000000000fo2", iter->key().ToString());
+
+      std::string seek_key = "2000000000000bar";
+      iter->Seek(seek_key);
+      ASSERT_TRUE(!iter->Valid() ||
+                  options.prefix_extractor->Transform(iter->key()) !=
+                      options.prefix_extractor->Transform(seek_key));
+
+      iter->Seek("1000000000000fo2");
+      ASSERT_TRUE(!iter->Valid());
+
+      iter->Seek("3000000000000fo2");
+      ASSERT_TRUE(!iter->Valid());
+
+      iter->Seek("8000000000000fo2");
+      ASSERT_TRUE(!iter->Valid());
+
+      delete iter;
+    }
+  }
+}
+
+TEST(PlainTableDBTest, NonExistingKeyToNonEmptyBucket) {
+  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
+  options.table_factory.reset(NewTotalOrderPlainTableFactory(16, 0, 5));
+  DestroyAndReopen(&options);
+  ASSERT_OK(Put("5000000000000fo0", "v1"));
+  ASSERT_OK(Put("5000000000000fo1", "v2"));
+  ASSERT_OK(Put("5000000000000fo2", "v3"));
+
+  dbfull()->TEST_FlushMemTable();
+
+  ASSERT_EQ("v1", Get("5000000000000fo0"));
+  ASSERT_EQ("v2", Get("5000000000000fo1"));
+  ASSERT_EQ("v3", Get("5000000000000fo2"));
+
+  ASSERT_EQ("NOT_FOUND", Get("8000000000000bar"));
+  ASSERT_EQ("NOT_FOUND", Get("1000000000000bar"));
+
+  Iterator* iter = dbfull()->NewIterator(ReadOptions());
+
+  iter->Seek("5000000000000bar");
+  ASSERT_TRUE(iter->Valid());
+  ASSERT_EQ("5000000000000fo0", iter->key().ToString());
+
+  iter->Seek("5000000000000fo8");
+  ASSERT_TRUE(!iter->Valid());
+
+  iter->Seek("1000000000000fo2");
+  ASSERT_TRUE(!iter->Valid());
+
+  iter->Seek("8000000000000fo2");
+  ASSERT_TRUE(!iter->Valid());
+
+  delete iter;
+}
+
+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(PlainTableDBTest, 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();
+}
@@ -1,75 +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.
-//
-// Wrap an underlying iterator, but exclude any results not starting
-// with a given prefix.  Seeking to keys not beginning with the prefix
-// is invalid, and SeekToLast is not implemented (that would be
-// non-trivial), but otherwise this iterator will behave just like the
-// underlying iterator would if there happened to be no non-matching
-// keys in the dataset.
-
-#pragma once
-#include "rocksdb/iterator.h"
-#include "rocksdb/slice.h"
-#include "rocksdb/slice_transform.h"
-
-namespace rocksdb {
-
-class PrefixFilterIterator : public Iterator {
- private:
-  Iterator* iter_;
-  const Slice &prefix_;
-  const SliceTransform *prefix_extractor_;
-  Status status_;
-
- public:
-  PrefixFilterIterator(Iterator* iter,
-                       const Slice &prefix,
-                       const SliceTransform* prefix_extractor)
-                             : iter_(iter), prefix_(prefix),
-                               prefix_extractor_(prefix_extractor),
-                               status_(Status::OK()) {
-    if (prefix_extractor == nullptr) {
-      status_ = Status::InvalidArgument("A prefix filter may not be used "
-                                        "unless a function is also defined "
-                                        "for extracting prefixes");
-    } else if (!prefix_extractor_->InRange(prefix)) {
-      status_ = Status::InvalidArgument("Must provide a slice for prefix which"
-                                        "is a prefix for some key");
-    }
-  }
-  ~PrefixFilterIterator() {
-    delete iter_;
-  }
-  Slice key() const { return iter_->key(); }
-  Slice value() const { return iter_->value(); }
-  Status status() const {
-    if (!status_.ok()) {
-      return status_;
-    }
-    return iter_->status();
-  }
-  void Next() { iter_->Next(); }
-  void Prev() { iter_->Prev(); }
-  void Seek(const Slice& k) {
-    if (prefix_extractor_->Transform(k) == prefix_) {
-      iter_->Seek(k);
-    } else {
-      status_ = Status::InvalidArgument("Seek must begin with target prefix");
-    }
-  }
-  void SeekToFirst() {
-    Seek(prefix_);
-  }
-  void SeekToLast() {
-    status_ = Status::NotSupported("SeekToLast is incompatible with prefixes");
-  }
-  bool Valid() const {
-    return (status_.ok() && iter_->Valid() &&
-            prefix_extractor_->Transform(iter_->key()) == prefix_);
-  }
-};
-
-}  // namespace rocksdb
@@ -1,3 +1,8 @@
+//  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 <algorithm>
 #include <iostream>
 #include <vector>
@@ -6,21 +11,27 @@
 #include "rocksdb/comparator.h"
 #include "rocksdb/db.h"
 #include "rocksdb/perf_context.h"
+#include "rocksdb/slice_transform.h"
+#include "rocksdb/memtablerep.h"
 #include "util/histogram.h"
 #include "util/stop_watch.h"
 #include "util/testharness.h"

-DEFINE_bool(use_prefix_hash_memtable, true, "");
 DEFINE_bool(trigger_deadlock, false,
            "issue delete in range scan to trigger PrefixHashMap deadlock");
 DEFINE_uint64(bucket_count, 100000, "number of buckets");
 DEFINE_uint64(num_locks, 10001, "number of locks");
 DEFINE_bool(random_prefix, false, "randomize prefix");
-DEFINE_uint64(total_prefixes, 1000, "total number of prefixes");
-DEFINE_uint64(items_per_prefix, 10, "total number of values per prefix");
-DEFINE_int64(write_buffer_size, 1000000000, "");
-DEFINE_int64(max_write_buffer_number, 8, "");
-DEFINE_int64(min_write_buffer_number_to_merge, 7, "");
+DEFINE_uint64(total_prefixes, 100000, "total number of prefixes");
+DEFINE_uint64(items_per_prefix, 1, "total number of values per prefix");
+DEFINE_int64(write_buffer_size, 33554432, "");
+DEFINE_int64(max_write_buffer_number, 2, "");
+DEFINE_int64(min_write_buffer_number_to_merge, 1, "");
+DEFINE_int32(skiplist_height, 4, "");
+DEFINE_int32(memtable_prefix_bloom_bits, 10000000, "");
+DEFINE_int32(memtable_prefix_bloom_probes, 10, "");
+DEFINE_int32(memtable_prefix_bloom_huge_page_tlb_size, 2 * 1024 * 1024, "");
+DEFINE_int32(value_size, 40, "");

 // Path to the database on file system
 const std::string kDbName = rocksdb::test::TmpDir() + "/prefix_test";
@@ -93,6 +104,38 @@ class TestKeyComparator : public Comparator {

 };

+namespace {
+void PutKey(DB* db, WriteOptions write_options, uint64_t prefix,
+            uint64_t suffix, const Slice& value) {
+  TestKey test_key(prefix, suffix);
+  Slice key = TestKeyToSlice(test_key);
+  ASSERT_OK(db->Put(write_options, key, value));
+}
+
+void SeekIterator(Iterator* iter, uint64_t prefix, uint64_t suffix) {
+  TestKey test_key(prefix, suffix);
+  Slice key = TestKeyToSlice(test_key);
+  iter->Seek(key);
+}
+
+const std::string kNotFoundResult = "NOT_FOUND";
+
+std::string Get(DB* db, const ReadOptions& read_options, uint64_t prefix,
+                uint64_t suffix) {
+  TestKey test_key(prefix, suffix);
+  Slice key = TestKeyToSlice(test_key);
+
+  std::string result;
+  Status s = db->Get(read_options, key, &result);
+  if (s.IsNotFound()) {
+    result = kNotFoundResult;
+  } else if (!s.ok()) {
+    result = s.ToString();
+  }
+  return result;
+}
+}  // namespace
+
 class PrefixTest {
 public:
  std::shared_ptr<DB> OpenDb() {
@@ -104,218 +147,333 @@ class PrefixTest {
    options.min_write_buffer_number_to_merge =
      FLAGS_min_write_buffer_number_to_merge;

-    options.comparator = new TestKeyComparator();
-    if (FLAGS_use_prefix_hash_memtable) {
-      auto prefix_extractor = NewFixedPrefixTransform(8);
-      options.prefix_extractor = prefix_extractor;
-      options.memtable_factory.reset(NewHashSkipListRepFactory(
-          prefix_extractor, FLAGS_bucket_count));
-    }
+    options.memtable_prefix_bloom_bits = FLAGS_memtable_prefix_bloom_bits;
+    options.memtable_prefix_bloom_probes = FLAGS_memtable_prefix_bloom_probes;
+    options.memtable_prefix_bloom_huge_page_tlb_size =
+        FLAGS_memtable_prefix_bloom_huge_page_tlb_size;

    Status s = DB::Open(options, kDbName,  &db);
    ASSERT_OK(s);
    return std::shared_ptr<DB>(db);
  }
+
+  void FirstOption() {
+    option_config_ = kBegin;
+  }
+
+  bool NextOptions(int bucket_count) {
+    // skip some options
+    option_config_++;
+    if (option_config_ < kEnd) {
+      options.prefix_extractor.reset(NewFixedPrefixTransform(8));
+      switch(option_config_) {
+        case kHashSkipList:
+          options.memtable_factory.reset(
+              NewHashSkipListRepFactory(bucket_count, FLAGS_skiplist_height));
+          return true;
+        case kHashLinkList:
+          options.memtable_factory.reset(
+              NewHashLinkListRepFactory(bucket_count));
+          return true;
+        case kHashLinkListHugePageTlb:
+          options.memtable_factory.reset(
+              NewHashLinkListRepFactory(bucket_count, 2 * 1024 * 1024));
+          return true;
+        default:
+          return false;
+      }
+    }
+    return false;
+  }
+
+  PrefixTest() : option_config_(kBegin) {
+    options.comparator = new TestKeyComparator();
+  }
  ~PrefixTest() {
    delete options.comparator;
  }
 protected:
+  enum OptionConfig {
+    kBegin,
+    kHashSkipList,
+    kHashLinkList,
+    kHashLinkListHugePageTlb,
+    kEnd
+  };
+  int option_config_;
  Options options;
 };

-TEST(PrefixTest, DynamicPrefixIterator) {
+TEST(PrefixTest, TestResult) {
+  for (int num_buckets = 1; num_buckets <= 2; num_buckets++) {
+    FirstOption();
+    while (NextOptions(num_buckets)) {
+      std::cout << "*** Mem table: " << options.memtable_factory->Name()
+                << " number of buckets: " << num_buckets
+                << std::endl;
+      DestroyDB(kDbName, Options());
+      auto db = OpenDb();
+      WriteOptions write_options;
+      ReadOptions read_options;

-  DestroyDB(kDbName, Options());
-  auto db = OpenDb();
-  WriteOptions write_options;
-  ReadOptions read_options;
+      // 1. Insert one row.
+      Slice v16("v16");
+      PutKey(db.get(), write_options, 1, 6, v16);
+      std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
+      SeekIterator(iter.get(), 1, 6);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v16 == iter->value());
+      SeekIterator(iter.get(), 1, 5);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v16 == iter->value());
+      SeekIterator(iter.get(), 1, 5);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v16 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(!iter->Valid());

-  std::vector<uint64_t> prefixes;
-  for (uint64_t i = 0; i < FLAGS_total_prefixes; ++i) {
-    prefixes.push_back(i);
-  }
+      SeekIterator(iter.get(), 2, 0);
+      ASSERT_TRUE(!iter->Valid());

-  if (FLAGS_random_prefix) {
-    std::random_shuffle(prefixes.begin(), prefixes.end());
-  }
+      ASSERT_EQ(v16.ToString(), Get(db.get(), read_options, 1, 6));
+      ASSERT_EQ(kNotFoundResult, Get(db.get(), read_options, 1, 5));
+      ASSERT_EQ(kNotFoundResult, Get(db.get(), read_options, 1, 7));
+      ASSERT_EQ(kNotFoundResult, Get(db.get(), read_options, 0, 6));
+      ASSERT_EQ(kNotFoundResult, Get(db.get(), read_options, 2, 6));

-  // insert x random prefix, each with y continuous element.
-  for (auto prefix : prefixes) {
-     for (uint64_t sorted = 0; sorted < FLAGS_items_per_prefix; sorted++) {
-      TestKey test_key(prefix, sorted);
+      // 2. Insert an entry for the same prefix as the last entry in the bucket.
+      Slice v17("v17");
+      PutKey(db.get(), write_options, 1, 7, v17);
+      iter.reset(db->NewIterator(read_options));
+      SeekIterator(iter.get(), 1, 7);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());

-      Slice key = TestKeyToSlice(test_key);
-      std::string value = "v" + std::to_string(sorted);
+      SeekIterator(iter.get(), 1, 6);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v16 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(!iter->Valid());

-      ASSERT_OK(db->Put(write_options, key, value));
+      SeekIterator(iter.get(), 2, 0);
+      ASSERT_TRUE(!iter->Valid());
+
+      // 3. Insert an entry for the same prefix as the head of the bucket.
+      Slice v15("v15");
+      PutKey(db.get(), write_options, 1, 5, v15);
+      iter.reset(db->NewIterator(read_options));
+
+      SeekIterator(iter.get(), 1, 7);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+
+      SeekIterator(iter.get(), 1, 5);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v15 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v16 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+
+      SeekIterator(iter.get(), 1, 5);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v15 == iter->value());
+
+      ASSERT_EQ(v15.ToString(), Get(db.get(), read_options, 1, 5));
+      ASSERT_EQ(v16.ToString(), Get(db.get(), read_options, 1, 6));
+      ASSERT_EQ(v17.ToString(), Get(db.get(), read_options, 1, 7));
+
+      // 4. Insert an entry with a larger prefix
+      Slice v22("v22");
+      PutKey(db.get(), write_options, 2, 2, v22);
+      iter.reset(db->NewIterator(read_options));
+
+      SeekIterator(iter.get(), 2, 2);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v22 == iter->value());
+      SeekIterator(iter.get(), 2, 0);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v22 == iter->value());
+
+      SeekIterator(iter.get(), 1, 5);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v15 == iter->value());
+
+      SeekIterator(iter.get(), 1, 7);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+
+      // 5. Insert an entry with a smaller prefix
+      Slice v02("v02");
+      PutKey(db.get(), write_options, 0, 2, v02);
+      iter.reset(db->NewIterator(read_options));
+
+      SeekIterator(iter.get(), 0, 2);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v02 == iter->value());
+      SeekIterator(iter.get(), 0, 0);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v02 == iter->value());
+
+      SeekIterator(iter.get(), 2, 0);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v22 == iter->value());
+
+      SeekIterator(iter.get(), 1, 5);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v15 == iter->value());
+
+      SeekIterator(iter.get(), 1, 7);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+
+      // 6. Insert to the beginning and the end of the first prefix
+      Slice v13("v13");
+      Slice v18("v18");
+      PutKey(db.get(), write_options, 1, 3, v13);
+      PutKey(db.get(), write_options, 1, 8, v18);
+      iter.reset(db->NewIterator(read_options));
+      SeekIterator(iter.get(), 1, 7);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+
+      SeekIterator(iter.get(), 1, 3);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v13 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v15 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v16 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v17 == iter->value());
+      iter->Next();
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v18 == iter->value());
+
+      SeekIterator(iter.get(), 0, 0);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v02 == iter->value());
+
+      SeekIterator(iter.get(), 2, 0);
+      ASSERT_TRUE(iter->Valid());
+      ASSERT_TRUE(v22 == iter->value());
+
+      ASSERT_EQ(v22.ToString(), Get(db.get(), read_options, 2, 2));
+      ASSERT_EQ(v02.ToString(), Get(db.get(), read_options, 0, 2));
+      ASSERT_EQ(v13.ToString(), Get(db.get(), read_options, 1, 3));
+      ASSERT_EQ(v15.ToString(), Get(db.get(), read_options, 1, 5));
+      ASSERT_EQ(v16.ToString(), Get(db.get(), read_options, 1, 6));
+      ASSERT_EQ(v17.ToString(), Get(db.get(), read_options, 1, 7));
+      ASSERT_EQ(v18.ToString(), Get(db.get(), read_options, 1, 8));
    }
  }
-
-  // test seek existing keys
-  HistogramImpl hist_seek_time;
-  HistogramImpl hist_seek_comparison;
-
-  if (FLAGS_use_prefix_hash_memtable) {
-    read_options.prefix_seek = true;
-  }
-  std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
-
-  for (auto prefix : prefixes) {
-    TestKey test_key(prefix, FLAGS_items_per_prefix / 2);
-    Slice key = TestKeyToSlice(test_key);
-    std::string value = "v" + std::to_string(0);
-
-    perf_context.Reset();
-    StopWatchNano timer(Env::Default(), true);
-    uint64_t total_keys = 0;
-    for (iter->Seek(key); iter->Valid(); iter->Next()) {
-      if (FLAGS_trigger_deadlock) {
-        std::cout << "Behold the deadlock!\n";
-        db->Delete(write_options, iter->key());
-      }
-      auto test_key = SliceToTestKey(iter->key());
-      if (test_key->prefix != prefix) break;
-      total_keys++;
-    }
-    hist_seek_time.Add(timer.ElapsedNanos());
-    hist_seek_comparison.Add(perf_context.user_key_comparison_count);
-    ASSERT_EQ(total_keys, FLAGS_items_per_prefix - FLAGS_items_per_prefix/2);
-  }
-
-  std::cout << "Seek key comparison: \n"
-            << hist_seek_comparison.ToString()
-            << "Seek time: \n"
-            << hist_seek_time.ToString();
-
-  // test non-existing keys
-  HistogramImpl hist_no_seek_time;
-  HistogramImpl hist_no_seek_comparison;
-
-  for (auto prefix = FLAGS_total_prefixes;
-       prefix < FLAGS_total_prefixes + 100;
-       prefix++) {
-    TestKey test_key(prefix, 0);
-    Slice key = TestKeyToSlice(test_key);
-
-    perf_context.Reset();
-    StopWatchNano timer(Env::Default(), true);
-    iter->Seek(key);
-    hist_no_seek_time.Add(timer.ElapsedNanos());
-    hist_no_seek_comparison.Add(perf_context.user_key_comparison_count);
-    ASSERT_TRUE(!iter->Valid());
-  }
-
-  std::cout << "non-existing Seek key comparison: \n"
-            << hist_no_seek_comparison.ToString()
-            << "non-existing Seek time: \n"
-            << hist_no_seek_time.ToString();
 }

-TEST(PrefixTest, PrefixHash) {
+TEST(PrefixTest, DynamicPrefixIterator) {
+  while (NextOptions(FLAGS_bucket_count)) {
+    std::cout << "*** Mem table: " << options.memtable_factory->Name()
+        << std::endl;
+    DestroyDB(kDbName, Options());
+    auto db = OpenDb();
+    WriteOptions write_options;
+    ReadOptions read_options;

-  DestroyDB(kDbName, Options());
-  auto db = OpenDb();
-  WriteOptions write_options;
-  ReadOptions read_options;
+    std::vector<uint64_t> prefixes;
+    for (uint64_t i = 0; i < FLAGS_total_prefixes; ++i) {
+      prefixes.push_back(i);
+    }

-  std::vector<uint64_t> prefixes;
-  for (uint64_t i = 0; i < FLAGS_total_prefixes; ++i) {
-    prefixes.push_back(i);
-  }
+    if (FLAGS_random_prefix) {
+      std::random_shuffle(prefixes.begin(), prefixes.end());
+    }

-  if (FLAGS_random_prefix) {
-    std::random_shuffle(prefixes.begin(), prefixes.end());
-  }
+    HistogramImpl hist_put_time;
+    HistogramImpl hist_put_comparison;

-  // insert x random prefix, each with y continuous element.
-  HistogramImpl hist_put_time;
-  HistogramImpl hist_put_comparison;
+    // insert x random prefix, each with y continuous element.
+    for (auto prefix : prefixes) {
+       for (uint64_t sorted = 0; sorted < FLAGS_items_per_prefix; sorted++) {
+        TestKey test_key(prefix, sorted);

-  for (auto prefix : prefixes) {
-     for (uint64_t sorted = 0; sorted < FLAGS_items_per_prefix; sorted++) {
-      TestKey test_key(prefix, sorted);
+        Slice key = TestKeyToSlice(test_key);
+        std::string value(FLAGS_value_size, 0);

+        perf_context.Reset();
+        StopWatchNano timer(Env::Default(), true);
+        ASSERT_OK(db->Put(write_options, key, value));
+        hist_put_time.Add(timer.ElapsedNanos());
+        hist_put_comparison.Add(perf_context.user_key_comparison_count);
+      }
+    }
+
+    std::cout << "Put key comparison: \n" << hist_put_comparison.ToString()
+              << "Put time: \n" << hist_put_time.ToString();
+
+    // test seek existing keys
+    HistogramImpl hist_seek_time;
+    HistogramImpl hist_seek_comparison;
+
+    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
+
+    for (auto prefix : prefixes) {
+      TestKey test_key(prefix, FLAGS_items_per_prefix / 2);
      Slice key = TestKeyToSlice(test_key);
-      std::string value = "v" + std::to_string(sorted);
+      std::string value = "v" + std::to_string(0);

      perf_context.Reset();
      StopWatchNano timer(Env::Default(), true);
-      ASSERT_OK(db->Put(write_options, key, value));
-      hist_put_time.Add(timer.ElapsedNanos());
-      hist_put_comparison.Add(perf_context.user_key_comparison_count);
-    }
-  }
-
-  std::cout << "Put key comparison: \n" << hist_put_comparison.ToString()
-            << "Put time: \n" << hist_put_time.ToString();
-
-
-  // test seek existing keys
-  HistogramImpl hist_seek_time;
-  HistogramImpl hist_seek_comparison;
-
-  for (auto prefix : prefixes) {
-    TestKey test_key(prefix, 0);
-    Slice key = TestKeyToSlice(test_key);
-    std::string value = "v" + std::to_string(0);
-
-    Slice key_prefix;
-    if (FLAGS_use_prefix_hash_memtable) {
-      key_prefix = options.prefix_extractor->Transform(key);
-      read_options.prefix = &key_prefix;
-    }
-    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
-
-    perf_context.Reset();
-    StopWatchNano timer(Env::Default(), true);
-    uint64_t total_keys = 0;
-    for (iter->Seek(key); iter->Valid(); iter->Next()) {
-      if (FLAGS_trigger_deadlock) {
-        std::cout << "Behold the deadlock!\n";
-        db->Delete(write_options, iter->key());
+      auto key_prefix = options.prefix_extractor->Transform(key);
+      uint64_t total_keys = 0;
+      for (iter->Seek(key);
+           iter->Valid() && iter->key().starts_with(key_prefix);
+           iter->Next()) {
+        if (FLAGS_trigger_deadlock) {
+          std::cout << "Behold the deadlock!\n";
+          db->Delete(write_options, iter->key());
+        }
+        total_keys++;
      }
-      auto test_key = SliceToTestKey(iter->key());
-      if (test_key->prefix != prefix) break;
-      total_keys++;
+      hist_seek_time.Add(timer.ElapsedNanos());
+      hist_seek_comparison.Add(perf_context.user_key_comparison_count);
+      ASSERT_EQ(total_keys, FLAGS_items_per_prefix - FLAGS_items_per_prefix/2);
    }
-    hist_seek_time.Add(timer.ElapsedNanos());
-    hist_seek_comparison.Add(perf_context.user_key_comparison_count);
-    ASSERT_EQ(total_keys, FLAGS_items_per_prefix);
-  }

-  std::cout << "Seek key comparison: \n"
-            << hist_seek_comparison.ToString()
-            << "Seek time: \n"
-            << hist_seek_time.ToString();
+    std::cout << "Seek key comparison: \n"
+              << hist_seek_comparison.ToString()
+              << "Seek time: \n"
+              << hist_seek_time.ToString();

-  // test non-existing keys
-  HistogramImpl hist_no_seek_time;
-  HistogramImpl hist_no_seek_comparison;
+    // test non-existing keys
+    HistogramImpl hist_no_seek_time;
+    HistogramImpl hist_no_seek_comparison;

-  for (auto prefix = FLAGS_total_prefixes;
-       prefix < FLAGS_total_prefixes + 100;
-       prefix++) {
-    TestKey test_key(prefix, 0);
-    Slice key = TestKeyToSlice(test_key);
+    for (auto prefix = FLAGS_total_prefixes;
+         prefix < FLAGS_total_prefixes + 10000;
+         prefix++) {
+      TestKey test_key(prefix, 0);
+      Slice key = TestKeyToSlice(test_key);

-    if (FLAGS_use_prefix_hash_memtable) {
-      Slice key_prefix = options.prefix_extractor->Transform(key);
-      read_options.prefix = &key_prefix;
+      perf_context.Reset();
+      StopWatchNano timer(Env::Default(), true);
+      iter->Seek(key);
+      hist_no_seek_time.Add(timer.ElapsedNanos());
+      hist_no_seek_comparison.Add(perf_context.user_key_comparison_count);
+      ASSERT_TRUE(!iter->Valid());
    }
-    std::unique_ptr<Iterator> iter(db->NewIterator(read_options));

-    perf_context.Reset();
-    StopWatchNano timer(Env::Default(), true);
-    iter->Seek(key);
-    hist_no_seek_time.Add(timer.ElapsedNanos());
-    hist_no_seek_comparison.Add(perf_context.user_key_comparison_count);
-    ASSERT_TRUE(!iter->Valid());
+    std::cout << "non-existing Seek key comparison: \n"
+              << hist_no_seek_comparison.ToString()
+              << "non-existing Seek time: \n"
+              << hist_no_seek_time.ToString();
  }
-
-  std::cout << "non-existing Seek key comparison: \n"
-            << hist_no_seek_comparison.ToString()
-            << "non-existing Seek time: \n"
-            << hist_no_seek_time.ToString();
 }

 }
@@ -29,6 +29,8 @@
 //   Store per-table metadata (smallest, largest, largest-seq#, ...)
 //   in the table's meta section to speed up ScanTable.

