Simplify bucketing logic in ldb-ttl

Summary: [start_time, end_time) is waht I'm following for the buckets and the whole time-range. Also cleaned up some code in db_ttl.* Not correcting the spacing/indenting convention for util/ldb_cmd.cc in this diff.

Test Plan: python ldb_test.py, make ttl_test, Run mcrocksdb-backup tool, Run the ldb tool on 2 mcrocksdb production backups form sigmafio033.prn1

Reviewers: vamsi, haobo

Reviewed By: vamsi

Differential Revision: https://reviews.facebook.net/D11433

Makefile

@@ -18,7 +18,7 @@ $(shell ./build_detect_platform build_config.mk)
 # this file is generated by the previous line to set build flags and sources
 include build_config.mk
 
-WARNING_FLAGS = -Wall -Werror -Wno-unused-parameter -Wno-sign-compare
+WARNING_FLAGS = -Wall -Werror
 CFLAGS += -g $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CCFLAGS) $(OPT)
 CXXFLAGS += -g $(WARNING_FLAGS) -I. -I./include $(PLATFORM_CXXFLAGS) $(OPT) -std=gnu++0x
 
@@ -53,23 +53,25 @@ TESTS = \
 	memenv_test \
 	skiplist_test \
 	table_test \
+	ttl_test \
 	block_test \
 	version_edit_test \
 	version_set_test \
 	reduce_levels_test \
 	write_batch_test \
 	auto_roll_logger_test \
-	filelock_test
-
+	filelock_test \
+	merge_test \
+	redis_test \
+	manual_compaction_test
 
 TOOLS = \
-	manifest_dump \
         sst_dump \
         db_stress \
         ldb \
 	db_repl_stress
 
-PROGRAMS = db_bench $(TESTS) $(TOOLS)
+PROGRAMS = db_bench signal_test $(TESTS) $(TOOLS)
 BENCHMARKS = db_bench_sqlite3 db_bench_tree_db
 
 LIBRARY = libleveldb.a
@@ -116,6 +118,14 @@ check: all $(PROGRAMS) $(TESTS) $(TOOLS) ldb_tests
 ldb_tests: all $(PROGRAMS) $(TOOLS)
 	python tools/ldb_test.py
 
+crash_test: blackbox_crash_test whitebox_crash_test
+
+blackbox_crash_test: db_stress
+	python -u tools/db_crashtest.py
+
+whitebox_crash_test: db_stress
+	python -u tools/db_crashtest2.py
+
 valgrind_check: all $(PROGRAMS) $(TESTS)
 	echo TESTS THAT HAVE VALGRIND ERRORS > $(VALGRIND_DIR)/valgrind_failed_tests; \
 	echo TIMES in seconds TAKEN BY TESTS ON VALGRIND > $(VALGRIND_DIR)/valgrind_tests_times; \
@@ -130,7 +140,7 @@ valgrind_check: all $(PROGRAMS) $(TESTS)
 	done
 
 clean:
-	-rm -f $(PROGRAMS) $(BENCHMARKS) $(LIBRARY) $(SHARED) $(MEMENVLIBRARY) $(THRIFTSERVER) */*.o */*/*.o ios-x86/*/*.o ios-arm/*/*.o build_config.mk
+	-rm -f $(PROGRAMS) $(BENCHMARKS) $(LIBRARY) $(SHARED) $(MEMENVLIBRARY) $(THRIFTSERVER) */*.o */*/*.o ios-x86/*/*.o ios-arm/*/*.o build_config.mk */*.d */*/*.d
 	-rm -rf ios-x86/* ios-arm/*
 
 $(LIBRARY): $(LIBOBJECTS)
@@ -153,6 +163,9 @@ db_bench_sqlite3: doc/bench/db_bench_sqlite3.o $(LIBOBJECTS) $(TESTUTIL)
 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
 
+signal_test: util/signal_test.o $(LIBOBJECTS)
+	$(CXX) util/signal_test.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
+
 arena_test: util/arena_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/arena_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
@@ -168,6 +181,12 @@ cache_test: util/cache_test.o $(LIBOBJECTS) $(TESTHARNESS)
 coding_test: util/coding_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/coding_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
+stringappend_test: utilities/merge_operators/string_append/stringappend_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) utilities/merge_operators/string_append/stringappend_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
+
+redis_test: utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) utilities/redis/redis_lists_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
+
 histogram_test: util/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/histogram_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o$@ $(LDFLAGS)
 