+#ifndef ROCKSDB_LITE
+
 #include "db/builder.h"
 #include "db/db_impl.h"
 #include "db/dbformat.h"
@@ -55,14 +57,20 @@ class Repairer {
        icmp_(options.comparator),
        ipolicy_(options.filter_policy),
        options_(SanitizeOptions(dbname, &icmp_, &ipolicy_, options)),
+        raw_table_cache_(
+            // TableCache can be small since we expect each table to be opened
+            // once.
+            NewLRUCache(10, options_.table_cache_numshardbits,
+                        options_.table_cache_remove_scan_count_limit)),
        next_file_number_(1) {
-    // TableCache can be small since we expect each table to be opened once.
-    table_cache_ = new TableCache(dbname_, &options_, storage_options_, 10);
+    table_cache_ = new TableCache(dbname_, &options_, storage_options_,
+                                  raw_table_cache_.get());
    edit_ = new VersionEdit();
  }

  ~Repairer() {
    delete table_cache_;
+    raw_table_cache_.reset();
    delete edit_;
  }

@@ -102,6 +110,7 @@ class Repairer {
  InternalKeyComparator const icmp_;
  InternalFilterPolicy const ipolicy_;
  Options const options_;
+  std::shared_ptr<Cache> raw_table_cache_;
  TableCache* table_cache_;
  VersionEdit* edit_;

@@ -119,7 +128,7 @@ class Repairer {
      return status;
    }
    if (filenames.empty()) {
-      return Status::IOError(dbname_, "repair found no files");
+      return Status::Corruption(dbname_, "repair found no files");
    }

    uint64_t number;
@@ -197,6 +206,7 @@ class Repairer {
    Slice record;
    WriteBatch batch;
    MemTable* mem = new MemTable(icmp_, options_);
+    auto cf_mems_default = new ColumnFamilyMemTablesDefault(mem, &options_);
    mem->Ref();
    int counter = 0;
    while (reader.ReadRecord(&record, &scratch)) {
@@ -206,7 +216,7 @@ class Repairer {
        continue;
      }
      WriteBatchInternal::SetContents(&batch, record);
-      status = WriteBatchInternal::InsertInto(&batch, mem, &options_);
+      status = WriteBatchInternal::InsertInto(&batch, cf_mems_default);
      if (status.ok()) {
        counter += WriteBatchInternal::Count(&batch);
      } else {
@@ -221,13 +231,13 @@ class Repairer {
    // since ExtractMetaData() will also generate edits.
    FileMetaData meta;
    meta.number = next_file_number_++;
-    Iterator* iter = mem->NewIterator();
-    status = BuildTable(dbname_, env_, options_, storage_options_,
-                        table_cache_, iter, &meta,
-                        icmp_.user_comparator(), 0, 0,
-                        kNoCompression);
+    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;
    delete mem->Unref();
+    delete cf_mems_default;
    mem = nullptr;
    if (status.ok()) {
      if (meta.file_size > 0) {
@@ -243,7 +253,6 @@ class Repairer {
  }

  void ExtractMetaData() {
-    std::vector<TableInfo> kept;
    for (size_t i = 0; i < table_numbers_.size(); i++) {
      TableInfo t;
      t.meta.number = table_numbers_[i];
@@ -265,8 +274,9 @@ class Repairer {
    int counter = 0;
    Status status = env_->GetFileSize(fname, &t->meta.file_size);
    if (status.ok()) {
+      FileMetaData dummy_meta(t->meta.number, t->meta.file_size);
      Iterator* iter = table_cache_->NewIterator(
-          ReadOptions(), storage_options_, t->meta.number, t->meta.file_size);
+          ReadOptions(), storage_options_, icmp_, dummy_meta);
      bool empty = true;
      ParsedInternalKey parsed;
      t->min_sequence = 0;
@@ -308,7 +318,8 @@ class Repairer {
  Status WriteDescriptor() {
    std::string tmp = TempFileName(dbname_, 1);
    unique_ptr<WritableFile> file;
-    Status status = env_->NewWritableFile(tmp, &file, storage_options_);
+    Status status = env_->NewWritableFile(
+        tmp, &file, env_->OptimizeForManifestWrite(storage_options_));
    if (!status.ok()) {
      return status;
    }
@@ -388,3 +399,5 @@ Status RepairDB(const std::string& dbname, const Options& options) {
 }

 }  // namespace rocksdb
+
+#endif  // ROCKSDB_LITE
@@ -22,6 +22,8 @@
 #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"
@@ -31,6 +33,7 @@

 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
@@ -80,22 +83,18 @@ public:
                     unique_ptr<RandomAccessFile> && file, uint64_t file_size,
                     unique_ptr<TableReader>* table_reader);

-  bool PrefixMayMatch(const Slice& internal_prefix) override;
-
  Iterator* NewIterator(const ReadOptions&) override;

-  Status Get(
-      const ReadOptions&, const Slice& key, void* arg,
-      bool (*handle_result)(void* arg, const Slice& k, const Slice& v, bool),
-      void (*mark_key_may_exist)(void*) = nullptr) override;
+  Status Get(const ReadOptions&, const Slice& key, void* arg,
+             bool (*handle_result)(void* arg, const ParsedInternalKey& k,
+                                   const Slice& v, bool),
+             void (*mark_key_may_exist)(void*) = nullptr) override;

  uint64_t ApproximateOffsetOf(const Slice& key) override;

-  bool TEST_KeyInCache(const ReadOptions& options, const Slice& key) override;
-
  void SetupForCompaction() override;

-  TableProperties& GetTableProperties() override;
+  std::shared_ptr<const TableProperties> GetTableProperties() const override;

  ~SimpleTableReader();

@@ -171,7 +170,7 @@ struct SimpleTableReader::Rep {
  unique_ptr<RandomAccessFile> file;
  uint64_t index_start_offset;
  int num_entries;
-  TableProperties table_properties;
+  std::shared_ptr<TableProperties> table_properties;

  const static int user_key_size = 16;
  const static int offset_length = 8;
@@ -214,14 +213,11 @@ Status SimpleTableReader::Open(const Options& options,
 void SimpleTableReader::SetupForCompaction() {
 }

-TableProperties& SimpleTableReader::GetTableProperties() {
+std::shared_ptr<const TableProperties> SimpleTableReader::GetTableProperties()
+    const {
  return rep_->table_properties;
 }

-bool SimpleTableReader::PrefixMayMatch(const Slice& internal_prefix) {
-  return true;
-}
-
 Iterator* SimpleTableReader::NewIterator(const ReadOptions& options) {
  return new SimpleTableIterator(this);
 }
@@ -244,7 +240,8 @@ Status SimpleTableReader::GetOffset(const Slice& target, uint64_t* offset) {
      return s;
    }

-    int compare_result = rep_->options.comparator->Compare(tmp_slice, target);
+    InternalKeyComparator ikc(rep_->options.comparator);
+    int compare_result = ikc.Compare(tmp_slice, target);

    if (compare_result < 0) {
      if (left == right) {
@@ -279,14 +276,20 @@ Status SimpleTableReader::GetOffset(const Slice& target, uint64_t* offset) {
  return s;
 }

-Status SimpleTableReader::Get(
-    const ReadOptions& options, const Slice& k, void* arg,
-    bool (*saver)(void*, const Slice&, const Slice&, bool),
-    void (*mark_key_may_exist)(void*)) {
+Status SimpleTableReader::Get(const ReadOptions& options, const Slice& k,
+                              void* arg,
+                              bool (*saver)(void*, const ParsedInternalKey&,
+                                            const Slice&, bool),
+                              void (*mark_key_may_exist)(void*)) {
  Status s;
  SimpleTableIterator* iter = new SimpleTableIterator(this);
  for (iter->Seek(k); iter->Valid(); iter->Next()) {
-    if (!(*saver)(arg, iter->key(), iter->value(), true)) {
+    ParsedInternalKey parsed_key;
+    if (!ParseInternalKey(iter->key(), &parsed_key)) {
+      return Status::Corruption(Slice());
+    }
+
+    if (!(*saver)(arg, parsed_key, iter->value(), true)) {
      break;
    }
  }
@@ -295,11 +298,6 @@ Status SimpleTableReader::Get(
  return s;
 }

-bool SimpleTableReader::TEST_KeyInCache(const ReadOptions& options,
-                                        const Slice& key) {
-  return false;
-}
-
 uint64_t SimpleTableReader::ApproximateOffsetOf(const Slice& key) {
  return 0;
 }
@@ -540,27 +538,30 @@ public:
  const char* Name() const override {
    return "SimpleTable";
  }
-  Status GetTableReader(const Options& options, const EnvOptions& soptions,
-                        unique_ptr<RandomAccessFile> && file,
-                        uint64_t file_size,
+  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* GetTableBuilder(const Options& options, WritableFile* file,
+  TableBuilder* NewTableBuilder(const Options& options,
+                                const InternalKeyComparator& internal_key,
+                                WritableFile* file,
                                CompressionType compression_type) const;
 };

-Status SimpleTableFactory::GetTableReader(
+Status SimpleTableFactory::NewTableReader(
    const Options& options, const EnvOptions& soptions,
-    unique_ptr<RandomAccessFile> && file, uint64_t file_size,
+    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::GetTableBuilder(
-    const Options& options, WritableFile* file,
-    CompressionType compression_type) const {
+TableBuilder* SimpleTableFactory::NewTableBuilder(
+    const Options& options, const InternalKeyComparator& internal_key,
+    WritableFile* file, CompressionType compression_type) const {
  return new SimpleTableBuilder(options, file, compression_type);
 }

@@ -33,9 +33,10 @@
 #pragma once
 #include <assert.h>
 #include <stdlib.h>
+#include "util/arena.h"
 #include "port/port.h"
+#include "util/arena.h"
 #include "util/random.h"
-#include "rocksdb/arena.h"

 namespace rocksdb {

@@ -48,7 +49,8 @@ class SkipList {
  // Create a new SkipList object that will use "cmp" for comparing keys,
  // and will allocate memory using "*arena".  Objects allocated in the arena
  // must remain allocated for the lifetime of the skiplist object.
-  explicit SkipList(Comparator cmp, Arena* arena);
+  explicit SkipList(Comparator cmp, Arena* arena,
+                    int32_t max_height = 12, int32_t branching_factor = 4);

  // Insert key into the list.
  // REQUIRES: nothing that compares equal to key is currently in the list.
@@ -102,7 +104,8 @@ class SkipList {
  };

 private:
-  enum { kMaxHeight = 12 };
+  const int32_t kMaxHeight_;
+  const int32_t kBranching_;

  // Immutable after construction
  Comparator const compare_;
@@ -115,8 +118,8 @@ class SkipList {
  port::AtomicPointer max_height_;   // Height of the entire list

  // Used for optimizing sequential insert patterns
-  Node* prev_[kMaxHeight];
-  int   prev_height_;
+  Node** prev_;
+  int32_t prev_height_;

  inline int GetMaxHeight() const {
    return static_cast<int>(
@@ -155,7 +158,7 @@ class SkipList {

 // Implementation details follow
 template<typename Key, class Comparator>
-struct SkipList<Key,Comparator>::Node {
+struct SkipList<Key, Comparator>::Node {
  explicit Node(const Key& k) : key(k) { }

  Key const key;
@@ -191,43 +194,43 @@ struct SkipList<Key,Comparator>::Node {
 };

 template<typename Key, class Comparator>
-typename SkipList<Key,Comparator>::Node*
-SkipList<Key,Comparator>::NewNode(const Key& key, int height) {
+typename SkipList<Key, Comparator>::Node*
+SkipList<Key, Comparator>::NewNode(const Key& key, int height) {
  char* mem = arena_->AllocateAligned(
      sizeof(Node) + sizeof(port::AtomicPointer) * (height - 1));
  return new (mem) Node(key);
 }

 template<typename Key, class Comparator>
-inline SkipList<Key,Comparator>::Iterator::Iterator(const SkipList* list) {
+inline SkipList<Key, Comparator>::Iterator::Iterator(const SkipList* list) {
  SetList(list);
 }

 template<typename Key, class Comparator>
-inline void SkipList<Key,Comparator>::Iterator::SetList(const SkipList* list) {
+inline void SkipList<Key, Comparator>::Iterator::SetList(const SkipList* list) {
  list_ = list;
  node_ = nullptr;
 }

 template<typename Key, class Comparator>
-inline bool SkipList<Key,Comparator>::Iterator::Valid() const {
+inline bool SkipList<Key, Comparator>::Iterator::Valid() const {
  return node_ != nullptr;
 }

 template<typename Key, class Comparator>
-inline const Key& SkipList<Key,Comparator>::Iterator::key() const {
+inline const Key& SkipList<Key, Comparator>::Iterator::key() const {
  assert(Valid());
  return node_->key;
 }

 template<typename Key, class Comparator>
-inline void SkipList<Key,Comparator>::Iterator::Next() {
+inline void SkipList<Key, Comparator>::Iterator::Next() {
  assert(Valid());
  node_ = node_->Next(0);
 }

 template<typename Key, class Comparator>
-inline void SkipList<Key,Comparator>::Iterator::Prev() {
+inline void SkipList<Key, Comparator>::Iterator::Prev() {
  // Instead of using explicit "prev" links, we just search for the
  // last node that falls before key.
  assert(Valid());
@@ -238,17 +241,17 @@ inline void SkipList<Key,Comparator>::Iterator::Prev() {
 }

 template<typename Key, class Comparator>
-inline void SkipList<Key,Comparator>::Iterator::Seek(const Key& target) {
+inline void SkipList<Key, Comparator>::Iterator::Seek(const Key& target) {
  node_ = list_->FindGreaterOrEqual(target, nullptr);
 }

 template<typename Key, class Comparator>
-inline void SkipList<Key,Comparator>::Iterator::SeekToFirst() {
+inline void SkipList<Key, Comparator>::Iterator::SeekToFirst() {
  node_ = list_->head_->Next(0);
 }

 template<typename Key, class Comparator>
-inline void SkipList<Key,Comparator>::Iterator::SeekToLast() {
+inline void SkipList<Key, Comparator>::Iterator::SeekToLast() {
  node_ = list_->FindLast();
  if (node_ == list_->head_) {
    node_ = nullptr;
@@ -256,27 +259,26 @@ inline void SkipList<Key,Comparator>::Iterator::SeekToLast() {
 }

 template<typename Key, class Comparator>
-int SkipList<Key,Comparator>::RandomHeight() {
+int SkipList<Key, Comparator>::RandomHeight() {
  // Increase height with probability 1 in kBranching
-  static const unsigned int kBranching = 4;
  int height = 1;
-  while (height < kMaxHeight && ((rnd_.Next() % kBranching) == 0)) {
+  while (height < kMaxHeight_ && ((rnd_.Next() % kBranching_) == 0)) {
    height++;
  }
  assert(height > 0);
-  assert(height <= kMaxHeight);
+  assert(height <= kMaxHeight_);
  return height;
 }

 template<typename Key, class Comparator>
-bool SkipList<Key,Comparator>::KeyIsAfterNode(const Key& key, Node* n) const {
+bool SkipList<Key, Comparator>::KeyIsAfterNode(const Key& key, Node* n) const {
  // nullptr n is considered infinite
  return (n != nullptr) && (compare_(n->key, key) < 0);
 }

 template<typename Key, class Comparator>
-typename SkipList<Key,Comparator>::Node* SkipList<Key,Comparator>::FindGreaterOrEqual(const Key& key, Node** prev)
-    const {
+typename SkipList<Key, Comparator>::Node* SkipList<Key, Comparator>::
+  FindGreaterOrEqual(const Key& key, Node** prev) const {
  // Use prev as an optimization hint and fallback to slow path
  if (prev && !KeyIsAfterNode(key, prev[0]->Next(0))) {
    Node* x = prev[0];
@@ -313,8 +315,8 @@ typename SkipList<Key,Comparator>::Node* SkipList<Key,Comparator>::FindGreaterOr
 }

 template<typename Key, class Comparator>
-typename SkipList<Key,Comparator>::Node*
-SkipList<Key,Comparator>::FindLessThan(const Key& key) const {
+typename SkipList<Key, Comparator>::Node*
+SkipList<Key, Comparator>::FindLessThan(const Key& key) const {
  Node* x = head_;
  int level = GetMaxHeight() - 1;
  while (true) {
@@ -334,7 +336,7 @@ SkipList<Key,Comparator>::FindLessThan(const Key& key) const {
 }

 template<typename Key, class Comparator>
-typename SkipList<Key,Comparator>::Node* SkipList<Key,Comparator>::FindLast()
+typename SkipList<Key, Comparator>::Node* SkipList<Key, Comparator>::FindLast()
    const {
  Node* x = head_;
  int level = GetMaxHeight() - 1;
@@ -354,21 +356,31 @@ typename SkipList<Key,Comparator>::Node* SkipList<Key,Comparator>::FindLast()
 }

 template<typename Key, class Comparator>
-SkipList<Key,Comparator>::SkipList(Comparator cmp, Arena* arena)
-    : compare_(cmp),
+SkipList<Key, Comparator>::SkipList(const Comparator cmp, Arena* arena,
+                                   int32_t max_height,
+                                   int32_t branching_factor)
+    : kMaxHeight_(max_height),
+      kBranching_(branching_factor),
+      compare_(cmp),
      arena_(arena),
-      head_(NewNode(0 /* any key will do */, kMaxHeight)),
+      head_(NewNode(0 /* any key will do */, max_height)),
      max_height_(reinterpret_cast<void*>(1)),
      prev_height_(1),
      rnd_(0xdeadbeef) {
-  for (int i = 0; i < kMaxHeight; i++) {
+  assert(kMaxHeight_ > 0);
+  assert(kBranching_ > 0);
+  // Allocate the prev_ Node* array, directly from the passed-in arena.
+  // prev_ does not need to be freed, as its life cycle is tied up with
+  // the arena as a whole.
+  prev_ = (Node**) arena_->AllocateAligned(sizeof(Node*) * kMaxHeight_);
+  for (int i = 0; i < kMaxHeight_; i++) {
    head_->SetNext(i, nullptr);
    prev_[i] = head_;
  }
 }

 template<typename Key, class Comparator>
-void SkipList<Key,Comparator>::Insert(const Key& key) {
+void SkipList<Key, Comparator>::Insert(const Key& key) {
  // TODO(opt): We can use a barrier-free variant of FindGreaterOrEqual()
  // here since Insert() is externally synchronized.
  Node* x = FindGreaterOrEqual(key, prev_);
@@ -405,7 +417,7 @@ void SkipList<Key,Comparator>::Insert(const Key& key) {
 }

 template<typename Key, class Comparator>
-bool SkipList<Key,Comparator>::Contains(const Key& key) const {
+bool SkipList<Key, Comparator>::Contains(const Key& key) const {
  Node* x = FindGreaterOrEqual(key, nullptr);
  if (x != nullptr && Equal(key, x->key)) {
    return true;
@@ -10,7 +10,7 @@
 #include "db/skiplist.h"
 #include <set>
 #include "rocksdb/env.h"
-#include "util/arena_impl.h"
+#include "util/arena.h"
 #include "util/hash.h"
 #include "util/random.h"
 #include "util/testharness.h"
@@ -34,9 +34,9 @@ struct TestComparator {
 class SkipTest { };

 TEST(SkipTest, Empty) {
-  ArenaImpl arena_impl;
+  Arena arena;
  TestComparator cmp;
-  SkipList<Key, TestComparator> list(cmp, &arena_impl);
+  SkipList<Key, TestComparator> list(cmp, &arena);
  ASSERT_TRUE(!list.Contains(10));

  SkipList<Key, TestComparator>::Iterator iter(&list);
@@ -54,9 +54,9 @@ TEST(SkipTest, InsertAndLookup) {
  const int R = 5000;
  Random rnd(1000);
  std::set<Key> keys;
-  ArenaImpl arena_impl;
+  Arena arena;
  TestComparator cmp;
-  SkipList<Key, TestComparator> list(cmp, &arena_impl);
+  SkipList<Key, TestComparator> list(cmp, &arena);
  for (int i = 0; i < N; i++) {
    Key key = rnd.Next() % R;
    if (keys.insert(key).second) {
@@ -209,14 +209,14 @@ class ConcurrentTest {
  // Current state of the test
  State current_;

-  ArenaImpl arena_impl_;
+  Arena arena_;

  // SkipList is not protected by mu_.  We just use a single writer
  // thread to modify it.
  SkipList<Key, TestComparator> list_;

 public:
-  ConcurrentTest() : list_(TestComparator(), &arena_impl_) { }
+  ConcurrentTest() : list_(TestComparator(), &arena_) {}

  // REQUIRES: External synchronization
  void WriteStep(Random* rnd) {
@@ -10,9 +10,10 @@
 #include "db/table_cache.h"

 #include "db/filename.h"
+#include "db/version_edit.h"

 #include "rocksdb/statistics.h"
-#include "rocksdb/table.h"
+#include "table/table_reader.h"
 #include "util/coding.h"
 #include "util/stop_watch.h"

@@ -34,23 +35,27 @@ static Slice GetSliceForFileNumber(uint64_t* file_number) {
               sizeof(*file_number));
 }

-TableCache::TableCache(const std::string& dbname,
-                       const Options* options,
-                       const EnvOptions& storage_options,
-                       int entries)
+TableCache::TableCache(const std::string& dbname, const Options* options,
+                       const EnvOptions& storage_options, Cache* const cache)
    : env_(options->env),
      dbname_(dbname),
      options_(options),
      storage_options_(storage_options),
-      cache_(
-        NewLRUCache(entries, options->table_cache_numshardbits,
-                    options->table_cache_remove_scan_count_limit)) {
-}
+      cache_(cache) {}

 TableCache::~TableCache() {
 }

+TableReader* TableCache::GetTableReaderFromHandle(Cache::Handle* handle) {
+  return reinterpret_cast<TableReader*>(cache_->Value(handle));
+}
+
+void TableCache::ReleaseHandle(Cache::Handle* handle) {
+  cache_->Release(handle);
+}
+
 Status TableCache::FindTable(const EnvOptions& toptions,
+                             const InternalKeyComparator& internal_comparator,
                             uint64_t file_number, uint64_t file_size,
                             Cache::Handle** handle, bool* table_io,
                             const bool no_io) {
@@ -74,9 +79,9 @@ Status TableCache::FindTable(const EnvOptions& toptions,
        file->Hint(RandomAccessFile::RANDOM);
      }
      StopWatch sw(env_, options_->statistics.get(), TABLE_OPEN_IO_MICROS);
-      s = options_->table_factory->GetTableReader(*options_, toptions,
-                                                  std::move(file), file_size,
-                                                  &table_reader);
+      s = options_->table_factory->NewTableReader(
+          *options_, toptions, internal_comparator, std::move(file), file_size,
+          &table_reader);
    }

    if (!s.ok()) {
@@ -94,25 +99,29 @@ Status TableCache::FindTable(const EnvOptions& toptions,

 Iterator* TableCache::NewIterator(const ReadOptions& options,
                                  const EnvOptions& toptions,
-                                  uint64_t file_number,
-                                  uint64_t file_size,
+                                  const InternalKeyComparator& icomparator,
+                                  const FileMetaData& file_meta,
                                  TableReader** table_reader_ptr,
                                  bool for_compaction) {
  if (table_reader_ptr != nullptr) {
    *table_reader_ptr = nullptr;
  }
-
+  TableReader* table_reader = file_meta.table_reader;
  Cache::Handle* handle = nullptr;
-  Status s = FindTable(toptions, file_number, file_size, &handle,
-                       nullptr, options.read_tier == kBlockCacheTier);
-  if (!s.ok()) {
-    return NewErrorIterator(s);
+  Status s;
+  if (table_reader == nullptr) {
+    s = FindTable(toptions, icomparator, file_meta.number, file_meta.file_size,
+                  &handle, nullptr, options.read_tier == kBlockCacheTier);
+    if (!s.ok()) {
+      return NewErrorIterator(s);
+    }
+    table_reader = GetTableReaderFromHandle(handle);
  }

-  TableReader* table_reader =
-    reinterpret_cast<TableReader*>(cache_->Value(handle));
  Iterator* result = table_reader->NewIterator(options);
-  result->RegisterCleanup(&UnrefEntry, cache_.get(), handle);
+  if (handle != nullptr) {
+    result->RegisterCleanup(&UnrefEntry, cache_, handle);
+  }
  if (table_reader_ptr != nullptr) {
    *table_reader_ptr = table_reader;
  }
@@ -125,22 +134,27 @@ Iterator* TableCache::NewIterator(const ReadOptions& options,
 }

 Status TableCache::Get(const ReadOptions& options,
-                       uint64_t file_number,
-                       uint64_t file_size,
-                       const Slice& k,
-                       void* arg,
-                       bool (*saver)(void*, const Slice&, const Slice&, bool),
-                       bool* table_io,
-                       void (*mark_key_may_exist)(void*)) {
+                       const InternalKeyComparator& internal_comparator,
+                       const FileMetaData& file_meta, const Slice& k, void* arg,
+                       bool (*saver)(void*, const ParsedInternalKey&,
+                                     const Slice&, bool),
+                       bool* table_io, void (*mark_key_may_exist)(void*)) {
+  TableReader* t = file_meta.table_reader;
+  Status s;
  Cache::Handle* handle = nullptr;
-  Status s = FindTable(storage_options_, file_number, file_size,
-                       &handle, table_io,
-                       options.read_tier == kBlockCacheTier);
+  if (!t) {
+    s = FindTable(storage_options_, internal_comparator, file_meta.number,
+                  file_meta.file_size, &handle, table_io,
+                  options.read_tier == kBlockCacheTier);
+    if (s.ok()) {
+      t = GetTableReaderFromHandle(handle);
+    }
+  }
  if (s.ok()) {
-    TableReader* t =
-      reinterpret_cast<TableReader*>(cache_->Value(handle));
    s = t->Get(options, k, arg, saver, mark_key_may_exist);
-    cache_->Release(handle);
+    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);
@@ -148,27 +162,36 @@ Status TableCache::Get(const ReadOptions& options,
  }
  return s;
 }
+Status TableCache::GetTableProperties(
+    const EnvOptions& toptions,
+    const InternalKeyComparator& internal_comparator,
+    const FileMetaData& file_meta,
+    std::shared_ptr<const TableProperties>* properties, bool no_io) {
+  Status s;
+  auto table_reader = file_meta.table_reader;
+  // table already been pre-loaded?
+  if (table_reader) {
+    *properties = table_reader->GetTableProperties();

-bool TableCache::PrefixMayMatch(const ReadOptions& options,
-                                uint64_t file_number,
-                                uint64_t file_size,
-                                const Slice& internal_prefix,
-                                bool* table_io) {
-  Cache::Handle* handle = nullptr;
-  Status s = FindTable(storage_options_, file_number,
-                       file_size, &handle, table_io);
-  bool may_match = true;
-  if (s.ok()) {
-    TableReader* t =
-      reinterpret_cast<TableReader*>(cache_->Value(handle));
-    may_match = t->PrefixMayMatch(internal_prefix);
-    cache_->Release(handle);
+    return s;
  }
-  return may_match;
+
+  bool table_io;
+  Cache::Handle* table_handle = nullptr;
+  s = FindTable(toptions, internal_comparator, file_meta.number,
+                file_meta.file_size, &table_handle, &table_io, no_io);
+  if (!s.ok()) {
+    return s;
+  }
+  assert(table_handle);
+  auto table = GetTableReaderFromHandle(table_handle);
+  *properties = table->GetTableProperties();
+  ReleaseHandle(table_handle);
+  return s;
 }

-void TableCache::Evict(uint64_t file_number) {
-  cache_->Erase(GetSliceForFileNumber(&file_number));
+void TableCache::Evict(Cache* cache, uint64_t file_number) {
+  cache->Erase(GetSliceForFileNumber(&file_number));
 }

 }  // namespace rocksdb
@@ -12,20 +12,25 @@
 #pragma once
 #include <string>
 #include <stdint.h>
+
 #include "db/dbformat.h"
-#include "rocksdb/env.h"
-#include "rocksdb/cache.h"
 #include "port/port.h"
+#include "rocksdb/cache.h"
+#include "rocksdb/env.h"
 #include "rocksdb/table.h"
+#include "table/table_reader.h"

 namespace rocksdb {

 class Env;
+struct FileMetaData;

+// TODO(sdong): try to come up with a better API to pass the file information
+//              other than simply passing FileMetaData.
 class TableCache {
 public:
  TableCache(const std::string& dbname, const Options* options,
-             const EnvOptions& storage_options, int entries);
+             const EnvOptions& storage_options, Cache* cache);
  ~TableCache();

  // Return an iterator for the specified file number (the corresponding
@@ -35,10 +40,9 @@ class TableCache {
  // the returned iterator.  The returned "*tableptr" object is owned by
  // the cache and should not be deleted, and is valid for as long as the
  // returned iterator is live.
-  Iterator* NewIterator(const ReadOptions& options,
-                        const EnvOptions& toptions,
-                        uint64_t file_number,
-                        uint64_t file_size,
+  Iterator* NewIterator(const ReadOptions& options, const EnvOptions& toptions,
+                        const InternalKeyComparator& internal_comparator,
+                        const FileMetaData& file_meta,
                        TableReader** table_reader_ptr = nullptr,
                        bool for_compaction = false);

@@ -46,33 +50,45 @@ class TableCache {
  // call (*handle_result)(arg, found_key, found_value) repeatedly until
  // it returns false.
  Status Get(const ReadOptions& options,
-             uint64_t file_number,
-             uint64_t file_size,
-             const Slice& k,
-             void* arg,
-             bool (*handle_result)(void*, const Slice&, const Slice&, bool),
-             bool* table_io,
-             void (*mark_key_may_exist)(void*) = nullptr);
-
-  // Determine whether the table may contain the specified prefix.  If
-  // the table index of blooms are not in memory, this may cause an I/O
-  bool PrefixMayMatch(const ReadOptions& options, uint64_t file_number,
-                      uint64_t file_size, const Slice& internal_prefix,
-                      bool* table_io);
+             const InternalKeyComparator& internal_comparator,
+             const FileMetaData& file_meta, const Slice& k, void* arg,
+             bool (*handle_result)(void*, const ParsedInternalKey&,
+                                   const Slice&, bool),
+             bool* table_io, void (*mark_key_may_exist)(void*) = nullptr);

  // Evict any entry for the specified file number
-  void Evict(uint64_t file_number);
+  static void Evict(Cache* cache, uint64_t file_number);
+
+  // Find table reader
+  Status FindTable(const EnvOptions& toptions,
+                   const InternalKeyComparator& internal_comparator,
+                   uint64_t file_number, uint64_t file_size, Cache::Handle**,
+                   bool* table_io = nullptr, const bool no_io = false);
+
+  // Get TableReader from a cache handle.
+  TableReader* GetTableReaderFromHandle(Cache::Handle* handle);
+
+  // Get the table properties of a given table.
+  // @no_io: indicates if we should load table to the cache if it is not present
+  //         in table cache yet.
+  // @returns: `properties` will be reset on success. Please note that we will
+  //            return Status::Incomplete() if table is not present in cache and
+  //            we set `no_io` to be true.
+  Status GetTableProperties(const EnvOptions& toptions,
+                            const InternalKeyComparator& internal_comparator,
+                            const FileMetaData& file_meta,
+                            std::shared_ptr<const TableProperties>* properties,
+                            bool no_io = false);
+
+  // Release the handle from a cache
+  void ReleaseHandle(Cache::Handle* handle);

 private:
  Env* const env_;
  const std::string dbname_;
  const Options* options_;
  const EnvOptions& storage_options_;
-  std::shared_ptr<Cache> cache_;
-
-  Status FindTable(const EnvOptions& toptions, uint64_t file_number,
-                   uint64_t file_size, Cache::Handle**, bool* table_io=nullptr,
-                   const bool no_io = false);
+  Cache* const cache_;
 };

 }  // namespace rocksdb
@@ -10,87 +10,6 @@

 namespace rocksdb {

-namespace {
-  void AppendProperty(
-      std::string& props,
-      const std::string& key,
-      const std::string& value,
-      const std::string& prop_delim,
-      const std::string& kv_delim) {
-    props.append(key);
-    props.append(kv_delim);
-    props.append(value);
-    props.append(prop_delim);
-  }
-
-  template <class TValue>
-  void AppendProperty(
-      std::string& props,
-      const std::string& key,
-      const TValue& value,
-      const std::string& prop_delim,
-      const std::string& kv_delim) {
-    AppendProperty(
-        props, key, std::to_string(value), prop_delim, kv_delim
-    );
-  }
-}
-
-std::string TableProperties::ToString(
-    const std::string& prop_delim,
-    const std::string& kv_delim) const {
-  std::string result;
-  result.reserve(1024);
-
-  // Basic Info
-  AppendProperty(
-      result, "# data blocks", num_data_blocks, prop_delim, kv_delim
-  );
-  AppendProperty(result, "# entries", num_entries, prop_delim, kv_delim);
-
-  AppendProperty(result, "raw key size", raw_key_size, prop_delim, kv_delim);
-  AppendProperty(
-      result,
-      "raw average key size",
-      num_entries != 0 ?  1.0 * raw_key_size / num_entries : 0.0,
-      prop_delim,
-      kv_delim
-  );
-  AppendProperty(
-      result, "raw value size", raw_value_size, prop_delim, kv_delim
-  );
-  AppendProperty(
-      result,
-      "raw average value size",
-      num_entries != 0 ?  1.0 * raw_value_size / num_entries : 0.0,
-      prop_delim,
-      kv_delim
-  );
-
-  AppendProperty(result, "data block size", data_size, prop_delim, kv_delim);
-  AppendProperty(result, "index block size", index_size, prop_delim, kv_delim);
-  AppendProperty(
-      result, "filter block size", filter_size, prop_delim, kv_delim
-  );
-  AppendProperty(
-      result,
-      "(estimated) table size",
-      data_size + index_size + filter_size,
-      prop_delim,
-      kv_delim
-  );
-
-  AppendProperty(
-      result,
-      "filter policy name",
-      filter_policy_name.empty() ? std::string("N/A") : filter_policy_name,
-      prop_delim,
-      kv_delim
-  );
-
-  return result;
-}
-
 Status InternalKeyPropertiesCollector::Add(
    const Slice& key, const Slice& value) {
  ParsedInternalKey ikey;
@@ -106,7 +25,7 @@ Status InternalKeyPropertiesCollector::Add(
 }

 Status InternalKeyPropertiesCollector::Finish(
-    TableProperties::UserCollectedProperties* properties) {
+    UserCollectedProperties* properties) {
  assert(properties);
  assert(properties->find(
        InternalKeyTablePropertiesNames::kDeletedKeys) == properties->end());
@@ -118,7 +37,7 @@ Status InternalKeyPropertiesCollector::Finish(
  return Status::OK();
 }

-TableProperties::UserCollectedProperties
+UserCollectedProperties
 InternalKeyPropertiesCollector::GetReadableProperties() const {
  return {
    { "kDeletedKeys", std::to_string(deleted_keys_) }
@@ -137,11 +56,11 @@ Status UserKeyTablePropertiesCollector::Add(
 }

 Status UserKeyTablePropertiesCollector::Finish(
-    TableProperties::UserCollectedProperties* properties) {
+    UserCollectedProperties* properties) {
  return collector_->Finish(properties);
 }

-TableProperties::UserCollectedProperties
+UserCollectedProperties
 UserKeyTablePropertiesCollector::GetReadableProperties() const {
  return collector_->GetReadableProperties();
 }
@@ -151,7 +70,7 @@ const std::string InternalKeyTablePropertiesNames::kDeletedKeys
  = "rocksdb.deleted.keys";

 uint64_t GetDeletedKeys(
-    const TableProperties::UserCollectedProperties& props) {
+    const UserCollectedProperties& props) {
  auto pos = props.find(InternalKeyTablePropertiesNames::kDeletedKeys);
  if (pos == props.end()) {
    return 0;
@@ -24,15 +24,13 @@ class InternalKeyPropertiesCollector : public TablePropertiesCollector {
 public:
  virtual Status Add(const Slice& key, const Slice& value) override;

-  virtual Status Finish(
-      TableProperties::UserCollectedProperties* properties) override;
+  virtual Status Finish(UserCollectedProperties* properties) override;

  virtual const char* Name() const override {
    return "InternalKeyPropertiesCollector";
  }

-  TableProperties::UserCollectedProperties
-    GetReadableProperties() const override;
+  UserCollectedProperties GetReadableProperties() const override;

 private:
  uint64_t deleted_keys_ = 0;
@@ -61,13 +59,11 @@ class UserKeyTablePropertiesCollector : public TablePropertiesCollector {

  virtual Status Add(const Slice& key, const Slice& value) override;

-  virtual Status Finish(
-      TableProperties::UserCollectedProperties* properties) override;
+  virtual Status Finish(UserCollectedProperties* properties) override;

  virtual const char* Name() const override { return collector_->Name(); }

-  TableProperties::UserCollectedProperties
-    GetReadableProperties() const override;
+  UserCollectedProperties GetReadableProperties() const override;

 protected:
  std::shared_ptr<TablePropertiesCollector> collector_;
@@ -7,12 +7,14 @@
 #include <memory>
 #include <string>

-#include "db/dbformat.h"
 #include "db/db_impl.h"
+#include "db/dbformat.h"
 #include "db/table_properties_collector.h"
-#include "rocksdb/table_properties.h"
 #include "rocksdb/table.h"
 #include "table/block_based_table_factory.h"
+#include "table/meta_blocks.h"
+#include "table/plain_table_factory.h"
+#include "table/table_builder.h"
 #include "util/coding.h"
 #include "util/testharness.h"
 #include "util/testutil.h"
@@ -20,8 +22,6 @@
 namespace rocksdb {

 class TablePropertiesTest {
- private:
-  unique_ptr<TableReader> table_reader_;
 };

 // TODO(kailiu) the following classes should be moved to some more general
@@ -83,41 +83,26 @@ class DumbLogger : public Logger {
 };

 // Utilities test functions
-void MakeBuilder(
-    const Options& options,
-    std::unique_ptr<FakeWritableFile>* writable,
-    std::unique_ptr<TableBuilder>* builder) {
+namespace {
+void MakeBuilder(const Options& options,
+                 const InternalKeyComparator& internal_comparator,
+                 std::unique_ptr<FakeWritableFile>* writable,
+                 std::unique_ptr<TableBuilder>* builder) {
  writable->reset(new FakeWritableFile);
-  builder->reset(
-      options.table_factory->GetTableBuilder(options, writable->get(),
-                                             options.compression));
-}
-
-void OpenTable(
-    const Options& options,
-    const std::string& contents,
-    std::unique_ptr<TableReader>* table_reader) {
-
-  std::unique_ptr<RandomAccessFile> file(new FakeRandomeAccessFile(contents));
-  auto s = options.table_factory->GetTableReader(
-      options,
-      EnvOptions(),
-      std::move(file),
-      contents.size(),
-      table_reader
-  );
-  ASSERT_OK(s);
+  builder->reset(options.table_factory->NewTableBuilder(
+      options, internal_comparator, writable->get(), options.compression));
 }
+}  // namespace

 // Collects keys that starts with "A" in a table.
 class RegularKeysStartWithA: public TablePropertiesCollector {
 public:
   const char* Name() const { return "RegularKeysStartWithA"; }

-   Status Finish(TableProperties::UserCollectedProperties* properties) {
+   Status Finish(UserCollectedProperties* properties) {
     std::string encoded;
     PutVarint32(&encoded, count_);
-     *properties = TableProperties::UserCollectedProperties {
+     *properties = UserCollectedProperties {
       { "TablePropertiesTest", "Rocksdb" },
       { "Count", encoded }
     };
@@ -132,9 +117,8 @@ class RegularKeysStartWithA: public TablePropertiesCollector {
     return Status::OK();
   }

-  virtual TableProperties::UserCollectedProperties
-    GetReadableProperties() const {
-      return {};
+  virtual UserCollectedProperties GetReadableProperties() const {
+    return UserCollectedProperties{};
  }


@@ -142,23 +126,68 @@ class RegularKeysStartWithA: public TablePropertiesCollector {
  uint32_t count_ = 0;
 };

-TEST(TablePropertiesTest, CustomizedTablePropertiesCollector) {
-  Options options;
-
+extern uint64_t kBlockBasedTableMagicNumber;
+extern uint64_t kPlainTableMagicNumber;
+namespace {
+void TestCustomizedTablePropertiesCollector(
+    uint64_t magic_number, bool encode_as_internal, const Options& options,
+    const InternalKeyComparator& internal_comparator) {
  // make sure the entries will be inserted with order.
  std::map<std::string, std::string> kvs = {
-    {"About",     "val5"},  // starts with 'A'
-    {"Abstract",  "val2"},  // starts with 'A'
-    {"Around",    "val7"},  // starts with 'A'
-    {"Beyond",    "val3"},
-    {"Builder",   "val1"},
-    {"Cancel",    "val4"},
-    {"Find",      "val6"},
+    {"About   ", "val5"},  // starts with 'A'
+    {"Abstract", "val2"},  // starts with 'A'
+    {"Around  ", "val7"},  // starts with 'A'
+    {"Beyond  ", "val3"},
+    {"Builder ", "val1"},
+    {"Cancel  ", "val4"},
+    {"Find    ", "val6"},
  };

+  // -- Step 1: build table
+  std::unique_ptr<TableBuilder> builder;
+  std::unique_ptr<FakeWritableFile> writable;
+  MakeBuilder(options, internal_comparator, &writable, &builder);
+
+  for (const auto& kv : kvs) {
+    if (encode_as_internal) {
+      InternalKey ikey(kv.first, 0, ValueType::kTypeValue);
+      builder->Add(ikey.Encode(), kv.second);
+    } else {
+      builder->Add(kv.first, kv.second);
+    }
+  }
+  ASSERT_OK(builder->Finish());
+
+  // -- Step 2: Read properties
+  FakeRandomeAccessFile readable(writable->contents());
+  TableProperties* props;
+  Status s = ReadTableProperties(
+      &readable,
+      writable->contents().size(),
+      magic_number,
+      Env::Default(),
+      nullptr,
+      &props
+  );
+  std::unique_ptr<TableProperties> props_guard(props);
+  ASSERT_OK(s);
+
+  auto user_collected = props->user_collected_properties;
+
+  ASSERT_EQ("Rocksdb", user_collected.at("TablePropertiesTest"));
+
+  uint32_t starts_with_A = 0;
+  Slice key(user_collected.at("Count"));
+  ASSERT_TRUE(GetVarint32(&key, &starts_with_A));
+  ASSERT_EQ(3u, starts_with_A);
+}
+}  // namespace
+
+TEST(TablePropertiesTest, CustomizedTablePropertiesCollector) {
  // Test properties collectors with internal keys or regular keys
+  // for block based table
  for (bool encode_as_internal : { true, false }) {
-    // -- Step 1: build table
+    Options options;
    auto collector = new RegularKeysStartWithA();
    if (encode_as_internal) {
      options.table_properties_collectors = {
@@ -168,95 +197,112 @@ TEST(TablePropertiesTest, CustomizedTablePropertiesCollector) {
      options.table_properties_collectors.resize(1);
      options.table_properties_collectors[0].reset(collector);
    }
-    std::unique_ptr<TableBuilder> builder;
-    std::unique_ptr<FakeWritableFile> writable;
-    MakeBuilder(options, &writable, &builder);
-
-    for (const auto& kv : kvs) {
-      if (encode_as_internal) {
-        InternalKey ikey(kv.first, 0, ValueType::kTypeValue);
-        builder->Add(ikey.Encode(), kv.second);
-      } else {
-        builder->Add(kv.first, kv.second);
-      }
-    }
-    ASSERT_OK(builder->Finish());
-
-    // -- Step 2: Open table
-    std::unique_ptr<TableReader> table_reader;
-    OpenTable(options, writable->contents(), &table_reader);
-    const auto& properties =
-      table_reader->GetTableProperties().user_collected_properties;
-
-    ASSERT_EQ("Rocksdb", properties.at("TablePropertiesTest"));
-
-    uint32_t starts_with_A = 0;
-    Slice key(properties.at("Count"));
-    ASSERT_TRUE(GetVarint32(&key, &starts_with_A));
-    ASSERT_EQ(3u, starts_with_A);
+    test::PlainInternalKeyComparator ikc(options.comparator);
+    TestCustomizedTablePropertiesCollector(kBlockBasedTableMagicNumber,
+                                           encode_as_internal, options, ikc);
  }
+
+  // test plain table
+  Options options;
+  options.table_properties_collectors.push_back(
+      std::make_shared<RegularKeysStartWithA>()
+  );
+  options.table_factory = std::make_shared<PlainTableFactory>(8, 8, 0);
+  test::PlainInternalKeyComparator ikc(options.comparator);
+  TestCustomizedTablePropertiesCollector(kPlainTableMagicNumber, true, options,
+                                         ikc);
 }

-TEST(TablePropertiesTest, InternalKeyPropertiesCollector) {
+namespace {
+void TestInternalKeyPropertiesCollector(
+    uint64_t magic_number,
+    bool sanitized,
+    std::shared_ptr<TableFactory> table_factory) {
  InternalKey keys[] = {
-    InternalKey("A", 0, ValueType::kTypeValue),
-    InternalKey("B", 0, ValueType::kTypeValue),
-    InternalKey("C", 0, ValueType::kTypeValue),
-    InternalKey("W", 0, ValueType::kTypeDeletion),
-    InternalKey("X", 0, ValueType::kTypeDeletion),
-    InternalKey("Y", 0, ValueType::kTypeDeletion),
-    InternalKey("Z", 0, ValueType::kTypeDeletion),
+    InternalKey("A       ", 0, ValueType::kTypeValue),
+    InternalKey("B       ", 0, ValueType::kTypeValue),
+    InternalKey("C       ", 0, ValueType::kTypeValue),
+    InternalKey("W       ", 0, ValueType::kTypeDeletion),
+    InternalKey("X       ", 0, ValueType::kTypeDeletion),
+    InternalKey("Y       ", 0, ValueType::kTypeDeletion),
+    InternalKey("Z       ", 0, ValueType::kTypeDeletion),
  };

-  for (bool sanitized : { false, true }) {
-    std::unique_ptr<TableBuilder> builder;
-    std::unique_ptr<FakeWritableFile> writable;
-    Options options;
-    if (sanitized) {
-      options.table_properties_collectors = {
-        std::make_shared<RegularKeysStartWithA>()
-      };
-      // with sanitization, even regular properties collector will be able to
-      // handle internal keys.
-      auto comparator = options.comparator;
-      // HACK: Set options.info_log to avoid writing log in
-      // SanitizeOptions().
-      options.info_log = std::make_shared<DumbLogger>();
-      options = SanitizeOptions(
-          "db",  // just a place holder
-          nullptr,  // with skip internal key comparator
-          nullptr,  // don't care filter policy
-          options
-      );
-      options.comparator = comparator;
-    } else {
-      options.table_properties_collectors = {
-        std::make_shared<InternalKeyPropertiesCollector>()
-      };
-    }
+  std::unique_ptr<TableBuilder> builder;
+  std::unique_ptr<FakeWritableFile> writable;
+  Options options;
+  test::PlainInternalKeyComparator pikc(options.comparator);

-    MakeBuilder(options, &writable, &builder);
-    for (const auto& k : keys) {
-      builder->Add(k.Encode(), "val");
-    }
-
-    ASSERT_OK(builder->Finish());
-
-    std::unique_ptr<TableReader> table_reader;
-    OpenTable(options, writable->contents(), &table_reader);
-    const auto& properties =
-      table_reader->GetTableProperties().user_collected_properties;
-
-    uint64_t deleted = GetDeletedKeys(properties);
-    ASSERT_EQ(4u, deleted);
-
-    if (sanitized) {
-      uint32_t starts_with_A = 0;
-      Slice key(properties.at("Count"));
-      ASSERT_TRUE(GetVarint32(&key, &starts_with_A));
-      ASSERT_EQ(1u, starts_with_A);
-    }
+  options.table_factory = table_factory;
+  if (sanitized) {
+    options.table_properties_collectors = {
+      std::make_shared<RegularKeysStartWithA>()
+    };
+    // with sanitization, even regular properties collector will be able to
+    // handle internal keys.
+    auto comparator = options.comparator;
+    // HACK: Set options.info_log to avoid writing log in
+    // SanitizeOptions().
+    options.info_log = std::make_shared<DumbLogger>();
+    options = SanitizeOptions("db",            // just a place holder
+                              &pikc, nullptr,  // don't care filter policy
+                              options);
+    options.comparator = comparator;
+  } else {
+    options.table_properties_collectors = {
+      std::make_shared<InternalKeyPropertiesCollector>()
+    };
  }
+
+  MakeBuilder(options, pikc, &writable, &builder);
+  for (const auto& k : keys) {
+    builder->Add(k.Encode(), "val");
+  }
+
+  ASSERT_OK(builder->Finish());
+
+  FakeRandomeAccessFile readable(writable->contents());
+  TableProperties* props;
+  Status s = ReadTableProperties(
+      &readable,
+      writable->contents().size(),
+      magic_number,
+      Env::Default(),
+      nullptr,
+      &props
+  );
+  ASSERT_OK(s);
+
+  std::unique_ptr<TableProperties> props_guard(props);
+  auto user_collected = props->user_collected_properties;
+  uint64_t deleted = GetDeletedKeys(user_collected);
+  ASSERT_EQ(4u, deleted);
+
+  if (sanitized) {
+    uint32_t starts_with_A = 0;
+    Slice key(user_collected.at("Count"));
+    ASSERT_TRUE(GetVarint32(&key, &starts_with_A));
+    ASSERT_EQ(1u, starts_with_A);
+  }
+}
+}  // namespace
+
+TEST(TablePropertiesTest, InternalKeyPropertiesCollector) {
+  TestInternalKeyPropertiesCollector(
+      kBlockBasedTableMagicNumber,
+      true /* sanitize */,
+      std::make_shared<BlockBasedTableFactory>()
+  );
+  TestInternalKeyPropertiesCollector(
+      kBlockBasedTableMagicNumber,
+      true /* not sanitize */,
+      std::make_shared<BlockBasedTableFactory>()
+  );
+  TestInternalKeyPropertiesCollector(
+      kPlainTableMagicNumber,
+      false /* not sanitize */,
+      std::make_shared<PlainTableFactory>(8, 8, 0)
+  );
 }

 }  // namespace rocksdb
@@ -3,22 +3,36 @@
 //  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(DBImpl* db, const ReadOptions& options,
-                                 const Comparator* comparator)
-    : db_(db), options_(options), comparator_(comparator),
-      version_number_(0), current_(nullptr),
+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;
 }
@@ -46,18 +60,17 @@ void TailingIterator::Seek(const Slice& target) {
  // need to do a seek if 'target' belongs to that interval (i.e. immutable_ is
  // already at the correct position)!
  //
-  // If options.prefix_seek is used and immutable_ is not valid, seek if target
-  // has a different prefix than prev_key.
+  // 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.

-  if (!is_prev_set_ ||
-      comparator_->Compare(prev_key_, target) >= !is_prev_inclusive_ ||
-      (immutable_->Valid() &&
-       comparator_->Compare(target, immutable_->key()) > 0) ||
-      (options_.prefix_seek && !IsSamePrefix(target))) {
+  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);
  }

@@ -119,14 +132,45 @@ 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() {
-  std::pair<Iterator*, Iterator*> iters =
-    db_->GetTailingIteratorPair(options_, &version_number_);
+  Cleanup();
+  super_version_= cfd_->GetReferencedSuperVersion(&(db_->mutex_));

-  assert(iters.first && iters.second);
+  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));

-  mutable_.reset(iters.first);
-  immutable_.reset(iters.second);
  current_ = nullptr;
  is_prev_set_ = false;
 }
@@ -137,9 +181,10 @@ void TailingIterator::UpdateCurrent() {
  if (mutable_->Valid()) {
    current_ = mutable_.get();
  }
+  const Comparator* cmp = cfd_->user_comparator();
  if (immutable_->Valid() &&
      (current_ == nullptr ||
-       comparator_->Compare(immutable_->key(), current_->key()) < 0)) {
+       cmp->Compare(immutable_->key(), current_->key()) < 0)) {
    current_ = immutable_.get();
  }

@@ -150,12 +195,12 @@ void TailingIterator::UpdateCurrent() {
 }

 bool TailingIterator::IsCurrentVersion() const {
-  return mutable_ != nullptr && immutable_ != nullptr &&
-    version_number_ == db_->CurrentVersionNumber();
+  return super_version_ != nullptr &&
+         super_version_->version_number == cfd_->GetSuperVersionNumber();
 }

 bool TailingIterator::IsSamePrefix(const Slice& target) const {
-  const SliceTransform* extractor = db_->options_.prefix_extractor;
+  const SliceTransform* extractor = cfd_->options()->prefix_extractor.get();

  assert(extractor);
  assert(is_prev_set_);
@@ -173,3 +218,4 @@ void TailingIterator::SeekImmutable(const Slice& target) {
 }

 }  // namespace rocksdb
+#endif  // ROCKSDB_LITE
@@ -4,6 +4,8 @@
 //  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"
@@ -13,6 +15,9 @@
 namespace rocksdb {

 class DBImpl;
+class Env;
+struct SuperVersion;
+class ColumnFamilyData;

 /**
 * TailingIterator is a special type of iterator that doesn't use an (implicit)
@@ -24,9 +29,9 @@ class DBImpl;
 */
 class TailingIterator : public Iterator {
 public:
-  TailingIterator(DBImpl* db, const ReadOptions& options,
-                  const Comparator* comparator);
-  virtual ~TailingIterator() {}
+  TailingIterator(Env* const env, DBImpl* db, const ReadOptions& read_options,
+                  ColumnFamilyData* cfd);
+  virtual ~TailingIterator();

  virtual bool Valid() const override;
  virtual void SeekToFirst() override;
@@ -39,10 +44,13 @@ class TailingIterator : public Iterator {
  virtual Status status() const override;

 private:
+  void Cleanup();
+
+  Env* const env_;
  DBImpl* const db_;
-  const ReadOptions options_;
-  const Comparator* const comparator_;
-  uint64_t version_number_;
+  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).
@@ -72,7 +80,7 @@ class TailingIterator : public Iterator {
  bool IsCurrentVersion() const;

  // check if SeekImmutable() is needed due to target having a different prefix
-  // than prev_key_ (used when options.prefix_seek is set)
+  // than prev_key_ (used when in prefix seek mode)
  bool IsSamePrefix(const Slice& target) const;

  // creates mutable_ and immutable_ iterators and updates version_number_
@@ -86,3 +94,4 @@ class TailingIterator : public Iterator {
 };

 }  // namespace rocksdb
+#endif  // ROCKSDB_LITE
@@ -2,30 +2,30 @@
 //  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/transaction_log_impl.h"
 #include "db/write_batch_internal.h"

 namespace rocksdb {

 TransactionLogIteratorImpl::TransactionLogIteratorImpl(
-                           const std::string& dir,
-                           const Options* options,
-                           const EnvOptions& soptions,
-                           const SequenceNumber seq,
-                           std::unique_ptr<VectorLogPtr> files,
-                           DBImpl const * const dbimpl) :
-    dir_(dir),
-    options_(options),
-    soptions_(soptions),
-    startingSequenceNumber_(seq),
-    files_(std::move(files)),
-    started_(false),
-    isValid_(false),
-    currentFileIndex_(0),
-    currentBatchSeq_(0),
-    currentLastSeq_(0),
-    dbimpl_(dbimpl) {
+    const std::string& dir, const DBOptions* options,
+    const TransactionLogIterator::ReadOptions& read_options,
+    const EnvOptions& soptions, const SequenceNumber seq,
+    std::unique_ptr<VectorLogPtr> files, DBImpl const* const dbimpl)
+    : dir_(dir),
+      options_(options),
+      read_options_(read_options),
+      soptions_(soptions),
+      startingSequenceNumber_(seq),
+      files_(std::move(files)),
+      started_(false),
+      isValid_(false),
+      currentFileIndex_(0),
+      currentBatchSeq_(0),
+      currentLastSeq_(0),
+      dbimpl_(dbimpl) {
  assert(files_ != nullptr);
  assert(dbimpl_ != nullptr);

@@ -49,9 +49,6 @@ Status TransactionLogIteratorImpl::OpenLogFile(
      //  Try the archive dir, as it could have moved in the meanwhile.
      fname = ArchivedLogFileName(dir_, logFile->LogNumber());
      status = env->NewSequentialFile(fname, file, soptions_);
-      if (!status.ok()) {
-        return Status::IOError("Requested file not present in the dir");
-      }
    }
    return status;
  }
@@ -190,7 +187,7 @@ void TransactionLogIteratorImpl::NextImpl(bool internal) {
      if (currentLastSeq_ == dbimpl_->GetLatestSequenceNumber()) {
        currentStatus_ = Status::OK();
      } else {
-        currentStatus_ = Status::IOError("NO MORE DATA LEFT");
+        currentStatus_ = Status::Corruption("NO MORE DATA LEFT");
      }
      return;
    }
@@ -256,9 +253,9 @@ Status TransactionLogIteratorImpl::OpenLogReader(const LogFile* logFile) {
    return status;
  }
  assert(file);
-  currentLogReader_.reset(
-    new log::Reader(std::move(file), &reporter_, true, 0)
-  );
+  currentLogReader_.reset(new log::Reader(std::move(file), &reporter_,
+                                          read_options_.verify_checksums_, 0));
  return Status::OK();
 }
 }  //  namespace rocksdb
+#endif  // ROCKSDB_LITE
@@ -2,7 +2,8 @@
 //  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
 #pragma once
 #include <vector>

@@ -66,12 +67,11 @@ class LogFileImpl : public LogFile {

 class TransactionLogIteratorImpl : public TransactionLogIterator {
 public:
-  TransactionLogIteratorImpl(const std::string& dir,
-                             const Options* options,
-                             const EnvOptions& soptions,
-                             const SequenceNumber seqNum,
-                             std::unique_ptr<VectorLogPtr> files,
-                             DBImpl const * const dbimpl);
+  TransactionLogIteratorImpl(
+      const std::string& dir, const DBOptions* options,
+      const TransactionLogIterator::ReadOptions& read_options,
+      const EnvOptions& soptions, const SequenceNumber seqNum,
+      std::unique_ptr<VectorLogPtr> files, DBImpl const* const dbimpl);

  virtual bool Valid();