@@ -180,6 +199,9 @@ 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)
 
+ttl_test: utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) utilities/ttl/ttl_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS)
+
 dbformat_test: db/dbformat_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/dbformat_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
@@ -216,6 +238,9 @@ reduce_levels_test: tools/reduce_levels_test.o $(LIBOBJECTS) $(TESTHARNESS)
 write_batch_test: db/write_batch_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) db/write_batch_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
+merge_test: db/merge_test.o $(LIBOBJECTS)
+	$(CXX) db/merge_test.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
+
 $(MEMENVLIBRARY) : $(MEMENVOBJECTS)
 	rm -f $@
 	$(AR) -rs $@ $(MEMENVOBJECTS)
@@ -223,15 +248,15 @@ $(MEMENVLIBRARY) : $(MEMENVOBJECTS)
 memenv_test : helpers/memenv/memenv_test.o $(MEMENVLIBRARY) $(LIBRARY) $(TESTHARNESS)
 	$(CXX) helpers/memenv/memenv_test.o $(MEMENVLIBRARY) $(LIBRARY) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
+manual_compaction_test: util/manual_compaction_test.o $(LIBOBJECTS) $(TESTHARNESS)
+	$(CXX) util/manual_compaction_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
+
 leveldb_shell: tools/shell/ShellContext.o tools/shell/ShellState.o tools/shell/LeveldbShell.o tools/shell/DBClientProxy.o tools/shell/ShellContext.h tools/shell/ShellState.h tools/shell/DBClientProxy.h $(LIBOBJECTS)
 	$(CXX) tools/shell/ShellContext.o tools/shell/ShellState.o tools/shell/LeveldbShell.o tools/shell/DBClientProxy.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
-DBClientProxy_test: tools/shell/test/DBClientProxyTest.o tools/shell/DBClientProxy.o $(LIBRARY) 
+DBClientProxy_test: tools/shell/test/DBClientProxyTest.o tools/shell/DBClientProxy.o $(LIBRARY)
 	$(CXX) tools/shell/test/DBClientProxyTest.o tools/shell/DBClientProxy.o $(LIBRARY) $(EXEC_LDFLAGS) $(EXEC_LDFLAGS) -o $@  $(LDFLAGS)
 
-manifest_dump: tools/manifest_dump.o $(LIBOBJECTS)
-	$(CXX) tools/manifest_dump.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
-
 filelock_test: util/filelock_test.o $(LIBOBJECTS) $(TESTHARNESS)
 	$(CXX) util/filelock_test.o $(LIBOBJECTS) $(TESTHARNESS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
@@ -244,6 +269,11 @@ sst_dump: tools/sst_dump.o $(LIBOBJECTS)
 ldb: tools/ldb.o $(LIBOBJECTS)
 	$(CXX) tools/ldb.o $(LIBOBJECTS) $(EXEC_LDFLAGS) -o $@ $(LDFLAGS)
 
+.PHONY: tags
+tags:
+	ctags `find . -name '*.cc'` `find . -name '*.h'`
+	cscope -b `find . -name '*.cc'` `find . -name '*.h'`
+
 ifeq ($(PLATFORM), IOS)
 # For iOS, create universal object files to be used on both the simulator and
 # a device.

README.fb

@@ -1,3 +1,3 @@
 * Detailed instructions on how to compile using fbcode and jemalloc
 
-* Latest release is 1.5.7.fb
+* Latest release is 1.5.9.fb

build_detect_platform

@@ -31,9 +31,9 @@ fi
 # Default to fbcode gcc on internal fb machines
 if [ -d /mnt/gvfs/third-party -a -z "$CXX" ]; then
     if [ -z "$USE_CLANG" ]; then
-        source fbcode.gcc471.sh
+        source ./fbcode.gcc471.sh
     else
-        source fbcode.clang31.sh
+        source ./fbcode.clang31.sh
     fi
 fi
 
@@ -64,6 +64,9 @@ PLATFORM_SHARED_LDFLAGS="${EXEC_LDFLAGS_SHARED} -shared -Wl,-soname -Wl,"
 PLATFORM_SHARED_CFLAGS="-fPIC"
 PLATFORM_SHARED_VERSIONED=true
 