@@ -83,7 +83,8 @@ class TransactionLogIteratorImpl : public TransactionLogIterator {

 private:
  const std::string& dir_;
-  const Options* options_;
+  const DBOptions* options_;
+  const TransactionLogIterator::ReadOptions read_options_;
  const EnvOptions& soptions_;
  SequenceNumber startingSequenceNumber_;
  std::unique_ptr<VectorLogPtr> files_;
@@ -116,3 +117,4 @@ class TransactionLogIteratorImpl : public TransactionLogIterator {
  Status OpenLogReader(const LogFile* file);
 };
 }  //  namespace rocksdb
+#endif  // ROCKSDB_LITE
@@ -11,6 +11,7 @@

 #include "db/version_set.h"
 #include "util/coding.h"
+#include "rocksdb/slice.h"

 namespace rocksdb {

@@ -28,7 +29,12 @@ enum Tag {
  kPrevLogNumber        = 9,

  // these are new formats divergent from open source leveldb
-  kNewFile2             = 100  // store smallest & largest seqno
+  kNewFile2             = 100,  // store smallest & largest seqno
+
+  kColumnFamily         = 200,  // specify column family for version edit
+  kColumnFamilyAdd      = 201,
+  kColumnFamilyDrop     = 202,
+  kMaxColumnFamily      = 203,
 };

 void VersionEdit::Clear() {
@@ -38,13 +44,19 @@ void VersionEdit::Clear() {
  prev_log_number_ = 0;
  last_sequence_ = 0;
  next_file_number_ = 0;
+  max_column_family_ = 0;
  has_comparator_ = false;
  has_log_number_ = false;
  has_prev_log_number_ = false;
  has_next_file_number_ = false;
  has_last_sequence_ = false;
+  has_max_column_family_ = false;
  deleted_files_.clear();
  new_files_.clear();
+  column_family_ = 0;
+  is_column_family_add_ = 0;
+  is_column_family_drop_ = 0;
+  column_family_name_.clear();
 }

 void VersionEdit::EncodeTo(std::string* dst) const {
@@ -68,13 +80,15 @@ void VersionEdit::EncodeTo(std::string* dst) const {
    PutVarint32(dst, kLastSequence);
    PutVarint64(dst, last_sequence_);
  }
+  if (has_max_column_family_) {
+    PutVarint32(dst, kMaxColumnFamily);
+    PutVarint32(dst, max_column_family_);
+  }

-  for (DeletedFileSet::const_iterator iter = deleted_files_.begin();
-       iter != deleted_files_.end();
-       ++iter) {
+  for (const auto& deleted : deleted_files_) {
    PutVarint32(dst, kDeletedFile);
-    PutVarint32(dst, iter->first);   // level
-    PutVarint64(dst, iter->second);  // file number
+    PutVarint32(dst, deleted.first /* level */);
+    PutVarint64(dst, deleted.second /* file number */);
  }

  for (size_t i = 0; i < new_files_.size(); i++) {
@@ -88,6 +102,21 @@ void VersionEdit::EncodeTo(std::string* dst) const {
    PutVarint64(dst, f.smallest_seqno);
    PutVarint64(dst, f.largest_seqno);
  }
+
+  // 0 is default and does not need to be explicitly written
+  if (column_family_ != 0) {
+    PutVarint32(dst, kColumnFamily);
+    PutVarint32(dst, column_family_);
+  }
+
+  if (is_column_family_add_) {
+    PutVarint32(dst, kColumnFamilyAdd);
+    PutLengthPrefixedSlice(dst, Slice(column_family_name_));
+  }
+
+  if (is_column_family_drop_) {
+    PutVarint32(dst, kColumnFamilyDrop);
+  }
 }

 static bool GetInternalKey(Slice* input, InternalKey* dst) {
@@ -169,6 +198,14 @@ Status VersionEdit::DecodeFrom(const Slice& src) {
        }
        break;

+      case kMaxColumnFamily:
+        if (GetVarint32(&input, &max_column_family_)) {
+          has_max_column_family_ = true;
+        } else {
+          msg = "max column family";
+        }
+        break;
+
      case kCompactPointer:
        if (GetLevel(&input, &level, &msg) &&
            GetInternalKey(&input, &key)) {
@@ -223,6 +260,29 @@ Status VersionEdit::DecodeFrom(const Slice& src) {
        }
        break;

+      case kColumnFamily:
+        if (!GetVarint32(&input, &column_family_)) {
+          if (!msg) {
+            msg = "set column family id";
+          }
+        }
+        break;
+
+      case kColumnFamilyAdd:
+        if (GetLengthPrefixedSlice(&input, &str)) {
+          is_column_family_add_ = true;
+          column_family_name_ = str.ToString();
+        } else {
+          if (!msg) {
+            msg = "column family add";
+          }
+        }
+        break;
+
+      case kColumnFamilyDrop:
+        is_column_family_drop_ = true;
+        break;
+
      default:
        msg = "unknown tag";
        break;
@@ -284,6 +344,19 @@ std::string VersionEdit::DebugString(bool hex_key) const {
    r.append(" .. ");
    r.append(f.largest.DebugString(hex_key));
  }
+  r.append("\n  ColumnFamily: ");
+  AppendNumberTo(&r, column_family_);
+  if (is_column_family_add_) {
+    r.append("\n  ColumnFamilyAdd: ");
+    r.append(column_family_name_);
+  }
+  if (is_column_family_drop_) {
+    r.append("\n  ColumnFamilyDrop");
+  }
+  if (has_max_column_family_) {
+    r.append("\n  MaxColumnFamily: ");
+    AppendNumberTo(&r, max_column_family_);
+  }
  r.append("\n}\n");
  return r;
 }
@@ -11,6 +11,8 @@
 #include <set>
 #include <utility>
 #include <vector>
+#include <string>
+#include "rocksdb/cache.h"
 #include "db/dbformat.h"

 namespace rocksdb {
@@ -28,8 +30,20 @@ struct FileMetaData {
  SequenceNumber smallest_seqno;// The smallest seqno in this file
  SequenceNumber largest_seqno; // The largest seqno in this file

-  FileMetaData() : refs(0), allowed_seeks(1 << 30), file_size(0),
-                   being_compacted(false) { }
+  // Needs to be disposed when refs becomes 0.
+  Cache::Handle* table_reader_handle;
+  // Table reader in table_reader_handle
+  TableReader* table_reader;
+
+  FileMetaData(uint64_t number, uint64_t file_size)
+      : refs(0),
+        allowed_seeks(1 << 30),
+        number(number),
+        file_size(file_size),
+        being_compacted(false),
+        table_reader_handle(nullptr),
+        table_reader(nullptr) {}
+  FileMetaData() : FileMetaData(0, 0) {}
 };

 class VersionEdit {
@@ -59,6 +73,10 @@ class VersionEdit {
    has_last_sequence_ = true;
    last_sequence_ = seq;
  }
+  void SetMaxColumnFamily(uint32_t max_column_family) {
+    has_max_column_family_ = true;
+    max_column_family_ = max_column_family;
+  }

  // Add the specified file at the specified number.
  // REQUIRES: This version has not been saved (see VersionSet::SaveTo)
@@ -69,6 +87,7 @@ class VersionEdit {
               const InternalKey& largest,
               const SequenceNumber& smallest_seqno,
               const SequenceNumber& largest_seqno) {
+    assert(smallest_seqno <= largest_seqno);
    FileMetaData f;
    f.number = file;
    f.file_size = file_size;
@@ -76,13 +95,12 @@ class VersionEdit {
    f.largest = largest;
    f.smallest_seqno = smallest_seqno;
    f.largest_seqno = largest_seqno;
-    assert(smallest_seqno <= largest_seqno);
    new_files_.push_back(std::make_pair(level, f));
  }

  // Delete the specified "file" from the specified "level".
  void DeleteFile(int level, uint64_t file) {
-    deleted_files_.insert(std::make_pair(level, file));
+    deleted_files_.insert({level, file});
  }

  // Number of edits
@@ -90,6 +108,31 @@ class VersionEdit {
    return new_files_.size() + deleted_files_.size();
  }

+  bool IsColumnFamilyManipulation() {
+    return is_column_family_add_ || is_column_family_drop_;
+  }
+
+  void SetColumnFamily(uint32_t column_family_id) {
+    column_family_ = column_family_id;
+  }
+
+  // set column family ID by calling SetColumnFamily()
+  void AddColumnFamily(const std::string& name) {
+    assert(!is_column_family_drop_);
+    assert(!is_column_family_add_);
+    assert(NumEntries() == 0);
+    is_column_family_add_ = true;
+    column_family_name_ = name;
+  }
+
+  // set column family ID by calling SetColumnFamily()
+  void DropColumnFamily() {
+    assert(!is_column_family_drop_);
+    assert(!is_column_family_add_);
+    assert(NumEntries() == 0);
+    is_column_family_drop_ = true;
+  }
+
  void EncodeTo(std::string* dst) const;
  Status DecodeFrom(const Slice& src);

@@ -98,7 +141,7 @@ class VersionEdit {
 private:
  friend class VersionSet;

-  typedef std::set< std::pair<int, uint64_t> > DeletedFileSet;
+  typedef std::set< std::pair<int, uint64_t>> DeletedFileSet;

  bool GetLevel(Slice* input, int* level, const char** msg);

@@ -107,15 +150,27 @@ class VersionEdit {
  uint64_t log_number_;
  uint64_t prev_log_number_;
  uint64_t next_file_number_;
+  uint32_t max_column_family_;
  SequenceNumber last_sequence_;
  bool has_comparator_;
  bool has_log_number_;
  bool has_prev_log_number_;
  bool has_next_file_number_;
  bool has_last_sequence_;
+  bool has_max_column_family_;

  DeletedFileSet deleted_files_;
-  std::vector<std::pair<int, FileMetaData> > new_files_;
+  std::vector<std::pair<int, FileMetaData>> new_files_;
+
+  // Each version edit record should have column_family_id set
+  // If it's not set, it is default (0)
+  uint32_t column_family_;
+  // a version edit can be either column_family add or
+  // column_family drop. If it's column family add,
+  // it also includes column family name.
+  bool is_column_family_drop_;
+  bool is_column_family_add_;
+  std::string column_family_name_;
 };

 }  // namespace rocksdb
@@ -45,6 +45,19 @@ TEST(VersionEditTest, EncodeDecode) {
  TestEncodeDecode(edit);
 }

+TEST(VersionEditTest, ColumnFamilyTest) {
+  VersionEdit edit;
+  edit.SetColumnFamily(2);
+  edit.AddColumnFamily("column_family");
+  edit.SetMaxColumnFamily(5);
+  TestEncodeDecode(edit);
+
+  edit.Clear();
+  edit.SetColumnFamily(3);
+  edit.DropColumnFamily();
+  TestEncodeDecode(edit);
+}
+
 }  // namespace rocksdb

 int main(int argc, char** argv) {
@@ -23,12 +23,17 @@
 #include <set>
 #include <vector>
 #include <deque>
+#include <atomic>
+#include <limits>
 #include "db/dbformat.h"
 #include "db/version_edit.h"
 #include "port/port.h"
 #include "db/table_cache.h"
 #include "db/compaction.h"
 #include "db/compaction_picker.h"
+#include "db/column_family.h"
+#include "db/log_reader.h"
+#include "db/file_indexer.h"

 namespace rocksdb {

@@ -37,11 +42,15 @@ namespace log { class Writer; }
 class Compaction;
 class CompactionPicker;
 class Iterator;
+class LogBuffer;
+class LookupKey;
 class MemTable;
-class TableCache;
 class Version;
 class VersionSet;
 class MergeContext;
+class ColumnFamilyData;
+class ColumnFamilySet;
+class TableCache;

 // Return the smallest index i such that files[i]->largest >= key.
 // Return files.size() if there is no such file.
@@ -80,9 +89,8 @@ class Version {
    int seek_file_level;
  };
  void Get(const ReadOptions&, const LookupKey& key, std::string* val,
-           Status* status, MergeContext* merge_context,
-           GetStats* stats, const Options& db_option, bool* value_found =
-               nullptr);
+           Status* status, MergeContext* merge_context, GetStats* stats,
+           bool* value_found = nullptr);

  // Adds "stats" into the current state.  Returns true if a new
  // compaction may need to be triggered, false otherwise.
@@ -91,13 +99,16 @@ class Version {

  // Updates internal structures that keep track of compaction scores
  // We use compaction scores to figure out which compaction to do next
-  // Also pre-sorts level0 files for Get()
-  void Finalize(std::vector<uint64_t>& size_being_compacted);
+  // 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);

  // Reference count management (so Versions do not disappear out from
  // under live iterators)
  void Ref();
-  void Unref();
+  // Decrease reference count. Delete the object if no reference left
+  // and return true. Otherwise, return false.
+  bool Unref();

  // Returns true iff some level needs a compaction.
  bool NeedsCompaction() const;
@@ -186,6 +197,12 @@ class Version {
  // Returns the version nuber of this version
  uint64_t GetVersionNumber() const { return version_number_; }

+  // REQUIRES: lock is held
+  // On success, *props will be populated with all SSTables' table properties.
+  // The keys of `props` are the sst file name, the values of `props` are the
+  // tables' propertis, represented as shared_ptr.
+  Status GetPropertiesOfAllTables(TablePropertiesCollection* props);
+
  // used to sort files by size
  struct Fsize {
    int index;
@@ -196,21 +213,28 @@ class Version {
  friend class Compaction;
  friend class VersionSet;
  friend class DBImpl;
+  friend class ColumnFamilyData;
  friend class CompactionPicker;
  friend class LevelCompactionPicker;
  friend class UniversalCompactionPicker;

  class LevelFileNumIterator;
-  Iterator* NewConcatenatingIterator(const ReadOptions&,
-                                     const EnvOptions& soptions,
-                                     int level) const;
-  bool PrefixMayMatch(const ReadOptions& options, const EnvOptions& soptions,
-                      const Slice& internal_prefix, Iterator* level_iter) const;
+  class LevelFileIteratorState;
+
+  bool PrefixMayMatch(const ReadOptions& options, Iterator* level_iter,
+                      const Slice& internal_prefix) const;

  // Sort all files for this version based on their file size and
  // record results in files_by_size_. The largest files are listed first.
  void UpdateFilesBySize();

+  ColumnFamilyData* cfd_;  // ColumnFamilyData to which this Version belongs
+  const InternalKeyComparator* internal_comparator_;
+  const Comparator* user_comparator_;
+  TableCache* table_cache_;
+  const MergeOperator* merge_operator_;
+  Logger* info_log_;
+  Statistics* db_statistics_;
  VersionSet* vset_;            // VersionSet to which this Version belongs
  Version* next_;               // Next version in linked list
  Version* prev_;               // Previous version in linked list
@@ -225,7 +249,7 @@ class Version {
  // but files in each level are now sorted based on file
  // size. The file with the largest size is at the front.
  // This vector stores the index of the file from files_.
-  std::vector< std::vector<int> > files_by_size_;
+  std::vector<std::vector<int>> files_by_size_;

  // An index into files_by_size_ that specifies the first
  // file that is not yet compacted
@@ -256,7 +280,8 @@ class Version {
  // used for debugging and logging purposes only.
  uint64_t version_number_;

-  explicit Version(VersionSet* vset, uint64_t version_number = 0);
+  Version(ColumnFamilyData* cfd, VersionSet* vset, uint64_t version_number = 0);
+  FileIndexer file_indexer_;

  ~Version();

@@ -273,23 +298,34 @@ class Version {

 class VersionSet {
 public:
-  VersionSet(const std::string& dbname, const Options* options,
-             const EnvOptions& storage_options, TableCache* table_cache,
-             const InternalKeyComparator*);
+  VersionSet(const std::string& dbname, const DBOptions* options,
+             const EnvOptions& storage_options, Cache* table_cache);
  ~VersionSet();

  // Apply *edit to the current version to form a new descriptor that
  // is both saved to persistent state and installed as the new
  // current version.  Will release *mu while actually writing to the file.
+  // 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(VersionEdit* edit, port::Mutex* mu,
-                     Directory* db_directory = nullptr,
-                     bool new_descriptor_log = false);
+  Status LogAndApply(ColumnFamilyData* column_family_data, VersionEdit* edit,
+                     port::Mutex* mu, Directory* db_directory = nullptr,
+                     bool new_descriptor_log = false,
+                     const ColumnFamilyOptions* column_family_options =
+                         nullptr);

  // Recover the last saved descriptor from persistent storage.
-  Status Recover();
+  // If read_only == true, Recover() will not complain if some column families
+  // are not opened
+  Status Recover(const std::vector<ColumnFamilyDescriptor>& column_families,
+                 bool read_only = false);

+  // Reads a manifest file and returns a list of column families in
+  // column_families.
+  static Status ListColumnFamilies(std::vector<std::string>* column_families,
+                                   const std::string& dbname, Env* env);
+
+#ifndef ROCKSDB_LITE
  // Try to reduce the number of levels. This call is valid when
  // only one level from the new max level to the old
  // max level containing files.
@@ -304,18 +340,19 @@ class VersionSet {
                                     const EnvOptions& storage_options,
                                     int new_levels);

-  // Return the current version.
-  Version* current() const { return current_; }
+  // printf contents (for debugging)
+  Status DumpManifest(Options& options, std::string& manifestFileName,
+                      bool verbose, bool hex = false);

-  // 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_;
-  }
+#endif  // ROCKSDB_LITE

  // Return the current manifest file number
  uint64_t ManifestFileNumber() const { return manifest_file_number_; }

+  uint64_t PendingManifestFileNumber() const {
+    return pending_manifest_file_number_;
+  }
+
  // Allocate and return a new file number
  uint64_t NewFileNumber() { return next_file_number_++; }

@@ -342,37 +379,21 @@ class VersionSet {
  // Mark the specified file number as used.
  void MarkFileNumberUsed(uint64_t number);

-  // Return the current log file number.
-  uint64_t LogNumber() const { return log_number_; }
-
  // Return the log file number for the log file that is currently
  // being compacted, or zero if there is no such log file.
  uint64_t PrevLogNumber() const { return prev_log_number_; }

-  int NumberLevels() const { return num_levels_; }
-
-  // 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.
-  Compaction* PickCompaction();
-
-  // Return a compaction object for compacting the range [begin,end] in
-  // the specified level.  Returns nullptr if there is nothing in that
-  // level that overlaps the specified range.  Caller should delete
-  // the result.
-  //
-  // The returned Compaction might not include the whole requested range.
-  // In that case, compaction_end will be set to the next key that needs
-  // compacting. In case the compaction will compact the whole range,
-  // compaction_end will be set to nullptr.
-  // Client is responsible for compaction_end storage -- when called,
-  // *compaction_end should point to valid InternalKey!
-  Compaction* CompactRange(int input_level,
-                           int output_level,
-                           const InternalKey* begin,
-                           const InternalKey* end,
-                           InternalKey** compaction_end);
+  // Returns the minimum log number such that all
+  // log numbers less than or equal to it can be deleted
+  uint64_t MinLogNumber() const {
+    uint64_t min_log_num = std::numeric_limits<uint64_t>::max();
+    for (auto cfd : *column_family_set_) {
+      if (min_log_num > cfd->GetLogNumber()) {
+        min_log_num = cfd->GetLogNumber();
+      }
+    }
+    return min_log_num;
+  }

  // Create an iterator that reads over the compaction inputs for "*c".
  // The caller should delete the iterator when no longer needed.
@@ -385,10 +406,6 @@ class VersionSet {
  // "key" as of version "v".
  uint64_t ApproximateOffsetOf(Version* v, const InternalKey& key);

-  // printf contents (for debugging)
-  Status DumpManifest(Options& options, std::string& manifestFileName,
-                      bool verbose, bool hex = false);
-
  // Return the size of the current manifest file
  uint64_t ManifestFileSize() const { return manifest_file_size_; }

@@ -398,60 +415,53 @@ class VersionSet {
  // pick the same files to compact.
  bool VerifyCompactionFileConsistency(Compaction* c);

-  double MaxBytesForLevel(int level);
-
-  // Get the max file size in a given level.
-  uint64_t MaxFileSizeForLevel(int level);
-
-  void ReleaseCompactionFiles(Compaction* c, Status status);
-
-  Status GetMetadataForFile(
-    uint64_t number, int *filelevel, FileMetaData *metadata);
+  Status GetMetadataForFile(uint64_t number, int* filelevel,
+                            FileMetaData** metadata, ColumnFamilyData** cfd);

  void GetLiveFilesMetaData(
    std::vector<LiveFileMetaData> *metadata);

  void GetObsoleteFiles(std::vector<FileMetaData*>* files);

+  ColumnFamilySet* GetColumnFamilySet() { return column_family_set_.get(); }
+
 private:
  class Builder;
  struct ManifestWriter;

-  friend class Compaction;
  friend class Version;

+  struct LogReporter : public log::Reader::Reporter {
+    Status* status;
+    virtual void Corruption(size_t bytes, const Status& s) {
+      if (this->status->ok()) *this->status = s;
+    }
+  };
+
  // Save current contents to *log
  Status WriteSnapshot(log::Writer* log);

-  void AppendVersion(Version* v);
+  void AppendVersion(ColumnFamilyData* column_family_data, Version* v);

-  bool ManifestContains(const std::string& record) const;
+  bool ManifestContains(uint64_t manifest_file_number,
+                        const std::string& record) const;
+
+  ColumnFamilyData* CreateColumnFamily(const ColumnFamilyOptions& options,
+                                       VersionEdit* edit);
+
+  std::unique_ptr<ColumnFamilySet> column_family_set_;

  Env* const env_;
  const std::string dbname_;
-  const Options* const options_;
-  TableCache* const table_cache_;
-  const InternalKeyComparator icmp_;
+  const DBOptions* const options_;
  uint64_t next_file_number_;
  uint64_t manifest_file_number_;
+  uint64_t pending_manifest_file_number_;
  std::atomic<uint64_t> last_sequence_;
-  uint64_t log_number_;
  uint64_t prev_log_number_;  // 0 or backing store for memtable being compacted

-  int num_levels_;
-
  // Opened lazily
  unique_ptr<log::Writer> descriptor_log_;
-  Version dummy_versions_;  // Head of circular doubly-linked list of versions.
-  Version* current_;        // == dummy_versions_.prev_
-
-  // A flag indicating whether we should delay writes because
-  // we have too many level 0 files
-  bool need_slowdown_for_num_level0_files_;
-
-  // An object that keeps all the compaction stats
-  // and picks the next compaction
-  std::unique_ptr<CompactionPicker> compaction_picker_;

  // generates a increasing version number for every new version
  uint64_t current_version_number_;
@@ -475,8 +485,9 @@ class VersionSet {
  VersionSet(const VersionSet&);
  void operator=(const VersionSet&);

-  void LogAndApplyHelper(Builder*b, Version* v,
-                           VersionEdit* edit, port::Mutex* mu);
+  void LogAndApplyCFHelper(VersionEdit* edit);
+  void LogAndApplyHelper(ColumnFamilyData* cfd, Builder* b, Version* v,
+                         VersionEdit* edit, port::Mutex* mu);
 };

 }  // namespace rocksdb
@@ -15,6 +15,9 @@
 //    kTypeValue varstring varstring
 //    kTypeMerge varstring varstring
 //    kTypeDeletion varstring
+//    kTypeColumnFamilyValue varint32 varstring varstring
+//    kTypeColumnFamilyMerge varint32 varstring varstring
+//    kTypeColumnFamilyDeletion varint32 varstring varstring
 // varstring :=
 //    len: varint32
 //    data: uint8[len]
@@ -45,10 +48,20 @@ WriteBatch::~WriteBatch() { }

 WriteBatch::Handler::~Handler() { }

+void WriteBatch::Handler::Put(const Slice& key, const Slice& value) {
+  // you need to either implement Put or PutCF
+  throw std::runtime_error("Handler::Put not implemented!");
+}
+
 void WriteBatch::Handler::Merge(const Slice& key, const Slice& value) {
  throw std::runtime_error("Handler::Merge not implemented!");
 }

+void WriteBatch::Handler::Delete(const Slice& key) {
+  // you need to either implement Delete or DeleteCF
+  throw std::runtime_error("Handler::Delete not implemented!");
+}
+
 void WriteBatch::Handler::LogData(const Slice& blob) {
  // If the user has not specified something to do with blobs, then we ignore
  // them.
@@ -76,31 +89,48 @@ Status WriteBatch::Iterate(Handler* handler) const {
  input.remove_prefix(kHeader);
  Slice key, value, blob;
  int found = 0;
-  while (!input.empty() && handler->Continue()) {
+  Status s;
+  while (s.ok() && !input.empty() && handler->Continue()) {
    char tag = input[0];
    input.remove_prefix(1);
+    uint32_t 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)) {
-          handler->Put(key, value);
+          s = handler->PutCF(column_family, key, value);
          found++;
        } else {
          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)) {
-          handler->Delete(key);
+          s = handler->DeleteCF(column_family, key);
          found++;
        } else {
          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)) {
-          handler->Merge(key, value);
+          s = handler->MergeCF(column_family, key, value);
          found++;
        } else {
          return Status::Corruption("bad WriteBatch Merge");
@@ -117,6 +147,9 @@ Status WriteBatch::Iterate(Handler* handler) const {
        return Status::Corruption("unknown WriteBatch tag");
    }
  }
+  if (!s.ok()) {
+    return s;
+  }
  if (found != WriteBatchInternal::Count(this)) {
    return Status::Corruption("WriteBatch has wrong count");
  } else {
@@ -140,31 +173,85 @@ void WriteBatchInternal::SetSequence(WriteBatch* b, SequenceNumber seq) {
  EncodeFixed64(&b->rep_[0], seq);
 }

-void WriteBatch::Put(const Slice& key, const Slice& value) {
-  WriteBatchInternal::SetCount(this, WriteBatchInternal::Count(this) + 1);
-  rep_.push_back(static_cast<char>(kTypeValue));
-  PutLengthPrefixedSlice(&rep_, key);
-  PutLengthPrefixedSlice(&rep_, value);
+void WriteBatchInternal::Put(WriteBatch* b, uint32_t column_family_id,
+                             const Slice& key, const Slice& value) {
+  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
+  if (column_family_id == 0) {
+    b->rep_.push_back(static_cast<char>(kTypeValue));
+  } else {
+    b->rep_.push_back(static_cast<char>(kTypeColumnFamilyValue));
+    PutVarint32(&b->rep_, column_family_id);
+  }
+  PutLengthPrefixedSlice(&b->rep_, key);
+  PutLengthPrefixedSlice(&b->rep_, value);
 }

-void WriteBatch::Put(const SliceParts& key, const SliceParts& value) {
-  WriteBatchInternal::SetCount(this, WriteBatchInternal::Count(this) + 1);
-  rep_.push_back(static_cast<char>(kTypeValue));
-  PutLengthPrefixedSliceParts(&rep_, key);
-  PutLengthPrefixedSliceParts(&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);
 }

-void WriteBatch::Delete(const Slice& key) {
-  WriteBatchInternal::SetCount(this, WriteBatchInternal::Count(this) + 1);
-  rep_.push_back(static_cast<char>(kTypeDeletion));
-  PutLengthPrefixedSlice(&rep_, key);
+void WriteBatchInternal::Put(WriteBatch* b, uint32_t column_family_id,
+                             const SliceParts& key, const SliceParts& value) {
+  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
+  if (column_family_id == 0) {
+    b->rep_.push_back(static_cast<char>(kTypeValue));
+  } else {
+    b->rep_.push_back(static_cast<char>(kTypeColumnFamilyValue));
+    PutVarint32(&b->rep_, column_family_id);
+  }
+  PutLengthPrefixedSliceParts(&b->rep_, key);
+  PutLengthPrefixedSliceParts(&b->rep_, value);
 }

-void WriteBatch::Merge(const Slice& key, const Slice& value) {
-  WriteBatchInternal::SetCount(this, WriteBatchInternal::Count(this) + 1);
-  rep_.push_back(static_cast<char>(kTypeMerge));
-  PutLengthPrefixedSlice(&rep_, key);
-  PutLengthPrefixedSlice(&rep_, value);
+void WriteBatch::Put(ColumnFamilyHandle* column_family, const SliceParts& key,
+                     const SliceParts& value) {
+  WriteBatchInternal::Put(this, GetColumnFamilyID(column_family), key, value);
+}
+
+void WriteBatchInternal::Delete(WriteBatch* b, uint32_t column_family_id,
+                                const Slice& key) {
+  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
+  if (column_family_id == 0) {
+    b->rep_.push_back(static_cast<char>(kTypeDeletion));
+  } else {
+    b->rep_.push_back(static_cast<char>(kTypeColumnFamilyDeletion));
+    PutVarint32(&b->rep_, column_family_id);
+  }
+  PutLengthPrefixedSlice(&b->rep_, key);
+}
+
+void WriteBatch::Delete(ColumnFamilyHandle* column_family, const Slice& key) {
+  WriteBatchInternal::Delete(this, GetColumnFamilyID(column_family), key);
+}
+
+void WriteBatchInternal::Merge(WriteBatch* b, uint32_t column_family_id,
+                               const Slice& key, const Slice& value) {
+  WriteBatchInternal::SetCount(b, WriteBatchInternal::Count(b) + 1);
+  if (column_family_id == 0) {
+    b->rep_.push_back(static_cast<char>(kTypeMerge));
+  } else {
+    b->rep_.push_back(static_cast<char>(kTypeColumnFamilyMerge));
+    PutVarint32(&b->rep_, column_family_id);
+  }
+  PutLengthPrefixedSlice(&b->rep_, key);
+  PutLengthPrefixedSlice(&b->rep_, value);
+}
+
+void WriteBatch::Merge(ColumnFamilyHandle* column_family, const Slice& key,
+                       const Slice& value) {
+  WriteBatchInternal::Merge(this, GetColumnFamilyID(column_family), key, value);
 }

 void WriteBatch::PutLogData(const Slice& blob) {
@@ -176,45 +263,128 @@ namespace {
 class MemTableInserter : public WriteBatch::Handler {
 public:
  SequenceNumber sequence_;
-  MemTable* mem_;
-  const Options* options_;
+  ColumnFamilyMemTables* cf_mems_;
+  bool recovery_;
+  uint64_t log_number_;
  DBImpl* db_;
-  const bool filter_deletes_;
+  const bool dont_filter_deletes_;

-  MemTableInserter(SequenceNumber sequence, MemTable* mem, const Options* opts,
-                   DB* db, const bool filter_deletes)
-    : sequence_(sequence),
-      mem_(mem),
-      options_(opts),
-      db_(reinterpret_cast<DBImpl*>(db)),
-      filter_deletes_(filter_deletes) {
-    assert(mem_);
-    if (filter_deletes_) {
-      assert(options_);
+  MemTableInserter(SequenceNumber sequence, ColumnFamilyMemTables* cf_mems,
+                   bool recovery, uint64_t log_number, DB* db,
+                   const bool dont_filter_deletes)
+      : sequence_(sequence),
+        cf_mems_(cf_mems),
+        recovery_(recovery),
+        log_number_(log_number),
+        db_(reinterpret_cast<DBImpl*>(db)),
+        dont_filter_deletes_(dont_filter_deletes) {
+    assert(cf_mems);
+    if (!dont_filter_deletes_) {
      assert(db_);
    }
  }

-  virtual void Put(const Slice& key, const Slice& value) {
-    if (options_->inplace_update_support
-        && mem_->Update(sequence_, kTypeValue, key, value)) {
-      RecordTick(options_->statistics.get(), NUMBER_KEYS_UPDATED);
-    } else {
-      mem_->Add(sequence_, kTypeValue, key, value);
+  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 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
+      // update
+      *s = Status::OK();
+      return false;
    }
-    sequence_++;
+    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 void Merge(const Slice& key, const Slice& value) {
+
+  virtual Status PutCF(uint32_t column_family_id, const Slice& key,
+                       const Slice& value) {
+    Status seek_status;
+    if (!SeekToColumnFamily(column_family_id, &seek_status)) {
+      ++sequence_;
+      return seek_status;
+    }
+    MemTable* mem = cf_mems_->GetMemTable();
+    const Options* options = cf_mems_->GetOptions();
+    if (!options->inplace_update_support) {
+      mem->Add(sequence_, kTypeValue, key, value);
+    } else if (options->inplace_callback == nullptr) {
+      mem->Update(sequence_, key, value);
+      RecordTick(options->statistics.get(), NUMBER_KEYS_UPDATED);
+    } else {
+      if (mem->UpdateCallback(sequence_, key, value, *options)) {
+      } else {
+        // key not found in memtable. Do sst get, update, add
+        SnapshotImpl read_from_snapshot;
+        read_from_snapshot.number_ = sequence_;
+        ReadOptions ropts;
+        ropts.snapshot = &read_from_snapshot;
+
+        std::string prev_value;
+        std::string merged_value;
+
+        auto cf_handle = cf_mems_->GetColumnFamilyHandle();
+        if (cf_handle == nullptr) {
+          cf_handle = db_->DefaultColumnFamily();
+        }
+        Status s = db_->Get(ropts, cf_handle, key, &prev_value);
+
+        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);
+        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);
+        } 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);
+        }
+      }
+    }
+    // Since all Puts are logged in trasaction logs (if enabled), always bump
+    // sequence number. Even if the update eventually fails and does not result
+    // in memtable add/update.
+    sequence_++;
+    return Status::OK();
+  }
+
+  virtual Status MergeCF(uint32_t column_family_id, const Slice& key,
+                         const Slice& value) {
+    Status seek_status;
+    if (!SeekToColumnFamily(column_family_id, &seek_status)) {
+      ++sequence_;
+      return seek_status;
+    }
+    MemTable* mem = cf_mems_->GetMemTable();
+    const Options* options = cf_mems_->GetOptions();
    bool perform_merge = false;

-    if (options_->max_successive_merges > 0 && db_ != nullptr) {
+    if (options->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);
+      size_t num_merges = mem->CountSuccessiveMergeEntries(lkey);

-      if (num_merges >= options_->max_successive_merges) {
+      if (num_merges >= options->max_successive_merges) {
        perform_merge = true;
      }
    }
@@ -230,62 +400,78 @@ class MemTableInserter : public WriteBatch::Handler {
      ReadOptions read_options;
      read_options.snapshot = &read_from_snapshot;

-      db_->Get(read_options, key, &get_value);
+      auto cf_handle = cf_mems_->GetColumnFamilyHandle();
+      if (cf_handle == nullptr) {
+        cf_handle = db_->DefaultColumnFamily();
+      }
+      db_->Get(read_options, cf_handle, key, &get_value);
      Slice get_value_slice = Slice(get_value);

      // 2) Apply this merge
-      auto merge_operator = options_->merge_operator.get();
+      auto merge_operator = options->merge_operator.get();
      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())) {
+      if (!merge_operator->FullMerge(key, &get_value_slice, operands,
+                                     &new_value, options->info_log.get())) {
          // Failed to merge!
-          RecordTick(options_->statistics.get(), NUMBER_MERGE_FAILURES);
+        RecordTick(options->statistics.get(), NUMBER_MERGE_FAILURES);

          // Store the delta in memtable
          perform_merge = false;
      } else {
        // 3) Add value to memtable
-        mem_->Add(sequence_, kTypeValue, key, new_value);
+        mem->Add(sequence_, kTypeValue, key, new_value);
      }
    }

    if (!perform_merge) {
      // Add merge operator to memtable
-      mem_->Add(sequence_, kTypeMerge, key, value);
+      mem->Add(sequence_, kTypeMerge, key, value);
    }

    sequence_++;
+    return Status::OK();
  }
-  virtual void Delete(const Slice& key) {
-    if (filter_deletes_) {
+
+  virtual Status DeleteCF(uint32_t column_family_id, const Slice& key) {
+    Status seek_status;
+    if (!SeekToColumnFamily(column_family_id, &seek_status)) {
+      ++sequence_;
+      return seek_status;
+    }
+    MemTable* mem = cf_mems_->GetMemTable();
+    const Options* options = cf_mems_->GetOptions();
+    if (!dont_filter_deletes_ && options->filter_deletes) {
      SnapshotImpl read_from_snapshot;
      read_from_snapshot.number_ = sequence_;
      ReadOptions ropts;
      ropts.snapshot = &read_from_snapshot;
      std::string value;
-      if (!db_->KeyMayExist(ropts, key, &value)) {
-        RecordTick(options_->statistics.get(), NUMBER_FILTERED_DELETES);
-        return;
+      auto cf_handle = cf_mems_->GetColumnFamilyHandle();
+      if (cf_handle == nullptr) {
+        cf_handle = db_->DefaultColumnFamily();
+      }
+      if (!db_->KeyMayExist(ropts, cf_handle, key, &value)) {
+        RecordTick(options->statistics.get(), NUMBER_FILTERED_DELETES);
+        return Status::OK();
      }
    }
-    mem_->Add(sequence_, kTypeDeletion, key, Slice());
+    mem->Add(sequence_, kTypeDeletion, key, Slice());
    sequence_++;
+    return Status::OK();
  }
 };
 }  // namespace

-Status WriteBatchInternal::InsertInto(const WriteBatch* b, MemTable* mem,
-                                      const Options* opts, DB* db,
-                                      const bool filter_deletes) {
-  MemTableInserter inserter(WriteBatchInternal::Sequence(b), mem, opts, db,
-                            filter_deletes);
+Status WriteBatchInternal::InsertInto(const WriteBatch* b,
+                                      ColumnFamilyMemTables* memtables,
+                                      bool recovery, uint64_t log_number,
+                                      DB* db, const bool dont_filter_deletes) {
+  MemTableInserter inserter(WriteBatchInternal::Sequence(b), memtables,
+                            recovery, log_number, db, dont_filter_deletes);
  return b->Iterate(&inserter);
 }

@@ -17,10 +17,66 @@ namespace rocksdb {

 class MemTable;

+class ColumnFamilyMemTables {
+ public:
+  virtual ~ColumnFamilyMemTables() {}
+  virtual bool Seek(uint32_t column_family_id) = 0;
+  // returns true if the update to memtable should be ignored
+  // (useful when recovering from log whose updates have already
+  // been processed)
+  virtual uint64_t GetLogNumber() const = 0;
+  virtual MemTable* GetMemTable() const = 0;
+  virtual const Options* GetOptions() const = 0;
+  virtual ColumnFamilyHandle* GetColumnFamilyHandle() = 0;
+};
+
+class ColumnFamilyMemTablesDefault : public ColumnFamilyMemTables {
+ public:
+  ColumnFamilyMemTablesDefault(MemTable* mem, const Options* options)
+      : ok_(false), mem_(mem), options_(options) {}
+
+  bool Seek(uint32_t column_family_id) override {
+    ok_ = (column_family_id == 0);
+    return ok_;
+  }
+
+  uint64_t GetLogNumber() const override { return 0; }
+
+  MemTable* GetMemTable() const override {
+    assert(ok_);
+    return mem_;
+  }
+
+  const Options* GetOptions() const override {
+    assert(ok_);
+    return options_;
+  }
+
+  ColumnFamilyHandle* GetColumnFamilyHandle() override { return nullptr; }
+
+ private:
+  bool ok_;
+  MemTable* mem_;
+  const Options* const options_;
+};
+
 // WriteBatchInternal provides static methods for manipulating a
 // WriteBatch that we don't want in the public WriteBatch interface.
 class WriteBatchInternal {
 public:
+  // WriteBatch methods with column_family_id instead of ColumnFamilyHandle*
+  static void Put(WriteBatch* batch, uint32_t column_family_id,
+                  const Slice& key, const Slice& value);
+
+  static void Put(WriteBatch* batch, uint32_t column_family_id,
+                  const SliceParts& key, const SliceParts& value);
+
+  static void Delete(WriteBatch* batch, uint32_t column_family_id,
+                     const Slice& key);
+
+  static void Merge(WriteBatch* batch, uint32_t column_family_id,
+                    const Slice& key, const Slice& value);
+
  // Return the number of entries in the batch.
  static int Count(const WriteBatch* batch);

@@ -45,11 +101,21 @@ class WriteBatchInternal {
  static void SetContents(WriteBatch* batch, const Slice& contents);

  // Inserts batch entries into memtable
-  // Drops deletes in batch if filter_del is set to true and
-  // db->KeyMayExist returns false
-  static Status InsertInto(const WriteBatch* batch, MemTable* memtable,
-                           const Options* opts, DB* db = nullptr,
-                           const bool filter_del = false);
+  // 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
+  // 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,
+                           const bool dont_filter_deletes = true);

  static void Append(WriteBatch* dst, const WriteBatch* src);
 };
@@ -11,6 +11,7 @@

 #include <memory>
 #include "db/memtable.h"
+#include "db/column_family.h"
 #include "db/write_batch_internal.h"
 #include "rocksdb/env.h"
 #include "rocksdb/memtablerep.h"
@@ -27,9 +28,10 @@ static std::string PrintContents(WriteBatch* b) {
  MemTable* mem = new MemTable(cmp, options);
  mem->Ref();
  std::string state;
-  Status s = WriteBatchInternal::InsertInto(b, mem, &options);
+  ColumnFamilyMemTablesDefault cf_mems_default(mem, &options);
+  Status s = WriteBatchInternal::InsertInto(b, &cf_mems_default);
  int count = 0;
-  Iterator* iter = mem->NewIterator();
+  Iterator* iter = mem->NewIterator(ReadOptions());
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
    ParsedInternalKey ikey;
    memset((void *)&ikey, 0, sizeof(ikey));
@@ -57,7 +59,7 @@ static std::string PrintContents(WriteBatch* b) {
        state.append(")");
        count++;
        break;
-      case kTypeLogData:
+      default:
        assert(false);
        break;
    }
@@ -144,17 +146,37 @@ TEST(WriteBatchTest, Append) {
 namespace {
  struct TestHandler : public WriteBatch::Handler {
    std::string seen;
-    virtual void Put(const Slice& key, const Slice& value) {
-      seen += "Put(" + key.ToString() + ", " + value.ToString() + ")";
+    virtual Status PutCF(uint32_t column_family_id, const Slice& key,
+                       const Slice& value) {
+      if (column_family_id == 0) {
+        seen += "Put(" + key.ToString() + ", " + value.ToString() + ")";
+      } else {
+        seen += "PutCF(" + std::to_string(column_family_id) + ", " +
+                key.ToString() + ", " + value.ToString() + ")";
+      }
+      return Status::OK();
    }
-    virtual void Merge(const Slice& key, const Slice& value) {
-      seen += "Merge(" + key.ToString() + ", " + value.ToString() + ")";
+    virtual Status MergeCF(uint32_t column_family_id, const Slice& key,
+                         const Slice& value) {
+      if (column_family_id == 0) {
+        seen += "Merge(" + key.ToString() + ", " + value.ToString() + ")";
+      } else {
+        seen += "MergeCF(" + std::to_string(column_family_id) + ", " +
+                key.ToString() + ", " + value.ToString() + ")";
+      }
+      return Status::OK();
    }
    virtual void LogData(const Slice& blob) {
      seen += "LogData(" + blob.ToString() + ")";
    }
-    virtual void Delete(const Slice& key) {
-      seen += "Delete(" + key.ToString() + ")";
+    virtual Status DeleteCF(uint32_t column_family_id, const Slice& key) {
+      if (column_family_id == 0) {
+        seen += "Delete(" + key.ToString() + ")";
+      } else {
+        seen += "DeleteCF(" + std::to_string(column_family_id) + ", " +
+                key.ToString() + ")";
+      }
+      return Status::OK();
    }
  };
 }
@@ -194,21 +216,23 @@ TEST(WriteBatchTest, Continue) {

  struct Handler : public TestHandler {
    int num_seen = 0;
-    virtual void Put(const Slice& key, const Slice& value) {
+    virtual Status PutCF(uint32_t column_family_id, const Slice& key,
+                       const Slice& value) {
      ++num_seen;
-      TestHandler::Put(key, value);
+      return TestHandler::PutCF(column_family_id, key, value);
    }
-    virtual void Merge(const Slice& key, const Slice& value) {
+    virtual Status MergeCF(uint32_t column_family_id, const Slice& key,
+                         const Slice& value) {
      ++num_seen;
-      TestHandler::Merge(key, value);
+      return TestHandler::MergeCF(column_family_id, key, value);
    }
    virtual void LogData(const Slice& blob) {
      ++num_seen;
      TestHandler::LogData(blob);
    }
-    virtual void Delete(const Slice& key) {
+    virtual Status DeleteCF(uint32_t column_family_id, const Slice& key) {
      ++num_seen;
-      TestHandler::Delete(key);
+      return TestHandler::DeleteCF(column_family_id, key);
    }
    virtual bool Continue() override {
      return num_seen < 3;
@@ -256,6 +280,42 @@ TEST(WriteBatchTest, PutGatherSlices) {
  ASSERT_EQ(3, batch.Count());
 }

+namespace {
+class ColumnFamilyHandleImplDummy : public ColumnFamilyHandleImpl {
+ public:
+  explicit ColumnFamilyHandleImplDummy(int id)
+      : ColumnFamilyHandleImpl(nullptr, nullptr, nullptr), id_(id) {}
+  uint32_t GetID() const override { return id_; }
+
+ private:
+  uint32_t id_;
+};
+}  // namespace anonymous
+
+TEST(WriteBatchTest, ColumnFamiliesBatchTest) {
+  WriteBatch 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"));
+
+  TestHandler handler;
+  batch.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) {
@@ -609,13 +609,13 @@ key-value during background compaction.
 </ul>
 <p>
 Other options impacting performance of compactions and when they get triggered
-are: 
+are:
 <ul>
 <p>
-<li> <code>Options::access_hint_on_compaction_start</code> - Specify the file access 
+<li> <code>Options::access_hint_on_compaction_start</code> - Specify the file access
 pattern once a compaction is started. It will be applied to all input files of a compaction. Default: NORMAL
 <p>
-<li> <code>Options::level0_file_num_compaction_trigger</code> -  Number of files to trigger level-0 compaction. 
+<li> <code>Options::level0_file_num_compaction_trigger</code> -  Number of files to trigger level-0 compaction.
 A negative value means that level-0 compaction will not be triggered by number of files at all.
 <p>
 <li> <code>Options::max_mem_compaction_level</code> -  Maximum level to which a new compacted memtable is pushed if it
@@ -623,7 +623,7 @@ does not create overlap.  We try to push to level 2 to avoid the relatively expe
 expensive manifest file operations.  We do not push all the way to the largest level since that can generate a lot of wasted disk
 space if the same key space is being repeatedly overwritten.
 <p>
-<li> <code>Options::target_file_size_base</code> and <code>Options::target_file_size_multiplier</code> - 
+<li> <code>Options::target_file_size_base</code> and <code>Options::target_file_size_multiplier</code> -
 Target file size for compaction.  target_file_size_base is per-file size for level-1.
 Target file size for level L can be calculated by target_file_size_base * (target_file_size_multiplier ^ (L-1))
 For example, if target_file_size_base is 2MB and target_file_size_multiplier is 10, then each file on level-1 will
@@ -698,7 +698,7 @@ Default: kCompactionStopStyleTotalSize
 A thread pool is associated with Env environment object. The client has to create a thread pool by setting the number of background
 threads using method <code>Env::SetBackgroundThreads()</code> defined in <code>rocksdb/env.h</code>.
 We use the thread pool for compactions and memtable flushes.
-Since memtable flushes are in critical code path (stalling memtable flush can stall writes, increasing p99), we suggest 
+Since memtable flushes are in critical code path (stalling memtable flush can stall writes, increasing p99), we suggest
 having two thread pools - with priorities HIGH and LOW. Memtable flushes can be set up to be scheduled on HIGH thread pool.
 There are two options available for configuration of background compactions and flushes:
 <ul>
@@ -779,9 +779,9 @@ implementation.  See <code>rocksdb/util/env_posix.h</code> for an example.
 <p>
 To be able to efficiently tune your application, it is always helpful if you
 have access to usage statistics. You can collect those statistics by setting
-<code>Options::table_stats_collectors</code> or
+<code>Options::table_properties_collectors</code> or
 <code>Options::statistics</code>. For more information, refer to
-<code>rocksdb/table_stats.h</code> and <code>rocksdb/statistics.h</code>.
+<code>rocksdb/table_properties.h</code> and <code>rocksdb/statistics.h</code>.
 These should not add significant overhead to your application and we
 recommend exporting them to other monitoring tools.

@@ -47,7 +47,7 @@ private:
 class HdfsEnv : public Env {

 public:
-  HdfsEnv(const std::string& fsname) : fsname_(fsname) {
+  explicit HdfsEnv(const std::string& fsname) : fsname_(fsname) {
    posixEnv = Env::Default();
    fileSys_ = connectToPath(fsname_);
  }
@@ -108,6 +108,13 @@ class HdfsEnv : public Env {
    posixEnv->StartThread(function, arg);
  }

+  virtual void WaitForJoin() { posixEnv->WaitForJoin(); }
+
+  virtual unsigned int GetThreadPoolQueueLen(Priority pri = LOW) const
+      override {
+    return posixEnv->GetThreadPoolQueueLen(pri);
+  }
+
  virtual Status GetTestDirectory(std::string* path) {
    return posixEnv->GetTestDirectory(path);
  }
@@ -161,7 +168,7 @@ class HdfsEnv : public Env {
   */
  hdfsFS connectToPath(const std::string& uri) {
    if (uri.empty()) {
-      return NULL;
+      return nullptr;
    }
    if (uri.find(kProto) != 0) {
      // uri doesn't start with hdfs:// -> use default:0, which is special
@@ -218,10 +225,10 @@ static const Status notsup;
 class HdfsEnv : public Env {

 public:
-  HdfsEnv(const std::string& fsname) {
+  explicit HdfsEnv(const std::string& fsname) {
    fprintf(stderr, "You have not build rocksdb with HDFS support\n");
    fprintf(stderr, "Please see hdfs/README for details\n");
-    throw new std::exception();
+    throw std::exception();
  }

  virtual ~HdfsEnv() {
@@ -288,6 +295,12 @@ class HdfsEnv : public Env {

  virtual void StartThread(void (*function)(void* arg), void* arg) {}

+  virtual void WaitForJoin() {}
+
+  virtual unsigned int GetThreadPoolQueueLen(Priority pri = LOW) const {
+    return 0;
+  }
+
  virtual Status GetTestDirectory(std::string* path) {return notsup;}

  virtual uint64_t NowMicros() {return 0;}
@@ -2,8 +2,6 @@
 // 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 "helpers/memenv/memenv.h"
-
 #include "rocksdb/env.h"
 #include "rocksdb/status.h"
 #include "port/port.h"
@@ -1,19 +0,0 @@
-// 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 STORAGE_ROCKSDB_HELPERS_MEMENV_MEMENV_H_
-#define STORAGE_ROCKSDB_HELPERS_MEMENV_MEMENV_H_
-namespace rocksdb {
-
-class Env;
-
-// Returns a new environment that stores its data in memory and delegates
-// all non-file-storage tasks to base_env. The caller must delete the result
-// when it is no longer needed.
-// *base_env must remain live while the result is in use.
-Env* NewMemEnv(Env* base_env);
-
-}  // namespace rocksdb
-
-#endif  // STORAGE_ROCKSDB_HELPERS_MEMENV_MEMENV_H_
@@ -2,8 +2,6 @@
 // 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 "helpers/memenv/memenv.h"
-
 #include "db/db_impl.h"
 #include "rocksdb/db.h"
 #include "rocksdb/env.h"
@@ -1,45 +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.
-//
-// Arena class defines memory allocation methods. It's used by memtable and
-// skiplist.
-
-#ifndef STORAGE_ROCKSDB_INCLUDE_ARENA_H_
-#define STORAGE_ROCKSDB_INCLUDE_ARENA_H_
-
-#include <limits>
-#include <memory>
-
-namespace rocksdb {
-
-class Arena {
- public:
-  Arena() {};
-  virtual ~Arena() {};
-
-  // Return a pointer to a newly allocated memory block of "bytes" bytes.
-  virtual char* Allocate(size_t bytes) = 0;
-
-  // Allocate memory with the normal alignment guarantees provided by malloc.
-  virtual char* AllocateAligned(size_t bytes) = 0;
-
-  // Returns an estimate of the total memory used by arena.
-  virtual const size_t ApproximateMemoryUsage() = 0;
-
-  // Returns the total number of bytes in all blocks allocated so far.
-  virtual const size_t MemoryAllocatedBytes() = 0;
-
- private:
-  // No copying allowed
-  Arena(const Arena&);
-  void operator=(const Arena&);
-};
-
-}  // namespace rocksdb
-
-#endif  // STORAGE_ROCKSDB_INCLUDE_ARENA_H_
@@ -54,23 +54,27 @@ extern "C" {

 /* Exported types */

-typedef struct rocksdb_t               rocksdb_t;
-typedef struct rocksdb_cache_t         rocksdb_cache_t;
-typedef struct rocksdb_comparator_t    rocksdb_comparator_t;
-typedef struct rocksdb_env_t           rocksdb_env_t;
-typedef struct rocksdb_filelock_t      rocksdb_filelock_t;
-typedef struct rocksdb_filterpolicy_t  rocksdb_filterpolicy_t;
-typedef struct rocksdb_iterator_t      rocksdb_iterator_t;
-typedef struct rocksdb_logger_t        rocksdb_logger_t;
-typedef struct rocksdb_options_t       rocksdb_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;
-typedef struct rocksdb_snapshot_t      rocksdb_snapshot_t;
-typedef struct rocksdb_writablefile_t  rocksdb_writablefile_t;
-typedef struct rocksdb_writebatch_t    rocksdb_writebatch_t;
-typedef struct rocksdb_writeoptions_t  rocksdb_writeoptions_t;
+typedef struct rocksdb_t                 rocksdb_t;
+typedef struct rocksdb_cache_t           rocksdb_cache_t;
+typedef struct rocksdb_comparator_t      rocksdb_comparator_t;
+typedef struct rocksdb_env_t             rocksdb_env_t;
+typedef struct rocksdb_filelock_t        rocksdb_filelock_t;
+typedef struct rocksdb_filterpolicy_t    rocksdb_filterpolicy_t;
+typedef struct rocksdb_flushoptions_t    rocksdb_flushoptions_t;
+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_randomfile_t      rocksdb_randomfile_t;
+typedef struct rocksdb_readoptions_t     rocksdb_readoptions_t;
+typedef struct rocksdb_seqfile_t         rocksdb_seqfile_t;
+typedef struct rocksdb_slicetransform_t  rocksdb_slicetransform_t;
+typedef struct rocksdb_snapshot_t        rocksdb_snapshot_t;
+typedef struct rocksdb_writablefile_t    rocksdb_writablefile_t;
+typedef struct rocksdb_writebatch_t      rocksdb_writebatch_t;
+typedef struct rocksdb_writeoptions_t    rocksdb_writeoptions_t;
 typedef struct rocksdb_universal_compaction_options_t rocksdb_universal_compaction_options_t;
+typedef struct rocksdb_livefiles_t     rocksdb_livefiles_t;

 /* DB operations */

@@ -79,6 +83,12 @@ extern rocksdb_t* rocksdb_open(
    const char* name,
    char** errptr);

+extern rocksdb_t* rocksdb_open_for_read_only(
+    const rocksdb_options_t* options,
+    const char* name,
+    unsigned char error_if_log_file_exist,
+    char** errptr);
+
 extern void rocksdb_close(rocksdb_t* db);

 extern void rocksdb_put(
@@ -94,6 +104,13 @@ extern void rocksdb_delete(
    const char* key, size_t keylen,
    char** errptr);

+extern void rocksdb_merge(
+    rocksdb_t* db,
+    const rocksdb_writeoptions_t* options,
+    const char* key, size_t keylen,
+    const char* val, size_t vallen,
+    char** errptr);
+
 extern void rocksdb_write(
    rocksdb_t* db,
    const rocksdb_writeoptions_t* options,
@@ -138,6 +155,27 @@ extern void rocksdb_compact_range(
    const char* start_key, size_t start_key_len,
    const char* limit_key, size_t limit_key_len);

+extern void rocksdb_delete_file(
+    rocksdb_t* db,
+    const char* name);
+
+extern const rocksdb_livefiles_t* rocksdb_livefiles(
+    rocksdb_t* db);
+
+extern void rocksdb_flush(
+    rocksdb_t* db,
+    const rocksdb_flushoptions_t* options,
+    char** errptr);
+
+extern void rocksdb_disable_file_deletions(
+    rocksdb_t* db,
+    char** errptr);
+
+extern void rocksdb_enable_file_deletions(
+    rocksdb_t* db,
+    unsigned char force,
+    char** errptr);
+
 /* Management operations */

 extern void rocksdb_destroy_db(
@@ -168,10 +206,15 @@ extern void rocksdb_iter_get_error(const rocksdb_iterator_t*, char** errptr);
 extern rocksdb_writebatch_t* rocksdb_writebatch_create();
 extern void rocksdb_writebatch_destroy(rocksdb_writebatch_t*);
 extern void rocksdb_writebatch_clear(rocksdb_writebatch_t*);
+extern int rocksdb_writebatch_count(rocksdb_writebatch_t*);
 extern void rocksdb_writebatch_put(
    rocksdb_writebatch_t*,
    const char* key, size_t klen,
    const char* val, size_t vlen);
+extern void rocksdb_writebatch_merge(
+    rocksdb_writebatch_t*,
+    const char* key, size_t klen,
+    const char* val, size_t vlen);
 extern void rocksdb_writebatch_delete(
    rocksdb_writebatch_t*,
    const char* key, size_t klen);
@@ -180,6 +223,7 @@ extern void rocksdb_writebatch_iterate(
    void* state,
    void (*put)(void*, const char* k, size_t klen, const char* v, size_t vlen),
    void (*deleted)(void*, const char* k, size_t klen));
+extern const char* rocksdb_writebatch_data(rocksdb_writebatch_t*, size_t *size);

 /* Options */

@@ -188,6 +232,8 @@ extern void rocksdb_options_destroy(rocksdb_options_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_compression_per_level(
  rocksdb_options_t* opt,
  int* level_values,
@@ -203,13 +249,18 @@ extern void rocksdb_options_set_paranoid_checks(
    rocksdb_options_t*, unsigned char);
 extern void rocksdb_options_set_env(rocksdb_options_t*, rocksdb_env_t*);
 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);
 extern void rocksdb_options_set_level0_file_num_compaction_trigger(
    rocksdb_options_t*, int);
@@ -217,29 +268,130 @@ extern void rocksdb_options_set_level0_slowdown_writes_trigger(
    rocksdb_options_t*, int);
 extern void rocksdb_options_set_level0_stop_writes_trigger(
    rocksdb_options_t*, int);
+extern void rocksdb_options_set_max_mem_compaction_level(
+    rocksdb_options_t*, int);
 extern void rocksdb_options_set_target_file_size_base(
    rocksdb_options_t*, uint64_t);
 extern void rocksdb_options_set_target_file_size_multiplier(
    rocksdb_options_t*, int);
+extern void rocksdb_options_set_max_bytes_for_level_base(
+    rocksdb_options_t*, uint64_t);
+extern void rocksdb_options_set_max_bytes_for_level_multiplier(
+    rocksdb_options_t*, int);
+extern void rocksdb_options_set_expanded_compaction_factor(
+    rocksdb_options_t*, int);
+extern void rocksdb_options_set_max_grandparent_overlap_factor(
+    rocksdb_options_t*, int);
+extern void rocksdb_options_set_max_bytes_for_level_multiplier_additional(
+    rocksdb_options_t*, int* level_values, size_t num_levels);
+extern void rocksdb_options_enable_statistics(rocksdb_options_t*);
+
 extern void rocksdb_options_set_max_write_buffer_number(rocksdb_options_t*, int);
 extern void rocksdb_options_set_min_write_buffer_number_to_merge(rocksdb_options_t*, int);
 extern void rocksdb_options_set_max_background_compactions(rocksdb_options_t*, int);
 extern void rocksdb_options_set_max_background_flushes(rocksdb_options_t*, int);
+extern void rocksdb_options_set_max_log_file_size(rocksdb_options_t*, size_t);
+extern void rocksdb_options_set_log_file_time_to_roll(rocksdb_options_t*, size_t);
+extern void rocksdb_options_set_keep_log_file_num(rocksdb_options_t*, size_t);
+extern void rocksdb_options_set_soft_rate_limit(rocksdb_options_t*, double);
+extern void rocksdb_options_set_hard_rate_limit(rocksdb_options_t*, double);
+extern void rocksdb_options_set_rate_limit_delay_max_milliseconds(
+    rocksdb_options_t*, unsigned int);
+extern void rocksdb_options_set_max_manifest_file_size(
+    rocksdb_options_t*, size_t);
+extern void rocksdb_options_set_no_block_cache(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_table_cache_numshardbits(
+    rocksdb_options_t*, int);
+extern void rocksdb_options_set_table_cache_remove_scan_count_limit(
+    rocksdb_options_t*, int);
+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(
+    rocksdb_options_t*, const char*);
+extern void rocksdb_options_set_WAL_ttl_seconds(
+    rocksdb_options_t*, uint64_t);
+extern void rocksdb_options_set_WAL_size_limit_MB(
+    rocksdb_options_t*, uint64_t);
+extern void rocksdb_options_set_manifest_preallocation_size(
+    rocksdb_options_t*, size_t);
+extern void rocksdb_options_set_purge_redundant_kvs_while_flush(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_allow_os_buffer(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_allow_mmap_reads(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_allow_mmap_writes(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_is_fd_close_on_exec(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_skip_log_error_on_recovery(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_stats_dump_period_sec(
+    rocksdb_options_t*, unsigned int);
+extern void rocksdb_options_set_block_size_deviation(
+    rocksdb_options_t*, int);
+extern void rocksdb_options_set_advise_random_on_open(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_access_hint_on_compaction_start(
+    rocksdb_options_t*, int);
+extern void rocksdb_options_set_use_adaptive_mutex(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_bytes_per_sync(
+    rocksdb_options_t*, uint64_t);
+extern void rocksdb_options_set_verify_checksums_in_compaction(
+    rocksdb_options_t*, unsigned char);
+extern void rocksdb_options_set_filter_deletes(
+    rocksdb_options_t*, unsigned char);
+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);
 extern void rocksdb_options_prepare_for_bulk_load(rocksdb_options_t*);
 extern void rocksdb_options_set_memtable_vector_rep(rocksdb_options_t*);
+extern void rocksdb_options_set_hash_skip_list_rep(rocksdb_options_t*, size_t, int32_t, int32_t);
+extern void rocksdb_options_set_hash_link_list_rep(rocksdb_options_t*, size_t);
+extern void rocksdb_options_set_plain_table_factory(rocksdb_options_t*, uint32_t, int, double, size_t);

+extern void rocksdb_options_set_max_bytes_for_level_base(rocksdb_options_t* opt, uint64_t n);
+extern void rocksdb_options_set_stats_dump_period_sec(rocksdb_options_t* opt, unsigned int sec);
+
+extern void rocksdb_options_set_min_level_to_compress(rocksdb_options_t* opt, int level);
+
+extern void rocksdb_options_set_memtable_prefix_bloom_bits(
+    rocksdb_options_t*, uint32_t);
+extern void rocksdb_options_set_memtable_prefix_bloom_probes(
+    rocksdb_options_t*, uint32_t);
+extern void rocksdb_options_set_max_successive_merges(
+    rocksdb_options_t*, size_t);
+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(
+    rocksdb_options_t*, size_t);

 enum {
  rocksdb_no_compression = 0,
  rocksdb_snappy_compression = 1,
-  rocksdb_zlib_compression = 1,
-  rocksdb_bz2_compression = 1
+  rocksdb_zlib_compression = 2,
+  rocksdb_bz2_compression = 3,
+  rocksdb_lz4_compression = 4,
+  rocksdb_lz4hc_compression = 5
 };
 extern void rocksdb_options_set_compression(rocksdb_options_t*, int);

@@ -275,12 +427,39 @@ extern rocksdb_filterpolicy_t* rocksdb_filterpolicy_create(
        void*,
        const char* key, size_t length,
        const char* filter, size_t filter_length),
+    void (*delete_filter)(
+        void*,
+        const char* filter, size_t filter_length),
    const char* (*name)(void*));
 extern void rocksdb_filterpolicy_destroy(rocksdb_filterpolicy_t*);

 extern rocksdb_filterpolicy_t* rocksdb_filterpolicy_create_bloom(
    int bits_per_key);

+/* Merge Operator */
+
+extern rocksdb_mergeoperator_t* rocksdb_mergeoperator_create(
+    void* state,
+    void (*destructor)(void*),
+    char* (*full_merge)(
+        void*,
+        const char* key, size_t key_length,
+        const char* existing_value, size_t existing_value_length,
+        const char* const* operands_list, const size_t* operands_list_length,
+        int num_operands,
+        unsigned char* success, size_t* new_value_length),
+    char* (*partial_merge)(
+        void*,
+        const char* key, size_t key_length,
+        const char* const* operands_list, const size_t* operands_list_length,
+        int num_operands,
+        unsigned char* success, size_t* new_value_length),
+    void (*delete_value)(
+        void*,
+        const char* value, size_t value_length),
+    const char* (*name)(void*));
+extern void rocksdb_mergeoperator_destroy(rocksdb_mergeoperator_t*);
+
 /* Read options */

 extern rocksdb_readoptions_t* rocksdb_readoptions_create();
@@ -293,6 +472,10 @@ 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_read_tier(
+    rocksdb_readoptions_t*, int);
+extern void rocksdb_readoptions_set_tailing(
+    rocksdb_readoptions_t*, unsigned char);

 /* Write options */

@@ -302,6 +485,13 @@ extern void rocksdb_writeoptions_set_sync(
    rocksdb_writeoptions_t*, unsigned char);
 extern void rocksdb_writeoptions_disable_WAL(rocksdb_writeoptions_t* opt, int disable);

+/* Flush options */
+
+extern rocksdb_flushoptions_t* rocksdb_flushoptions_create();
+extern void rocksdb_flushoptions_destroy(rocksdb_flushoptions_t*);
+extern void rocksdb_flushoptions_set_wait(
+    rocksdb_flushoptions_t*, unsigned char);
+
 /* Cache */

 extern rocksdb_cache_t* rocksdb_cache_create_lru(size_t capacity);
@@ -314,6 +504,25 @@ extern void rocksdb_env_set_background_threads(rocksdb_env_t* env, int n);
 extern void rocksdb_env_set_high_priority_background_threads(rocksdb_env_t* env, int n);
 extern void rocksdb_env_destroy(rocksdb_env_t*);

+/* SliceTransform */
+
+extern rocksdb_slicetransform_t* rocksdb_slicetransform_create(
+    void* state,
+    void (*destructor)(void*),
+    char* (*transform)(
+        void*,
+        const char* key, size_t length,
+        size_t* dst_length),
+    unsigned char (*in_domain)(
+        void*,
+        const char* key, size_t length),
+    unsigned char (*in_range)(
+        void*,
+        const char* key, size_t length),
+    const char* (*name)(void*));
+extern rocksdb_slicetransform_t* rocksdb_slicetransform_create_fixed_prefix(size_t);
+extern void rocksdb_slicetransform_destroy(rocksdb_slicetransform_t*);
+
 /* Universal Compaction options */

 enum {
@@ -337,6 +546,28 @@ extern void rocksdb_universal_compaction_options_set_stop_style(
 extern void rocksdb_universal_compaction_options_destroy(
  rocksdb_universal_compaction_options_t*);

+extern int rocksdb_livefiles_count(
+  const rocksdb_livefiles_t*);
+extern const char* rocksdb_livefiles_name(
+  const rocksdb_livefiles_t*,
+  int index);
+extern int rocksdb_livefiles_level(
+  const rocksdb_livefiles_t*,
+  int index);
+extern size_t rocksdb_livefiles_size(
+  const rocksdb_livefiles_t*,
+  int index);
+extern const char* rocksdb_livefiles_smallestkey(
+  const rocksdb_livefiles_t*,
+  int index,
+  size_t* size);
+extern const char* rocksdb_livefiles_largestkey(
+  const rocksdb_livefiles_t*,
+  int index,
+  size_t* size);
+extern void rocksdb_livefiles_destroy(
+  const rocksdb_livefiles_t*);
+
 #ifdef __cplusplus
 }  /* end extern "C" */
 #endif
@@ -102,7 +102,10 @@ class Cache {
  virtual uint64_t NewId() = 0;

  // returns the maximum configured capacity of the cache
-  virtual size_t GetCapacity() = 0;
+  virtual size_t GetCapacity() const = 0;
+
+  // returns the memory size for the entries residing in the cache.
+  virtual size_t GetUsage() const = 0;

  // Call this on shutdown if you want to speed it up. Cache will disown
  // any underlying data and will not free it on delete. This call will leak
@@ -113,6 +116,12 @@ class Cache {
    // default implementation is noop
  };

+  // Apply callback to all entries in the cache
+  // If thread_safe is true, it will also lock the accesses. Otherwise, it will
+  // access the cache without the lock held
+  virtual void ApplyToAllCacheEntries(void (*callback)(void*, size_t),
+                                      bool thread_safe) = 0;
+
 private:
  void LRU_Remove(Handle* e);
  void LRU_Append(Handle* e);
@@ -10,21 +10,34 @@
 #define STORAGE_ROCKSDB_INCLUDE_COMPACTION_FILTER_H_

 #include <string>
+#include <vector>

 namespace rocksdb {

 class Slice;
+class SliceTransform;
+
+// Context information of a compaction run
+struct CompactionFilterContext {
+  // Does this compaction run include all data files
+  bool is_full_compaction;
+  // Is this compaction requested by the client (true),
+  // or is it occurring as an automatic compaction process
+  bool is_manual_compaction;
+};

 // CompactionFilter allows an application to modify/delete a key-value at
 // the time of compaction.

 class CompactionFilter {
 public:
-
  // Context information of a compaction run
  struct Context {
    // Does this compaction run include all data files
    bool is_full_compaction;
+    // Is this compaction requested by the client (true),
+    // or is it occurring as an automatic compaction process
+    bool is_manual_compaction;
  };

  virtual ~CompactionFilter() {}
@@ -61,14 +74,47 @@ class CompactionFilter {
  virtual const char* Name() const = 0;
 };

+// CompactionFilterV2 that buffers kv pairs sharing the same prefix and let
+// application layer to make individual decisions for all the kv pairs in the
+// buffer.
+class CompactionFilterV2 {
+ public:
+  virtual ~CompactionFilterV2() {}
+
+  // The compaction process invokes this method for all the kv pairs
+  // sharing the same prefix. It is a "roll-up" version of CompactionFilter.
+  //
+  // Each entry in the return vector indicates if the corresponding kv should
+  // be preserved in the output of this compaction run. The application can
+  // inspect the exisitng values of the keys and make decision based on it.
+  //
+  // When a value is to be preserved, the application has the option
+  // to modify the entry in existing_values and pass it back through an entry
+  // in new_values. A corresponding values_changed entry needs to be set to
+  // true in this case. Note that the new_values vector contains only changed
+  // values, i.e. new_values.size() <= values_changed.size().
+  //
+  typedef std::vector<Slice> SliceVector;
+  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 = 0;
+
+  // Returns a name that identifies this compaction filter.
+  // The name will be printed to LOG file on start up for diagnosis.
+  virtual const char* Name() const = 0;
+};
+
 // Each compaction will create a new CompactionFilter allowing the
 // application to know about different campactions
 class CompactionFilterFactory {
 public:
-  virtual ~CompactionFilterFactory() { };
+  virtual ~CompactionFilterFactory() { }

  virtual std::unique_ptr<CompactionFilter> CreateCompactionFilter(
-    const CompactionFilter::Context& context) = 0;
+      const CompactionFilter::Context& context) = 0;

  // Returns a name that identifies this compaction filter factory.
  virtual const char* Name() const = 0;
@@ -78,8 +124,8 @@ class CompactionFilterFactory {
 // return any filter
 class DefaultCompactionFilterFactory : public CompactionFilterFactory {
 public:
-  virtual std::unique_ptr<CompactionFilter>
-  CreateCompactionFilter(const CompactionFilter::Context& context) override {
+  virtual std::unique_ptr<CompactionFilter> CreateCompactionFilter(
+      const CompactionFilter::Context& context) override {
    return std::unique_ptr<CompactionFilter>(nullptr);
  }

@@ -88,6 +134,65 @@ class DefaultCompactionFilterFactory : public CompactionFilterFactory {
  }
 };

+// Each compaction will create a new CompactionFilterV2
+//
+// CompactionFilterFactoryV2 enables application to specify a prefix and use
+// CompactionFilterV2 to filter kv-pairs in batches. Each batch contains all
+// the kv-pairs sharing the same prefix.
+//
+// This is useful for applications that require grouping kv-pairs in
+// compaction filter to make a purge/no-purge decision. For example, if the
+// key prefix is user id and the rest of key represents the type of value.
+// This batching filter will come in handy if the application's compaction
+// filter requires knowledge of all types of values for any user id.
+//
+class CompactionFilterFactoryV2 {
+ public:
+  // NOTE: CompactionFilterFactoryV2 will not delete prefix_extractor
+  explicit CompactionFilterFactoryV2(const SliceTransform* prefix_extractor)
+    : prefix_extractor_(prefix_extractor) { }
+
+  virtual ~CompactionFilterFactoryV2() { }
+
+  virtual std::unique_ptr<CompactionFilterV2> CreateCompactionFilterV2(
+    const CompactionFilterContext& context) = 0;
+
+  // Returns a name that identifies this compaction filter factory.
+  virtual const char* Name() const = 0;
+
+  const SliceTransform* GetPrefixExtractor() const {
+    return prefix_extractor_;
+  }
+
+  void SetPrefixExtractor(const SliceTransform* prefix_extractor) {
+    prefix_extractor_ = prefix_extractor;
+  }
+
+ private:
+  // Prefix extractor for compaction filter v2
+  // Keys sharing the same prefix will be buffered internally.
+  // Client can implement a Filter callback function to operate on the buffer
+  const SliceTransform* prefix_extractor_;
+};
+
+// Default implementaion of CompactionFilterFactoryV2 which does not
+// return any filter
+class DefaultCompactionFilterFactoryV2 : public CompactionFilterFactoryV2 {
+ public:
+  explicit DefaultCompactionFilterFactoryV2()
+      : CompactionFilterFactoryV2(nullptr) { }
+
+  virtual std::unique_ptr<CompactionFilterV2>
+  CreateCompactionFilterV2(
+      const CompactionFilterContext& context) override {
+    return std::unique_ptr<CompactionFilterV2>(nullptr);
+  }
+
+  virtual const char* Name() const override {
+    return "DefaultCompactionFilterFactoryV2";
+  }
+};
+
 }  // namespace rocksdb

 #endif  // STORAGE_ROCKSDB_INCLUDE_COMPACTION_FILTER_H_
@@ -13,6 +13,9 @@
 #include <stdio.h>
 #include <memory>
 #include <vector>
+#include <string>
+#include <unordered_map>
+#include "rocksdb/version.h"
 #include "rocksdb/iterator.h"
 #include "rocksdb/options.h"
 #include "rocksdb/types.h"
@@ -22,19 +25,37 @@ namespace rocksdb {

 using std::unique_ptr;

-// Update Makefile if you change these
-static const int kMajorVersion = 2;
-static const int kMinorVersion = 0;
+class ColumnFamilyHandle {
+ public:
+  virtual ~ColumnFamilyHandle() {}
+};
+extern const std::string kDefaultColumnFamilyName;
+
+struct ColumnFamilyDescriptor {
+  std::string name;
+  ColumnFamilyOptions options;
+  ColumnFamilyDescriptor()
+      : name(kDefaultColumnFamilyName), options(ColumnFamilyOptions()) {}
+  ColumnFamilyDescriptor(const std::string& _name,
+                         const ColumnFamilyOptions& _options)
+      : name(_name), options(_options) {}
+};
+
+static const int kMajorVersion = __ROCKSDB_MAJOR__;
+static const int kMinorVersion = __ROCKSDB_MINOR__;

 struct Options;
 struct ReadOptions;
 struct WriteOptions;
 struct FlushOptions;
+struct TableProperties;
 class WriteBatch;
+class Env;

 // Metadata associated with each SST file.
 struct LiveFileMetaData {
-  std::string name;        // Name of the file
+  std::string column_family_name;  // Name of the column family
+  std::string name;                // Name of the file
  int level;               // Level at which this file resides.
  size_t size;             // File size in bytes.
  std::string smallestkey; // Smallest user defined key in the file.
@@ -60,6 +81,12 @@ struct Range {
  Range(const Slice& s, const Slice& l) : start(s), limit(l) { }
 };

+// A collections of table properties objects, where
+//  key: is the table's file name.
+//  value: the table properties object of the given table.
+typedef std::unordered_map<std::string, std::shared_ptr<const TableProperties>>
+    TablePropertiesCollection;
+
 // A DB is a persistent ordered map from keys to values.
 // A DB is safe for concurrent access from multiple threads without
 // any external synchronization.
@@ -82,29 +109,88 @@ class DB {
      const std::string& name, DB** dbptr,
      bool error_if_log_file_exist = false);

+  // Open the database for read only with column families. When opening DB with
+  // read only, you can specify only a subset of column families in the
+  // database that should be opened. However, you always need to specify default
+  // column family. The default column family name is 'default' and it's stored
+  // in rocksdb::kDefaultColumnFamilyName
+  static Status OpenForReadOnly(
+      const DBOptions& db_options, const std::string& name,
+      const std::vector<ColumnFamilyDescriptor>& column_families,
+      std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
+      bool error_if_log_file_exist = false);
+
+  // Open DB with column families.
+  // db_options specify database specific options
+  // column_families is the vector of all column families in the databse,
+  // 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
+  // for read-only access.
+  // The default column family name is 'default' and it's stored
+  // in rocksdb::kDefaultColumnFamilyName.
+  // If everything is OK, handles will on return be the same size
+  // as column_families --- handles[i] will be a handle that you
+  // will use to operate on column family column_family[i]
+  static Status Open(const DBOptions& db_options, const std::string& name,
+                     const std::vector<ColumnFamilyDescriptor>& column_families,
+                     std::vector<ColumnFamilyHandle*>* handles, DB** dbptr);
+
+  // ListColumnFamilies will open the DB specified by argument name
+  // and return the list of all column families in that DB
+  // through column_families argument. The ordering of
+  // column families in column_families is unspecified.
+  static Status ListColumnFamilies(const DBOptions& db_options,
+                                   const std::string& name,
+                                   std::vector<std::string>* column_families);
+
  DB() { }
  virtual ~DB();

+  // Create a column_family and return the handle of column family
+  // through the argument handle.
+  virtual Status CreateColumnFamily(const ColumnFamilyOptions& options,
+                                    const std::string& column_family_name,
+                                    ColumnFamilyHandle** handle);
+
+  // Drop a column family specified by column_family handle. This call
+  // only records a drop record in the manifest and prevents the column
+  // family from flushing and compacting.
+  virtual Status DropColumnFamily(ColumnFamilyHandle* column_family);
+
  // Set the database entry for "key" to "value".
  // Returns OK on success, and a non-OK status on error.
  // Note: consider setting options.sync = true.
  virtual Status Put(const WriteOptions& options,
-                     const Slice& key,
+                     ColumnFamilyHandle* column_family, const Slice& key,
                     const Slice& value) = 0;
+  virtual Status Put(const WriteOptions& options, const Slice& key,
+                     const Slice& value) {
+    return Put(options, DefaultColumnFamily(), key, value);
+  }

  // Remove the database entry (if any) for "key".  Returns OK on
  // success, and a non-OK status on error.  It is not an error if "key"
  // did not exist in the database.
  // Note: consider setting options.sync = true.
-  virtual Status Delete(const WriteOptions& options, const Slice& key) = 0;
+  virtual Status Delete(const WriteOptions& options,
+                        ColumnFamilyHandle* column_family,
+                        const Slice& key) = 0;
+  virtual Status Delete(const WriteOptions& options, const Slice& key) {
+    return Delete(options, DefaultColumnFamily(), key);
+  }

  // Merge the database entry for "key" with "value".  Returns OK on success,
  // and a non-OK status on error. The semantics of this operation is
  // determined by the user provided merge_operator when opening DB.
  // Note: consider setting options.sync = true.
  virtual Status Merge(const WriteOptions& options,
-                       const Slice& key,
+                       ColumnFamilyHandle* column_family, const Slice& key,
                       const Slice& value) = 0;
+  virtual Status Merge(const WriteOptions& options, const Slice& key,
+                       const Slice& value) {
+    return Merge(options, DefaultColumnFamily(), key, value);
+  }

  // Apply the specified updates to the database.
  // Returns OK on success, non-OK on failure.
@@ -119,8 +205,11 @@ class DB {
  //
  // May return some other Status on an error.
  virtual Status Get(const ReadOptions& options,
-                     const Slice& key,
+                     ColumnFamilyHandle* column_family, const Slice& key,
                     std::string* value) = 0;
+  virtual Status Get(const ReadOptions& options, const Slice& key, std::string* value) {
+    return Get(options, DefaultColumnFamily(), key, value);
+  }

  // If keys[i] does not exist in the database, then the i'th returned
  // status will be one for which Status::IsNotFound() is true, and
@@ -132,9 +221,17 @@ class DB {
  // Similarly, the number of returned statuses will be the number of keys.
  // Note: keys will not be "de-duplicated". Duplicate keys will return
  // duplicate values in order.
+  virtual std::vector<Status> MultiGet(
+      const ReadOptions& options,
+      const std::vector<ColumnFamilyHandle*>& column_family,
+      const std::vector<Slice>& keys, std::vector<std::string>* values) = 0;
  virtual std::vector<Status> MultiGet(const ReadOptions& options,
                                       const std::vector<Slice>& keys,
-                                       std::vector<std::string>* values) = 0;
+                                       std::vector<std::string>* values) {
+    return MultiGet(options, std::vector<ColumnFamilyHandle*>(
+                                 keys.size(), DefaultColumnFamily()),
+                    keys, values);
+  }

  // If the key definitely does not exist in the database, then this method
  // returns false, else true. If the caller wants to obtain value when the key
@@ -144,14 +241,17 @@ class DB {
  // to make this lighter weight is to avoid doing any IOs.
  // Default implementation here returns true and sets 'value_found' to false
  virtual bool KeyMayExist(const ReadOptions& options,
-                           const Slice& key,
-                           std::string* value,
-                           bool* value_found = nullptr) {
+                           ColumnFamilyHandle* column_family, const Slice& key,
+                           std::string* value, bool* value_found = nullptr) {
    if (value_found != nullptr) {
      *value_found = false;
    }
    return true;
  }
+  virtual bool KeyMayExist(const ReadOptions& options, const Slice& key,
+                           std::string* value, bool* value_found = nullptr) {
+    return KeyMayExist(options, DefaultColumnFamily(), key, value, value_found);
+  }

  // Return a heap-allocated iterator over the contents of the database.
  // The result of NewIterator() is initially invalid (caller must
@@ -159,12 +259,26 @@ class DB {
  //
  // Caller should delete the iterator when it is no longer needed.
  // The returned iterator should be deleted before this db is deleted.
-  virtual Iterator* NewIterator(const ReadOptions& options) = 0;
+  virtual Iterator* NewIterator(const ReadOptions& options,
+                                ColumnFamilyHandle* column_family) = 0;
+  virtual Iterator* NewIterator(const ReadOptions& options) {
+    return NewIterator(options, DefaultColumnFamily());
+  }
+  // Returns iterators from a consistent database state across multiple
+  // column families. Iterators are heap allocated and need to be deleted
+  // before the db is deleted
+  virtual Status NewIterators(
+      const ReadOptions& options,
+      const std::vector<ColumnFamilyHandle*>& column_families,
+      std::vector<Iterator*>* iterators) = 0;

  // Return a handle to the current DB state.  Iterators created with
  // this handle will all observe a stable snapshot of the current DB
  // state.  The caller must call ReleaseSnapshot(result) when the
  // snapshot is no longer needed.
+  //
+  // nullptr will be returned if the DB fails to take a snapshot or does
+  // not support snapshot.
  virtual const Snapshot* GetSnapshot() = 0;

  // Release a previously acquired snapshot.  The caller must not
@@ -185,7 +299,11 @@ class DB {
  //     about the internal operation of the DB.
  //  "rocksdb.sstables" - returns a multi-line string that describes all
  //     of the sstables that make up the db contents.
-  virtual bool GetProperty(const Slice& property, std::string* value) = 0;
+  virtual bool GetProperty(ColumnFamilyHandle* column_family,
+                           const Slice& property, std::string* value) = 0;
+  virtual bool GetProperty(const Slice& property, std::string* value) {
+    return GetProperty(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)".
@@ -195,8 +313,12 @@ class DB {
  // sizes will be one-tenth the size of the corresponding user data size.
  //
  // The results may not include the sizes of recently written data.
-  virtual void GetApproximateSizes(const Range* range, int n,
+  virtual void GetApproximateSizes(ColumnFamilyHandle* column_family,
+                                   const Range* range, int n,
                                   uint64_t* sizes) = 0;
+  virtual void GetApproximateSizes(const Range* range, int n, uint64_t* sizes) {
+    GetApproximateSizes(DefaultColumnFamily(), range, n, sizes);
+  }

  // Compact the underlying storage for the key range [*begin,*end].
  // The actual compaction interval might be superset of [*begin, *end].
@@ -215,19 +337,33 @@ 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).
-  virtual Status CompactRange(const Slice* begin, const Slice* end,
+  virtual Status CompactRange(ColumnFamilyHandle* column_family,
+                              const Slice* begin, const Slice* end,
                              bool reduce_level = false,
                              int target_level = -1) = 0;
+  virtual Status CompactRange(const Slice* begin, const Slice* end,
+                              bool reduce_level = false,
+                              int target_level = -1) {
+    return CompactRange(DefaultColumnFamily(), begin, end, reduce_level,
+                        target_level);
+  }

  // Number of levels used for this DB.
-  virtual int NumberLevels() = 0;
+  virtual int NumberLevels(ColumnFamilyHandle* column_family) = 0;
+  virtual int NumberLevels() { return NumberLevels(DefaultColumnFamily()); }

  // Maximum level to which a new compacted memtable is pushed if it
  // does not create overlap.
-  virtual int MaxMemCompactionLevel() = 0;
+  virtual int MaxMemCompactionLevel(ColumnFamilyHandle* column_family) = 0;
+  virtual int MaxMemCompactionLevel() {
+    return MaxMemCompactionLevel(DefaultColumnFamily());
+  }

  // Number of files in level-0 that would stop writes.
-  virtual int Level0StopWriteTrigger() = 0;
+  virtual int Level0StopWriteTrigger(ColumnFamilyHandle* column_family) = 0;
+  virtual int Level0StopWriteTrigger() {
+    return Level0StopWriteTrigger(DefaultColumnFamily());
+  }

  // Get DB name -- the exact same name that was provided as an argument to
  // DB::Open()
@@ -237,10 +373,23 @@ class DB {
  virtual Env* GetEnv() const = 0;

  // Get DB Options that we use
-  virtual const Options& GetOptions() const = 0;
+  virtual const Options& GetOptions(ColumnFamilyHandle* column_family)
+      const = 0;
+  virtual const Options& GetOptions() const {
+    return GetOptions(DefaultColumnFamily());
+  }

  // Flush all mem-table data.
-  virtual Status Flush(const FlushOptions& options) = 0;
+  virtual Status Flush(const FlushOptions& options,
+                       ColumnFamilyHandle* column_family) = 0;
+  virtual Status Flush(const FlushOptions& options) {
+    return Flush(options, DefaultColumnFamily());
+  }
+
+  // The sequence number of the most recent transaction.
+  virtual SequenceNumber GetLatestSequenceNumber() const = 0;
+
+#ifndef ROCKSDB_LITE

  // Prevent file deletions. Compactions will continue to occur,
  // but no obsolete files will be deleted. Calling this multiple
@@ -260,7 +409,7 @@ class DB {

  // GetLiveFiles followed by GetSortedWalFiles can generate a lossless backup

-  // THIS METHOD IS DEPRECATED. Use the GetTableMetaData to get more
+  // THIS METHOD IS DEPRECATED. Use the GetLiveFilesMetaData to get more
  // detailed information on the live files.
  // Retrieve the list of all files in the database. The files are
  // relative to the dbname and are not absolute paths. The valid size of the
@@ -270,9 +419,12 @@ class DB {
  // Setting flush_memtable to true does Flush before recording the live files.
  // Setting flush_memtable to false is useful when we don't want to wait for
  // flush which may have to wait for compaction to complete taking an
-  // indeterminate time. But this will have to use GetSortedWalFiles after
-  // GetLiveFiles to compensate for memtables missed in this snapshot due to the
-  // absence of Flush, by WAL files to recover the database consistently later
+  // indeterminate time.
+  //
+  // In case you have multiple column families, even if flush_memtable is true,
+  // you still need to call GetSortedWalFiles after GetLiveFiles to compensate
+  // for new data that arrived to already-flushed column families while other
+  // column families were flushing
  virtual Status GetLiveFiles(std::vector<std::string>&,
                              uint64_t* manifest_file_size,
                              bool flush_memtable = true) = 0;
@@ -280,9 +432,6 @@ class DB {
  // Retrieve the sorted list of all wal files with earliest file first
  virtual Status GetSortedWalFiles(VectorLogPtr& files) = 0;

-  // The sequence number of the most recent transaction.
-  virtual SequenceNumber GetLatestSequenceNumber() const = 0;
-
  // Sets iter to an iterator that is positioned at a write-batch containing
  // seq_number. If the sequence number is non existent, it returns an iterator
  // at the first available seq_no after the requested seq_no
@@ -291,8 +440,10 @@ class DB {
  // use this api, else the WAL files will get
  // cleared aggressively and the iterator might keep getting invalid before
  // an update is read.
-  virtual Status GetUpdatesSince(SequenceNumber seq_number,
-                                 unique_ptr<TransactionLogIterator>* iter) = 0;
+  virtual Status GetUpdatesSince(
+      SequenceNumber seq_number, unique_ptr<TransactionLogIterator>* iter,
+      const TransactionLogIterator::ReadOptions&
+          read_options = TransactionLogIterator::ReadOptions()) = 0;

  // Delete the file name from the db directory and update the internal state to
  // reflect that. Supports deletion of sst and log files only. 'name' must be
@@ -301,15 +452,26 @@ class DB {

  // Returns a list of all table files with their level, start key
  // and end key
-  virtual void GetLiveFilesMetaData(
-    std::vector<LiveFileMetaData> *metadata) {
-  }
+  virtual void GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {}
+
+#endif  // ROCKSDB_LITE

  // Sets the globally unique ID created at database creation time by invoking
  // Env::GenerateUniqueId(), in identity. Returns Status::OK if identity could
  // be set properly
  virtual Status GetDbIdentity(std::string& identity) = 0;

+  // Returns default column family handle
+  virtual ColumnFamilyHandle* DefaultColumnFamily() const = 0;
+
+#ifndef ROCKSDB_LITE
+  virtual Status GetPropertiesOfAllTables(ColumnFamilyHandle* column_family,
+                                          TablePropertiesCollection* props) = 0;
+  virtual Status GetPropertiesOfAllTables(TablePropertiesCollection* props) {
+    return GetPropertiesOfAllTables(DefaultColumnFamily(), props);
+  }
+#endif  // ROCKSDB_LITE
+
 private:
  // No copying allowed
  DB(const DB&);
@@ -320,11 +482,13 @@ class DB {
 // Be very careful using this method.
 Status DestroyDB(const std::string& name, const Options& options);

+#ifndef ROCKSDB_LITE
 // If a DB cannot be opened, you may attempt to call this method to
 // resurrect as much of the contents of the database as possible.
 // Some data may be lost, so be careful when calling this function
 // on a database that contains important information.
 Status RepairDB(const std::string& dbname, const Options& options);
+#endif

 }  // namespace rocksdb

@@ -34,7 +34,7 @@ class Slice;
 class WritableFile;
 class RandomRWFile;
 class Directory;
-struct Options;
+struct DBOptions;

 using std::unique_ptr;
 using std::shared_ptr;
@@ -47,7 +47,7 @@ struct EnvOptions {
  EnvOptions();

  // construct from Options
-  explicit EnvOptions(const Options& options);
+  explicit EnvOptions(const DBOptions& options);

  // If true, then allow caching of data in environment buffers
  bool use_os_buffer = true;
@@ -59,13 +59,21 @@ struct EnvOptions {
  bool use_mmap_writes = true;

  // If true, set the FD_CLOEXEC on open fd.
-  bool set_fd_cloexec= true;
+  bool set_fd_cloexec = true;

  // Allows OS to incrementally sync files to disk while they are being
  // written, in the background. Issue one request for every bytes_per_sync
  // written. 0 turns it off.
  // Default: 0
  uint64_t bytes_per_sync = 0;
+
+  // If true, we will preallocate the file with FALLOC_FL_KEEP_SIZE flag, which
+  // means that file size won't change as part of preallocation.
+  // If false, preallocation will also change the file size. This option will
+  // improve the performance in workloads where you sync the data on every
+  // write. By default, we set it to true for MANIFEST writes and false for
+  // WAL writes
+  bool fallocate_with_keep_size = true;
 };

 class Env {
@@ -205,6 +213,14 @@ class Env {
  // When "function(arg)" returns, the thread will be destroyed.
  virtual void StartThread(void (*function)(void* arg), void* arg) = 0;

+  // Wait for all threads started by StartThread to terminate.
+  virtual void WaitForJoin() {}
+
+  // Get thread pool queue length for specific thrad pool.
+  virtual unsigned int GetThreadPoolQueueLen(Priority pri = LOW) const {
+    return 0;
+  }
+
  // *path is set to a temporary directory that can be used for testing. It may
  // or many not have just been created. The directory may or may not differ
  // between runs of the same process, but subsequent calls will return the
@@ -250,6 +266,16 @@ class Env {
  // Generates a unique id that can be used to identify a db
  virtual std::string GenerateUniqueId();

+  // OptimizeForLogWrite will create a new EnvOptions object that is a copy of
+  // the EnvOptions in the parameters, but is optimized for writing log files.
+  // Default implementation returns the copy of the same object.
+  virtual EnvOptions OptimizeForLogWrite(const EnvOptions& env_options) const;
+  // OptimizeForManifestWrite will create a new EnvOptions object that is a copy
+  // of the EnvOptions in the parameters, but is optimized for writing manifest
+  // files. Default implementation returns the copy of the same object.
+  virtual EnvOptions OptimizeForManifestWrite(const EnvOptions& env_options)
+      const;
+
 private:
  // No copying allowed
  Env(const Env&);
@@ -508,25 +534,66 @@ class Directory {
  virtual Status Fsync() = 0;
 };

+enum InfoLogLevel : unsigned char {
+  DEBUG_LEVEL = 0,
+  INFO_LEVEL,
+  WARN_LEVEL,
+  ERROR_LEVEL,
+  FATAL_LEVEL,
+  NUM_INFO_LOG_LEVELS,
+};
+
 // An interface for writing log messages.
 class Logger {
 public:
  enum { DO_NOT_SUPPORT_GET_LOG_FILE_SIZE = -1 };
-  Logger() { }
+  explicit Logger(const InfoLogLevel log_level = InfoLogLevel::INFO_LEVEL)
+      : log_level_(log_level) {}
  virtual ~Logger();

  // Write an entry to the log file with the specified format.
  virtual void Logv(const char* format, va_list ap) = 0;
+
+  // Write an entry to the log file with the specified log level
+  // and format.  Any log with level under the internal log level
+  // of *this (see @SetInfoLogLevel and @GetInfoLogLevel) will not be
+  // printed.
+  void Logv(const InfoLogLevel log_level, const char* format, va_list ap) {
+    static const char* kInfoLogLevelNames[5] = {"DEBUG", "INFO", "WARN",
+                                                "ERROR", "FATAL"};
+    if (log_level < log_level_) {
+      return;
+    }
+
+    if (log_level == InfoLogLevel::INFO_LEVEL) {
+      // Doesn't print log level if it is INFO level.
+      // This is to avoid unexpected performance regression after we add
+      // the feature of log level. All the logs before we add the feature
+      // are INFO level. We don't want to add extra costs to those existing
+      // logging.
+      Logv(format, ap);
+    } else {
+      char new_format[500];
+      snprintf(new_format, sizeof(new_format) - 1, "[%s] %s",
+               kInfoLogLevelNames[log_level], format);
+      Logv(new_format, ap);
+    }
+  }
  virtual size_t GetLogFileSize() const {
    return DO_NOT_SUPPORT_GET_LOG_FILE_SIZE;
  }
  // Flush to the OS buffers
  virtual void Flush() {}
+  virtual InfoLogLevel GetInfoLogLevel() const { return log_level_; }
+  virtual void SetInfoLogLevel(const InfoLogLevel log_level) {
+    log_level_ = log_level;
+  }

 private:
  // No copying allowed
  Logger(const Logger&);
  void operator=(const Logger&);
+  InfoLogLevel log_level_;
 };


@@ -541,10 +608,20 @@ class FileLock {
  void operator=(const FileLock&);
 };

-
 extern void LogFlush(const shared_ptr<Logger>& info_log);

+extern void Log(const InfoLogLevel log_level,
+                const shared_ptr<Logger>& info_log, const char* format, ...);
+
+// a set of log functions with different log levels.
+extern void Debug(const shared_ptr<Logger>& info_log, const char* format, ...);
+extern void Info(const shared_ptr<Logger>& info_log, const char* format, ...);
+extern void Warn(const shared_ptr<Logger>& info_log, const char* format, ...);
+extern void Error(const shared_ptr<Logger>& info_log, const char* format, ...);
+extern void Fatal(const shared_ptr<Logger>& info_log, const char* format, ...);
+
 // Log the specified data to *info_log if info_log is non-nullptr.
+// The default info log level is InfoLogLevel::ERROR.
 extern void Log(const shared_ptr<Logger>& info_log, const char* format, ...)
 #   if defined(__GNUC__) || defined(__clang__)
    __attribute__((__format__ (__printf__, 2, 3)))
@@ -553,15 +630,27 @@ extern void Log(const shared_ptr<Logger>& info_log, const char* format, ...)

 extern void LogFlush(Logger *info_log);

+extern void Log(const InfoLogLevel log_level, Logger* info_log,
+                const char* format, ...);
+
+// The default info log level is InfoLogLevel::ERROR.
 extern void Log(Logger* info_log, const char* format, ...)
 #   if defined(__GNUC__) || defined(__clang__)
    __attribute__((__format__ (__printf__, 2, 3)))
 #   endif
    ;

+// a set of log functions with different log levels.
+extern void Debug(Logger* info_log, const char* format, ...);
+extern void Info(Logger* info_log, const char* format, ...);
+extern void Warn(Logger* info_log, const char* format, ...);
+extern void Error(Logger* info_log, const char* format, ...);
+extern void Fatal(Logger* info_log, const char* format, ...);
+
 // A utility routine: write "data" to the named file.
 extern Status WriteStringToFile(Env* env, const Slice& data,
-                                const std::string& fname);
+                                const std::string& fname,
+                                bool should_sync = false);

 // A utility routine: read contents of named file into *data
 extern Status ReadFileToString(Env* env, const std::string& fname,
@@ -634,6 +723,10 @@ class EnvWrapper : public Env {
  void StartThread(void (*f)(void*), void* a) {
    return target_->StartThread(f, a);
  }
+  void WaitForJoin() { return target_->WaitForJoin(); }
+  virtual unsigned int GetThreadPoolQueueLen(Priority pri = LOW) const {
+    return target_->GetThreadPoolQueueLen(pri);
+  }
  virtual Status GetTestDirectory(std::string* path) {
    return target_->GetTestDirectory(path);
  }
@@ -668,6 +761,12 @@ class EnvWrapper : public Env {
  Env* target_;
 };

+// Returns a new environment that stores its data in memory and delegates
+// all non-file-storage tasks to base_env. The caller must delete the result
+// when it is no longer needed.
+// *base_env must remain live while the result is in use.
+Env* NewMemEnv(Env* base_env);
+
 }  // namespace rocksdb

 #endif  // STORAGE_ROCKSDB_INCLUDE_ENV_H_
@@ -11,6 +11,7 @@ namespace rocksdb {

 class Slice;
 class BlockBuilder;
+struct Options;

 // FlushBlockPolicy provides a configurable way to determine when to flush a
 // block in the block based tables,
@@ -36,29 +37,22 @@ class FlushBlockPolicyFactory {
  // Callers must delete the result after any database that is using the
  // result has been closed.
  virtual FlushBlockPolicy* NewFlushBlockPolicy(
-      const BlockBuilder& data_block_builder) const = 0;
+      const Options& options, const BlockBuilder& data_block_builder) const = 0;

  virtual ~FlushBlockPolicyFactory() { }
 };

 class FlushBlockBySizePolicyFactory : public FlushBlockPolicyFactory {
 public:
-  FlushBlockBySizePolicyFactory(const uint64_t block_size,
-                                const uint64_t block_size_deviation) :
-      block_size_(block_size),
-      block_size_deviation_(block_size_deviation) {
-  }
+  FlushBlockBySizePolicyFactory() {}

  virtual const char* Name() const override {
    return "FlushBlockBySizePolicyFactory";
  }

  virtual FlushBlockPolicy* NewFlushBlockPolicy(
+      const Options& options,
      const BlockBuilder& data_block_builder) const override;
-
- private:
-  const uint64_t block_size_;
-  const uint64_t block_size_deviation_;
 };

 }  // rocksdb
@@ -2,8 +2,8 @@
 // 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 STORAGE_ROCKSDB_INCLUDE_LDB_TOOL_H
-#define STORAGE_ROCKSDB_INCLUDE_LDB_TOOL_H
+#ifndef ROCKSDB_LITE
+#pragma once
 #include "rocksdb/options.h"

 namespace rocksdb {
@@ -15,4 +15,4 @@ class LDBTool {

 } // namespace rocksdb

-#endif // STORAGE_ROCKSDB_INCLUDE_LDB_TOOL_H
+#endif  // ROCKSDB_LITE
@@ -21,7 +21,7 @@
 // types built in:
 //  - SkipListRep: This is the default; it is backed by a skip list.
 //  - HashSkipListRep: The memtable rep that is best used for keys that are
-//  structured like "prefix:suffix" where iteration withing a prefix is
+//  structured like "prefix:suffix" where iteration within a prefix is
 //  common and iteration across different prefixes is rare. It is backed by
 //  a hash map where each bucket is a skip list.
 //  - VectorRep: This is backed by an unordered std::vector. On iteration, the
@@ -33,17 +33,20 @@
 // iteration over the entire collection is rare since doing so requires all the
 // keys to be copied into a sorted data structure.

-#ifndef STORAGE_ROCKSDB_DB_MEMTABLEREP_H_
-#define STORAGE_ROCKSDB_DB_MEMTABLEREP_H_
+#pragma once

 #include <memory>
+#include <stdint.h>

 namespace rocksdb {

 class Arena;
+class LookupKey;
 class Slice;
 class SliceTransform;

+typedef void* KeyHandle;
+
 class MemTableRep {
 public:
  // KeyComparator provides a means to compare keys, which are internal keys
@@ -52,16 +55,28 @@ class MemTableRep {
   public:
    // Compare a and b. Return a negative value if a is less than b, 0 if they
    // are equal, and a positive value if a is greater than b
-    virtual int operator()(const char* a, const char* b) const = 0;
+    virtual int operator()(const char* prefix_len_key1,
+                           const char* prefix_len_key2) const = 0;
+
+    virtual int operator()(const char* prefix_len_key,
+                           const Slice& key) const = 0;

    virtual ~KeyComparator() { }
  };

+  explicit MemTableRep(Arena* arena) : arena_(arena) {}
+
+  // Allocate a buf of len size for storing key. The idea is that a specific
+  // memtable representation knows its underlying data structure better. By
+  // allowing it to allocate memory, it can possibly put correlated stuff
+  // in consecutive memory area to make processor prefetching more efficient.
+  virtual KeyHandle Allocate(const size_t len, char** buf);
+
  // Insert key into the collection. (The caller will pack key and value into a
-  // single buffer and pass that in as the parameter to Insert)
+  // single buffer and pass that in as the parameter to Insert).
  // REQUIRES: nothing that compares equal to key is currently in the
  // collection.
-  virtual void Insert(const char* key) = 0;
+  virtual void Insert(KeyHandle handle) = 0;

  // Returns true iff an entry that compares equal to key is in the collection.
  virtual bool Contains(const char* key) const = 0;
@@ -70,6 +85,20 @@ class MemTableRep {
  // nothing.
  virtual void MarkReadOnly() { }

+  // Look up key from the mem table, since the first key in the mem table whose
+  // user_key matches the one given k, call the function callback_func(), with
+  // callback_args directly forwarded as the first parameter, and the mem table
+  // key as the second parameter. If the return value is false, then terminates.
+  // Otherwise, go through the next key.
+  // It's safe for Get() to terminate after having finished all the potential
+  // key for the k.user_key(), or not.
+  //
+  // Default:
+  // Get() function with a default value of dynamically construct an iterator,
+  // seek and call the call back function.
+  virtual void Get(const LookupKey& k, void* callback_args,
+                   bool (*callback_func)(void* arg, const char* entry));
+
  // Report an approximation of how much memory has been used other than memory
  // that was allocated through the arena.
  virtual size_t ApproximateMemoryUsage() = 0;
@@ -82,7 +111,7 @@ class MemTableRep {
    // Initialize an iterator over the specified collection.
    // The returned iterator is not valid.
    // explicit Iterator(const MemTableRep* collection);
-    virtual ~Iterator() { };
+    virtual ~Iterator() {}

    // Returns true iff the iterator is positioned at a valid node.
    virtual bool Valid() const = 0;
@@ -100,7 +129,7 @@ class MemTableRep {
    virtual void Prev() = 0;

    // Advance to the first entry with a key >= target
-    virtual void Seek(const char* target) = 0;
+    virtual void Seek(const Slice& internal_key, const char* memtable_key) = 0;

    // Position at the first entry in collection.
    // Final state of iterator is Valid() iff collection is not empty.
@@ -119,33 +148,46 @@ class MemTableRep {
  // GetIterator().
  virtual Iterator* GetIterator(const Slice& user_key) { return GetIterator(); }

-  // Return an iterator over at least the keys with the specified prefix. The
-  // iterator may also allow access to other keys, but doesn't have to. Default:
-  // GetIterator().
-  virtual Iterator* GetPrefixIterator(const Slice& prefix) {
-    return GetIterator();
-  }
-
  // Return an iterator that has a special Seek semantics. The result of
  // a Seek might only include keys with the same prefix as the target key.
  virtual Iterator* GetDynamicPrefixIterator() { return GetIterator(); }

+  // Return true if the current MemTableRep supports merge operator.
+  // Default: true
+  virtual bool IsMergeOperatorSupported() const { return true; }
+
+  // Return true if the current MemTableRep supports snapshot
+  // Default: true
+  virtual bool IsSnapshotSupported() const { return true; }
+
 protected:
  // When *key is an internal key concatenated with the value, returns the
  // user key.
  virtual Slice UserKey(const char* key) const;
+
+  Arena* arena_;
 };

 // This is the base class for all factories that are used by RocksDB to create
 // new MemTableRep objects
 class MemTableRepFactory {
 public:
-  virtual ~MemTableRepFactory() { };
-  virtual MemTableRep* CreateMemTableRep(MemTableRep::KeyComparator&,
-                                         Arena*) = 0;
+  virtual ~MemTableRepFactory() {}
+  virtual MemTableRep* CreateMemTableRep(const MemTableRep::KeyComparator&,
+      Arena*, const SliceTransform*) = 0;
  virtual const char* Name() const = 0;
 };

+// This uses a skip list to store keys. It is the default.
+class SkipListFactory : public MemTableRepFactory {
+ public:
+  virtual MemTableRep* CreateMemTableRep(const MemTableRep::KeyComparator&,
+                                         Arena*,
+                                         const SliceTransform*) override;
+  virtual const char* Name() const override { return "SkipListFactory"; }
+};
+
+#ifndef ROCKSDB_LITE
 // This creates MemTableReps that are backed by an std::vector. On iteration,
 // the vector is sorted. This is useful for workloads where iteration is very
 // rare and writes are generally not issued after reads begin.
@@ -156,45 +198,73 @@ class MemTableRepFactory {
 //     bytes reserved for usage.
 class VectorRepFactory : public MemTableRepFactory {
  const size_t count_;
-public:
+
+ public:
  explicit VectorRepFactory(size_t count = 0) : count_(count) { }
-  virtual MemTableRep* CreateMemTableRep(MemTableRep::KeyComparator&,
-                                         Arena*) override;
+  virtual MemTableRep* CreateMemTableRep(
+      const MemTableRep::KeyComparator&, Arena*,
+      const SliceTransform*) override;
  virtual const char* Name() const override {
    return "VectorRepFactory";
  }
 };

-// This uses a skip list to store keys. It is the default.
-class SkipListFactory : public MemTableRepFactory {
-public:
- virtual MemTableRep* CreateMemTableRep(MemTableRep::KeyComparator&,
-                                        Arena*) override;
-  virtual const char* Name() const override {
-    return "SkipListFactory";
-  }
-};
+// This class contains a fixed array of buckets, each
+// pointing to a skiplist (null if the bucket is empty).
+// bucket_count: number of fixed array buckets
+// skiplist_height: the max height of the skiplist
+// skiplist_branching_factor: probabilistic size ratio between adjacent
+//                            link lists in the skiplist
+extern MemTableRepFactory* NewHashSkipListRepFactory(
+    size_t bucket_count = 1000000, int32_t skiplist_height = 4,
+    int32_t skiplist_branching_factor = 4
+);

-// HashSkipListRep is backed by hash map of buckets. Each bucket is a skip
-// list. All the keys with the same prefix will be in the same bucket.
-// The prefix is determined using user supplied SliceTransform. It has
-// to match prefix_extractor in options.prefix_extractor.
+// The factory is to create memtables with a hashed linked list:
+// it contains a fixed array of buckets, each pointing to a sorted single
+// linked list (null if the bucket is empty).
+// @bucket_count: number of fixed array buckets
+// @huge_page_tlb_size: if <=0, allocate the hash table bytes from malloc.
+//                      Otherwise from huge page TLB. The user needs to reserve
+//                      huge pages for it to be allocated, like:
+//                          sysctl -w vm.nr_hugepages=20
+//                      See linux doc Documentation/vm/hugetlbpage.txt
+extern MemTableRepFactory* NewHashLinkListRepFactory(
+    size_t bucket_count = 50000, size_t huge_page_tlb_size = 0);
+
+// This factory creates a cuckoo-hashing based mem-table representation.
+// Cuckoo-hash is a closed-hash strategy, in which all key/value pairs
+// are stored in the bucket array itself intead of in some data structures
+// external to the bucket array.  In addition, each key in cuckoo hash
+// has a constant number of possible buckets in the bucket array.  These
+// two properties together makes cuckoo hash more memory efficient and
+// a constant worst-case read time.  Cuckoo hash is best suitable for
+// point-lookup workload.
 //
-// Iteration over the entire collection is implemented by dumping all the keys
-// into a separate skip list. Thus, these data structures are best used when
-// iteration over the entire collection is rare.
+// When inserting a key / value, it first checks whether one of its possible
+// buckets is empty.  If so, the key / value will be inserted to that vacant
+// bucket.  Otherwise, one of the keys originally stored in one of these
+// possible buckets will be "kicked out" and move to one of its possible
+// buckets (and possibly kicks out another victim.)  In the current
+// implementation, such "kick-out" path is bounded.  If it cannot find a
+// "kick-out" path for a specific key, this key will be stored in a backup
+// structure, and the current memtable to be forced to immutable.
+//
+// Note that currently this mem-table representation does not support
+// snapshot (i.e., it only queries latest state) and iterators.  In addition,
+// MultiGet operation might also lose its atomicity due to the lack of
+// snapshot support.
 //
 // Parameters:
-//   transform: The prefix extractor that returns prefix when supplied a user
-//     key. Has to match options.prefix_extractor
-//   bucket_count: Number of buckets in a hash_map. Each bucket needs
-//     8 bytes. By default, we set buckets to one million, which
-//     will take 8MB of memory. If you know the number of keys you'll
-//     keep in hash map, set bucket count to be approximately twice
-//     the number of keys
-extern MemTableRepFactory* NewHashSkipListRepFactory(
-    const SliceTransform* transform, size_t bucket_count = 1000000);
-
-}
-
-#endif // STORAGE_ROCKSDB_DB_MEMTABLEREP_H_
+//   write_buffer_size: the write buffer size in bytes.
+//   average_data_size: the average size of key + value in bytes.  This value
+//     together with write_buffer_size will be used to compute the number
+//     of buckets.
+//   hash_function_count: the number of hash functions that will be used by
+//     the cuckoo-hash.  The number also equals to the number of possible
+//     buckets each key will have.
+extern MemTableRepFactory* NewHashCuckooRepFactory(
+    size_t write_buffer_size, size_t average_data_size = 64,
+    unsigned int hash_function_count = 4);
+#endif  // ROCKSDB_LITE
+}  // namespace rocksdb
@@ -32,9 +32,9 @@ class Logger;
 //
 //  b) MergeOperator - the generic class for all the more abstract / complex
 //    operations; one method (FullMerge) to merge a Put/Delete value with a
-//    merge operand; and another method (PartialMerge) that merges two
-//    operands together. this is especially useful if your key values have a
-//    complex structure but you would still like to support client-specific
+//    merge operand; and another method (PartialMerge) that merges multiple
+//    operands together. this is especially useful if your key values have
+//    complex structures but you would still like to support client-specific
 //    incremental updates.
 //
 // AssociativeMergeOperator is simpler to implement. MergeOperator is simply
@@ -80,6 +80,13 @@ class MergeOperator {
  // DB::Merge(key, *new_value) would yield the same result as a call
  // to DB::Merge(key, left_op) followed by DB::Merge(key, right_op).
  //
+  // The default implementation of PartialMergeMulti will use this function
+  // as a helper, for backward compatibility.  Any successor class of
+  // MergeOperator should either implement PartialMerge or PartialMergeMulti,
+  // although implementing PartialMergeMulti is suggested as it is in general
+  // more effective to merge multiple operands at a time instead of two
+  // operands at a time.
+  //
  // If it is impossible or infeasible to combine the two operations,
  // leave new_value unchanged and return false. The library will
  // internally keep track of the operations, and apply them in the
@@ -89,12 +96,38 @@ class MergeOperator {
  // and simply "return false". For now, the client should simply return
  // false in any case it cannot perform partial-merge, regardless of reason.
  // If there is corruption in the data, handle it in the FullMerge() function,
-  // and return false there.
-  virtual bool PartialMerge(const Slice& key,
-                            const Slice& left_operand,
-                            const Slice& right_operand,
-                            std::string* new_value,
-                            Logger* logger) const = 0;
+  // and return false there.  The default implementation of PartialMerge will
+  // always return false.
+  virtual bool PartialMerge(const Slice& key, const Slice& left_operand,
+                            const Slice& right_operand, std::string* new_value,
+                            Logger* logger) const {
+    return false;
+  }
+
+  // This function performs merge when all the operands are themselves merge
+  // operation types that you would have passed to a DB::Merge() call in the
+  // same order (front() first)
+  // (i.e. DB::Merge(key, operand_list[0]), followed by
+  //  DB::Merge(key, operand_list[1]), ...)
+  //
+  // PartialMergeMulti should combine them into a single merge operation that is
+  // saved into *new_value, and then it should return true.  *new_value should
+  // be constructed such that a call to DB::Merge(key, *new_value) would yield
+  // the same result as subquential individual calls to DB::Merge(key, operand)
+  // for each operand in operand_list from front() to back().
+  //
+  // The PartialMergeMulti function will be called only when the list of
+  // operands are long enough. The minimum amount of operands that will be
+  // passed to the function are specified by the "min_partial_merge_operands"
+  // option.
+  //
+  // In the default implementation, PartialMergeMulti will invoke PartialMerge
+  // multiple times, where each time it only merges two operands.  Developers
+  // should either implement PartialMergeMulti, or implement PartialMerge which
+  // is served as the helper function of the default PartialMergeMulti.
+  virtual bool PartialMergeMulti(const Slice& key,
+                                 const std::deque<Slice>& operand_list,
+                                 std::string* new_value, Logger* logger) const;

  // The name of the MergeOperator. Used to check for MergeOperator
  // mismatches (i.e., a DB created with one MergeOperator is
@@ -15,6 +15,7 @@
 #include <vector>
 #include <stdint.h>

+#include "rocksdb/version.h"
 #include "rocksdb/universal_compaction.h"

 namespace rocksdb {
@@ -22,8 +23,10 @@ namespace rocksdb {
 class Cache;
 class CompactionFilter;
 class CompactionFilterFactory;
+class CompactionFilterFactoryV2;
 class Comparator;
 class Env;
+enum InfoLogLevel : unsigned char;
 class FilterPolicy;
 class Logger;
 class MergeOperator;
@@ -34,6 +37,7 @@ class TablePropertiesCollector;
 class Slice;
 class SliceTransform;
 class Statistics;
+class InternalKeyComparator;

 using std::shared_ptr;

@@ -44,10 +48,8 @@ using std::shared_ptr;
 enum CompressionType : char {
  // NOTE: do not change the values of existing entries, as these are
  // part of the persistent format on disk.
-  kNoCompression = 0x0,
-  kSnappyCompression = 0x1,
-  kZlibCompression = 0x2,
-  kBZip2Compression = 0x3
+  kNoCompression = 0x0, kSnappyCompression = 0x1, kZlibCompression = 0x2,
+  kBZip2Compression = 0x3, kLZ4Compression = 0x4, kLZ4HCCompression = 0x5
 };

 enum CompactionStyle : char {
@@ -61,12 +63,19 @@ struct CompressionOptions {
  int level;
  int strategy;
  CompressionOptions() : window_bits(-14), level(-1), strategy(0) {}
-  CompressionOptions(int wbits, int lev, int strategy)
-      : window_bits(wbits), level(lev), strategy(strategy) {}
+  CompressionOptions(int wbits, int _lev, int _strategy)
+      : window_bits(wbits), level(_lev), strategy(_strategy) {}
 };

-// Options to control the behavior of a database (passed to DB::Open)
-struct Options {
+enum UpdateStatus {    // Return status For inplace update callback
+  UPDATE_FAILED   = 0, // Nothing to update
+  UPDATED_INPLACE = 1, // Value updated inplace
+  UPDATED         = 2, // No inplace update. Merged value set
+};
+
+struct Options;
+
+struct ColumnFamilyOptions {
  // -------------------
  // Parameters that affect behavior

@@ -117,35 +126,11 @@ struct Options {
  // Default: a factory that doesn't provide any object
  std::shared_ptr<CompactionFilterFactory> compaction_filter_factory;

-  // If true, the database will be created if it is missing.
-  // Default: false
-  bool create_if_missing;
-
-  // If true, an error is raised if the database already exists.
-  // Default: false
-  bool error_if_exists;
-
-  // If true, the implementation will do aggressive checking of the
-  // data it is processing and will stop early if it detects any
-  // errors.  This may have unforeseen ramifications: for example, a
-  // corruption of one DB entry may cause a large number of entries to
-  // become unreadable or for the entire DB to become unopenable.
-  // If any of the  writes to the database fails (Put, Delete, Merge, Write),
-  // the database will switch to read-only mode and fail all other
-  // Write operations.
-  // Default: false
-  bool paranoid_checks;
-
-  // Use the specified object to interact with the environment,
-  // e.g. to read/write files, schedule background work, etc.
-  // Default: Env::Default()
-  Env* env;
-
-  // Any internal progress/error information generated by the db will
-  // be written to info_log if it is non-nullptr, or to a file stored
-  // in the same directory as the DB contents if info_log is nullptr.
-  // Default: nullptr
-  shared_ptr<Logger> info_log;
+  // Version TWO of the compaction_filter_factory
+  // It supports rolling compaction
+  //
+  // Default: a factory that doesn't provide any object
+  std::shared_ptr<CompactionFilterFactoryV2> compaction_filter_factory_v2;

  // -------------------
  // Parameters that affect performance
@@ -178,13 +163,6 @@ struct Options {
  // individual write buffers.  Default: 1
  int min_write_buffer_number_to_merge;

-  // Number of open files that can be used by the DB.  You may need to
-  // increase this if your database has a large working set (budget
-  // one open file per 2MB of working set).
-  //
-  // Default: 1000
-  int max_open_files;
-
  // Control over blocks (user data is stored in a set of blocks, and
  // a block is the unit of reading from disk).

@@ -267,7 +245,7 @@ struct Options {
  // 4) prefix(prefix(key)) == prefix(key)
  //
  // Default: nullptr
-  const SliceTransform* prefix_extractor;
+  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.
@@ -280,6 +258,8 @@ struct Options {

  // Number of files to trigger level-0 compaction. A value <0 means that
  // level-0 compaction will not be triggered by number of files at all.
+  //
+  // Default: 4
  int level0_file_num_compaction_trigger;

  // Soft limit on number of level-0 files. We start slowing down writes at this
@@ -352,6 +332,268 @@ struct Options {
  // stop building a single file in a level->level+1 compaction.
  int max_grandparent_overlap_factor;

+  // Disable compaction triggered by seek.
+  // With bloomfilter and fast storage, a miss on one level
+  // is very cheap if the file handle is cached in table cache
+  // (which is true if max_open_files is large).
+  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
+  // hold, RocksDB will set soft_rate_limit = hard_rate_limit
+  // Default: 0 (disabled)
+  double soft_rate_limit;
+
+  // Puts are delayed 1ms at a time when any level has a compaction score that
+  // exceeds hard_rate_limit. This is ignored when <= 1.0.
+  // 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
+  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).
+  //
+  // There are two additonal restriction of the The specified size:
+  // (1) size should be in the range of [4096, 2 << 30] and
+  // (2) be the multiple of the CPU word (which helps with the memory
+  // alignment).
+  //
+  // We'll automatically check and adjust the size number to make sure it
+  // conforms to the restrictions.
+  //
+  // Default: 0
+  size_t arena_block_size;
+
+  // Disable automatic compactions. Manual compactions can still
+  // be issued on this column family
+  bool disable_auto_compactions;
+
+  // Purge duplicate/deleted keys when a memtable is flushed to storage.
+  // 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;
+
+  // If true, compaction will verify checksum on every read that happens
+  // as part of compaction
+  // Default: true
+  bool verify_checksums_in_compaction;
+
+  // The options needed to support Universal Style compactions
+  CompactionOptionsUniversal compaction_options_universal;
+
+  // Use KeyMayExist API to filter deletes when this is true.
+  // If KeyMayExist returns false, i.e. the key definitely does not exist, then
+  // the delete is a noop. KeyMayExist only incurs in-memory look up.
+  // This optimization avoids writing the delete to storage when appropriate.
+  // Default: false
+  bool filter_deletes;
+
+  // An iteration->Next() sequentially skips over keys with the same
+  // user-key unless this option is set. This number specifies the number
+  // of keys (with the same userkey) that will be sequentially
+  // skipped before a reseek is issued.
+  // Default: 8
+  uint64_t max_sequential_skip_in_iterations;
+
+  // This is a factory that provides MemTableRep objects.
+  // Default: a factory that provides a skip-list-based implementation of
+  // MemTableRep.
+  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.
+  std::shared_ptr<TableFactory> table_factory;
+
+  // This option allows user to to collect their own interested statistics of
+  // the tables.
+  // Default: emtpy vector -- no user-defined statistics collection will be
+  // performed.
+  typedef std::vector<std::shared_ptr<TablePropertiesCollector>>
+      TablePropertiesCollectors;
+  TablePropertiesCollectors table_properties_collectors;
+
+  // Allows thread-safe inplace updates.
+  // If inplace_callback function is not set,
+  //   Put(key, new_value) will update inplace the existing_value iff
+  //   * key exists in current memtable
+  //   * new sizeof(new_value) <= sizeof(existing_value)
+  //   * existing_value for that key is a put i.e. kTypeValue
+  // If inplace_callback function is set, check doc for inplace_callback.
+  // Default: false.
+  bool inplace_update_support;
+
+  // Number of locks used for inplace update
+  // Default: 10000, if inplace_update_support = true, else 0.
+  size_t inplace_update_num_locks;
+
+  // existing_value - pointer to previous value (from both memtable and sst).
+  //                  nullptr if key doesn't exist
+  // existing_value_size - pointer to size of existing_value).
+  //                       nullptr if key doesn't exist
+  // delta_value - Delta value to be merged with the existing_value.
+  //               Stored in transaction logs.
+  // merged_value - Set when delta is applied on the previous value.
+
+  // Applicable only when inplace_update_support is true,
+  // this callback function is called at the time of updating the memtable
+  // as part of a Put operation, lets say Put(key, delta_value). It allows the
+  // 'delta_value' specified as part of the Put operation to be merged with
+  // an 'existing_value' of the key in the database.
+
+  // If the merged value is smaller in size that the 'existing_value',
+  // then this function can update the 'existing_value' buffer inplace and
+  // the corresponding 'existing_value'_size pointer, if it wishes to.
+  // The callback should return UpdateStatus::UPDATED_INPLACE.
+  // In this case. (In this case, the snapshot-semantics of the rocksdb
+  // Iterator is not atomic anymore).
+
+  // If the merged value is larger in size than the 'existing_value' or the
+  // application does not wish to modify the 'existing_value' buffer inplace,
+  // then the merged value should be returned via *merge_value. It is set by
+  // merging the 'existing_value' and the Put 'delta_value'. The callback should
+  // return UpdateStatus::UPDATED in this case. This merged value will be added
+  // to the memtable.
+
+  // If merging fails or the application does not wish to take any action,
+  // then the callback should return UpdateStatus::UPDATE_FAILED.
+
+  // Please remember that the original call from the application is Put(key,
+  // delta_value). So the transaction log (if enabled) will still contain (key,
+  // delta_value). The 'merged_value' is not stored in the transaction log.
+  // Hence the inplace_callback function should be consistent across db reopens.
+
+  // Default: nullptr
+  UpdateStatus (*inplace_callback)(char* existing_value,
+                                   uint32_t* existing_value_size,
+                                   Slice delta_value,
+                                   std::string* merged_value);
+
+  // if prefix_extractor is set and bloom_bits is not 0, create prefix bloom
+  // for memtable
+  uint32_t memtable_prefix_bloom_bits;
+
+  // number of hash probes per key
+  uint32_t memtable_prefix_bloom_probes;
+
+  // Page size for huge page TLB for bloom in memtable. If <=0, not allocate
+  // from huge page TLB but from malloc.
+  // Need to reserve huge pages for it to be allocated. For example:
+  //      sysctl -w vm.nr_hugepages=20
+  // See linux doc Documentation/vm/hugetlbpage.txt
+
+  size_t memtable_prefix_bloom_huge_page_tlb_size;
+
+  // Control locality of bloom filter probes to improve cache miss rate.
+  // This option only applies to memtable prefix bloom and plaintable
+  // prefix bloom. It essentially limits the max number of cache lines each
+  // bloom filter check can touch.
+  // This optimization is turned off when set to 0. The number should never
+  // be greater than number of probes. This option can boost performance
+  // for in-memory workload but should use with care since it can cause
+  // higher false positive rate.
+  // Default: 0
+  uint32_t bloom_locality;
+
+  // Maximum number of successive merge operations on a key in the memtable.
+  //
+  // When a merge operation is added to the memtable and the maximum number of
+  // successive merges is reached, the value of the key will be calculated and
+  // inserted into the memtable instead of the merge operation. This will
+  // ensure that there are never more than max_successive_merges merge
+  // operations in the memtable.
+  //
+  // Default: 0 (disabled)
+  size_t max_successive_merges;
+
+  // The number of partial merge operands to accumulate before partial
+  // merge will be performed. Partial merge will not be called
+  // if the list of values to merge is less than min_partial_merge_operands.
+  //
+  // If min_partial_merge_operands < 2, then it will be treated as 2.
+  //
+  // Default: 2
+  uint32_t min_partial_merge_operands;
+
+  // Create ColumnFamilyOptions with default values for all fields
+  ColumnFamilyOptions();
+  // Create ColumnFamilyOptions from Options
+  explicit ColumnFamilyOptions(const Options& options);
+
+  void Dump(Logger* log) const;
+};
+
+struct DBOptions {
+  // If true, the database will be created if it is missing.
+  // Default: false
+  bool create_if_missing;
+
+  // If true, an error is raised if the database already exists.
+  // Default: false
+  bool error_if_exists;
+
+  // If true, the implementation will do aggressive checking of the
+  // data it is processing and will stop early if it detects any
+  // errors.  This may have unforeseen ramifications: for example, a
+  // corruption of one DB entry may cause a large number of entries to
+  // become unreadable or for the entire DB to become unopenable.
+  // If any of the  writes to the database fails (Put, Delete, Merge, Write),
+  // the database will switch to read-only mode and fail all other
+  // Write operations.
+  // Default: true
+  bool paranoid_checks;
+
+  // Use the specified object to interact with the environment,
+  // e.g. to read/write files, schedule background work, etc.
+  // Default: Env::Default()
+  Env* env;
+
+  // Any internal progress/error information generated by the db will
+  // be written to info_log if it is non-nullptr, or to a file stored
+  // in the same directory as the DB contents if info_log is nullptr.
+  // Default: nullptr
+  shared_ptr<Logger> info_log;
+
+  InfoLogLevel info_log_level;
+
+  // Number of open files that can be used by the DB.  You may need to
+  // increase this if your database has a large working set. Value -1 means
+  // files opened are always kept open. You can estimate number of files based
+  // on target_file_size_base and target_file_size_multiplier for level-based
+  // compaction. For universal-style compaction, you can usually set it to -1.
+  // Default: 5000
+  int max_open_files;
+
+  // Once write-ahead logs exceed this size, we will start forcing the flush of
+  // column families whose memtables are backed by the oldest live WAL file
+  // (i.e. the ones that are causing all the space amplification). If set to 0
+  // (default), we will dynamically choose the WAL size limit to be
+  // [sum of all write_buffer_size * max_write_buffer_number] * 2
+  // Default: 0
+  uint64_t max_total_wal_size;
+
  // If non-null, then we should collect metrics about database operations
  // Statistics objects should not be shared between DB instances as
  // it does not use any locks to prevent concurrent updates.
@@ -393,25 +635,23 @@ struct Options {
  //   all log files in wal_dir and the dir itself is deleted
  std::string wal_dir;

-  // Disable compaction triggered by seek.
-  // With bloomfilter and fast storage, a miss on one level
-  // is very cheap if the file handle is cached in table cache
-  // (which is true if max_open_files is large).
-  bool disable_seek_compaction;
-
  // The periodicity when obsolete files get deleted. The default
  // value is 6 hours. The files that get out of scope by compaction
  // process will still get automatically delete on every compaction,
  // regardless of this setting
  uint64_t delete_obsolete_files_period_micros;

-  // Maximum number of concurrent background jobs, submitted to
-  // the default LOW priority thread pool
+  // Maximum number of concurrent background compaction jobs, submitted to
+  // the default LOW priority thread pool.
+  // If you're increasing this, also consider increasing number of threads in
+  // LOW priority thread pool. For more information, see
+  // Env::SetBackgroundThreads
  // Default: 1
  int max_background_compactions;

  // Maximum number of concurrent background memtable flush jobs, submitted to
  // the HIGH priority thread pool.
+  //
  // By default, all background jobs (major compaction and memtable flush) go
  // to the LOW priority pool. If this option is set to a positive number,
  // memtable flush jobs will be submitted to the HIGH priority pool.
@@ -419,7 +659,11 @@ struct Options {
  // Without a separate pool, long running major compaction jobs could
  // potentially block memtable flush jobs of other db instances, leading to
  // unnecessary Put stalls.
-  // Default: 0
+  //
+  // If you're increasing this, also consider increasing number of threads in
+  // HIGH priority thread pool. For more information, see
+  // Env::SetBackgroundThreads
+  // Default: 1
  int max_background_flushes;

  // Specify the maximal size of the info log file. If the log file
@@ -439,34 +683,11 @@ struct Options {
  // Default: 1000
  size_t keep_log_file_num;

-  // 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
-  // hold, RocksDB will set soft_rate_limit = hard_rate_limit
-  // Default: 0 (disabled)
-  double soft_rate_limit;
-
-  // Puts are delayed 1ms at a time when any level has a compaction score that
-  // exceeds hard_rate_limit. This is ignored when <= 1.0.
-  // 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
-  unsigned int rate_limit_delay_max_milliseconds;
-
  // manifest file is rolled over on reaching this limit.
  // The older manifest file be deleted.
  // The default value is MAX_INT so that roll-over does not take place.
  uint64_t max_manifest_file_size;

-  // 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;
-
  // Number of shards used for table cache.
  int table_cache_numshardbits;

@@ -479,40 +700,6 @@ struct Options {
  // order.
  int table_cache_remove_scan_count_limit;

-  // Size of one block in arena memory allocation.
-  //
-  // If <= 0, a proper value is automatically calculated (usually about 1/10 of
-  // writer_buffer_size).
-  //
-  // There are two additonal restriction of the The specified size:
-  // (1) size should be in the range of [4096, 2 << 30] and
-  // (2) be the multiple of the CPU word (which helps with the memory
-  // alignment).
-  //
-  // We'll automatically check and adjust the size number to make sure it
-  // conforms to the restrictions.
-  //
-  // Default: 0
-  size_t arena_block_size;
-
-  // Create an Options object with default values for all fields.
-  Options();
-
-  void Dump(Logger* log) const;
-
-  // Set appropriate parameters for bulk loading.
-  // The reason that this is a function that returns "this" instead of a
-  // constructor is to enable chaining of multiple similar calls in the future.
-  //
-  // All data will be in level 0 without any automatic compaction.
-  // It's recommended to manually call CompactRange(NULL, NULL) before reading
-  // from the database, because otherwise the read can be very slow.
-  Options* PrepareForBulkLoad();
-
-  // Disable automatic compactions. Manual compactions can still
-  // be issued on this database.
-  bool disable_auto_compactions;
-
  // The following two fields affect how archived logs will be deleted.
  // 1. If both set to 0, logs will be deleted asap and will not get into
  //    the archive.
@@ -534,10 +721,6 @@ struct Options {
  // large amounts of data (such as xfs's allocsize option).
  size_t manifest_preallocation_size;

-  // Purge duplicate/deleted keys when a memtable is flushed to storage.
-  // Default: true
-  bool purge_redundant_kvs_while_flush;
-
  // Data being read from file storage may be buffered in the OS
  // Default: true
  bool allow_os_buffer;
@@ -545,7 +728,7 @@ struct Options {
  // 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: true
+  // Allow the OS to mmap file for writing. Default: false
  bool allow_mmap_writes;

  // Disable child process inherit open files. Default: true
@@ -560,14 +743,6 @@ struct Options {
  // Default: 3600 (1 hour)
  unsigned int stats_dump_period_sec;

-  // 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;
-
  // If set true, will hint the underlying file system that the file
  // access pattern is random, when a sst file is opened.
  // Default: true
@@ -596,64 +771,40 @@ struct Options {
  // Default: 0
  uint64_t bytes_per_sync;

-  // The compaction style. Default: kCompactionStyleLevel
-  CompactionStyle compaction_style;
+  // Allow RocksDB to use thread local storage to optimize performance.
+  // Default: true
+  bool allow_thread_local;

-  // The options needed to support Universal Style compactions
-  CompactionOptionsUniversal compaction_options_universal;
+  // Create DBOptions with default values for all fields
+  DBOptions();
+  // Create DBOptions from Options
+  explicit DBOptions(const Options& options);

-  // Use KeyMayExist API to filter deletes when this is true.
-  // If KeyMayExist returns false, i.e. the key definitely does not exist, then
-  // the delete is a noop. KeyMayExist only incurs in-memory look up.
-  // This optimization avoids writing the delete to storage when appropriate.
-  // Default: false
-  bool filter_deletes;
+  void Dump(Logger* log) const;
+};

-  // An iteration->Next() sequentially skips over keys with the same
-  // user-key unless this option is set. This number specifies the number
-  // of keys (with the same userkey) that will be sequentially
-  // skipped before a reseek is issued.
-  // Default: 8
-  uint64_t max_sequential_skip_in_iterations;
+// Options to control the behavior of a database (passed to DB::Open)
+struct Options : public DBOptions, public ColumnFamilyOptions {
+  // Create an Options object with default values for all fields.
+  Options() :
+    DBOptions(),
+    ColumnFamilyOptions() {}

-  // This is a factory that provides MemTableRep objects.
-  // Default: a factory that provides a skip-list-based implementation of
-  // MemTableRep.
-  std::shared_ptr<MemTableRepFactory> memtable_factory;
+  Options(const DBOptions& db_options,
+          const ColumnFamilyOptions& column_family_options)
+      : DBOptions(db_options), ColumnFamilyOptions(column_family_options) {}

-  // This is a factory that provides TableFactory objects.
-  // Default: a factory that provides a default implementation of
-  // Table and TableBuilder.
-  std::shared_ptr<TableFactory> table_factory;
+  void Dump(Logger* log) const;

-  // This option allows user to to collect their own interested statistics of
-  // the tables.
-  // Default: emtpy vector -- no user-defined statistics collection will be
-  // performed.
-  std::vector<std::shared_ptr<TablePropertiesCollector>>
-  table_properties_collectors;
-
-  // Allows thread-safe inplace updates. Requires Updates iff
-  // * key exists in current memtable
-  // * new sizeof(new_value) <= sizeof(old_value)
-  // * old_value for that key is a put i.e. kTypeValue
-  // Default: false.
-  bool inplace_update_support;
-
-  // Number of locks used for inplace update
-  // Default: 10000, if inplace_update_support = true, else 0.
-  size_t inplace_update_num_locks;
-
-  // Maximum number of successive merge operations on a key in the memtable.
+  // Set appropriate parameters for bulk loading.
+  // The reason that this is a function that returns "this" instead of a
+  // constructor is to enable chaining of multiple similar calls in the future.
  //
-  // When a merge operation is added to the memtable and the maximum number of
-  // successive merges is reached, the value of the key will be calculated and
-  // inserted into the memtable instead of the merge operation. This will
-  // ensure that there are never more than max_successive_merges merge
-  // operations in the memtable.
-  //
-  // Default: 0 (disabled)
-  size_t max_successive_merges;
+
+  // All data will be in level 0 without any automatic compaction.
+  // It's recommended to manually call CompactRange(NULL, NULL) before reading
+  // from the database, because otherwise the read can be very slow.
+  Options* PrepareForBulkLoad();
 };

 //
@@ -672,7 +823,7 @@ enum ReadTier {
 struct ReadOptions {
  // If true, all data read from underlying storage will be
  // verified against corresponding checksums.
-  // Default: false
+  // Default: true
  bool verify_checksums;

  // Should the "data block"/"index block"/"filter block" read for this
@@ -683,7 +834,10 @@ struct ReadOptions {

  // If this option is set and memtable implementation allows, Seek
  // might only return keys with the same prefix as the seek-key
-  bool prefix_seek;
+  //
+  // ! DEPRECATED: prefix_seek is on by default when prefix_extractor
+  // is configured
+  // bool prefix_seek;

  // If "snapshot" is non-nullptr, read as of the supplied snapshot
  // (which must belong to the DB that is being read and which must
@@ -703,7 +857,9 @@ struct ReadOptions {
  // prefix, and SeekToLast() is not supported.  prefix filter with this
  // option will sometimes reduce the number of read IOPs.
  // Default: nullptr
-  const Slice* prefix;
+  //
+  // ! DEPRECATED
+  // const Slice* prefix;

  // Specify if this read request should process data that ALREADY
  // resides on a particular cache. If the required data is not
@@ -713,23 +869,22 @@ struct ReadOptions {

  // Specify to create a tailing iterator -- a special iterator that has a
  // view of the complete database (i.e. it can also be used to read newly
-  // added data) and is optimized for sequential reads.
+  // added data) and is optimized for sequential reads. It will return records
+  // that were inserted into the database after the creation of the iterator.
+  // Default: false
+  // Not supported in ROCKSDB_LITE mode!
  bool tailing;

  ReadOptions()
-      : verify_checksums(false),
+      : verify_checksums(true),
        fill_cache(true),
-        prefix_seek(false),
        snapshot(nullptr),
-        prefix(nullptr),
        read_tier(kReadAllTier),
        tailing(false) {}
  ReadOptions(bool cksum, bool cache)
      : verify_checksums(cksum),
        fill_cache(cache),
-        prefix_seek(false),
        snapshot(nullptr),
-        prefix(nullptr),
        read_tier(kReadAllTier),
        tailing(false) {}
 };