+# generic port files (working on all platform by #ifdef) go directly in /port
+GENERIC_PORT_FILES=`find port -name '*.cc' | tr "\n" " "`
+
 # On GCC, we pick libc's memcmp over GCC's memcmp via -fno-builtin-memcmp
 case "$TARGET_OS" in
     Darwin)
@@ -71,7 +74,7 @@ case "$TARGET_OS" in
         COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -DOS_MACOSX"
         PLATFORM_SHARED_EXT=dylib
         PLATFORM_SHARED_LDFLAGS="-dynamiclib -install_name "
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/darwin/darwin_specific.cc
         ;;
     Linux)
         PLATFORM=OS_LINUX
@@ -80,43 +83,43 @@ case "$TARGET_OS" in
             COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp"
         fi
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/linux/linux_specific.cc
         ;;
     SunOS)
         PLATFORM=OS_SOLARIS
         COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_SOLARIS"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lrt"
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/sunos/sunos_specific.cc
         ;;
     FreeBSD)
         PLATFORM=OS_FREEBSD
         COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_FREEBSD"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/freebsd/freebsd_specific.cc
         ;;
     NetBSD)
         PLATFORM=OS_NETBSD
         COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_NETBSD"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread -lgcc_s"
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/netbsd/netbsd_specific.cc
         ;;
     OpenBSD)
         PLATFORM=OS_OPENBSD
         COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_OPENBSD"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -pthread"
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/openbsd/openbsd_specific.cc
         ;;
     DragonFly)
         PLATFORM=OS_DRAGONFLYBSD
         COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_DRAGONFLYBSD"
         PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS -lpthread"
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/dragonfly/dragonfly_specific.cc
         ;;
     OS_ANDROID_CROSSCOMPILE)
         PLATFORM=OS_ANDROID
 	COMMON_FLAGS="$COMMON_FLAGS -fno-builtin-memcmp -D_REENTRANT -DOS_ANDROID -DLEVELDB_PLATFORM_POSIX"
 	PLATFORM_LDFLAGS="$PLATFORM_LDFLAGS "  # All pthread features are in the Android C library
-        PORT_FILE=port/port_posix.cc
+        # PORT_FILES=port/android/android.cc
         CROSS_COMPILE=true
         ;;
     *)
@@ -130,7 +133,7 @@ esac
 # except for the test and benchmark files. By default, find will output a list
 # of all files matching either rule, so we need to append -print to make the
 # prune take effect.
-DIRS="util db table"
+DIRS="util db table utilities"
 if test "$USE_THRIFT"; then
   DIRS="$DIRS thrift/server_utils.cpp thrift/gen-cpp "
   THRIFTSERVER=leveldb_server
@@ -149,7 +152,7 @@ set +f # re-enable globbing
 
 # The sources consist of the portable files, plus the platform-specific port
 # file.
-echo "SOURCES=$PORTABLE_FILES $PORT_FILE" >> $OUTPUT
+echo "SOURCES=$PORTABLE_FILES $GENERIC_PORT_FILES $PORT_FILES" >> $OUTPUT
 echo "SOURCESCPP=$PORTABLE_CPP" >> $OUTPUT
 echo "MEMENV_SOURCES=helpers/memenv/memenv.cc" >> $OUTPUT
 

db/builder.cc

@@ -6,17 +6,20 @@
 
 #include "db/filename.h"
 #include "db/dbformat.h"
+#include "db/merge_helper.h"
 #include "db/table_cache.h"
 #include "db/version_edit.h"
 #include "leveldb/db.h"
 #include "leveldb/env.h"
 #include "leveldb/iterator.h"
+#include "util/stop_watch.h"
 
 namespace leveldb {
 
 Status BuildTable(const std::string& dbname,
                   Env* env,
                   const Options& options,
+                  const EnvOptions& soptions,
                   TableCache* table_cache,
                   Iterator* iter,
                   FileMetaData* meta,
@@ -38,7 +41,7 @@ Status BuildTable(const std::string& dbname,
   std::string fname = TableFileName(dbname, meta->number);
   if (iter->Valid()) {
     unique_ptr<WritableFile> file;
-    s = env->NewWritableFile(fname, &file);
+    s = env->NewWritableFile(fname, &file, soptions);
     if (!s.ok()) {
       return s;
     }
@@ -48,36 +51,75 @@ Status BuildTable(const std::string& dbname,
     Slice key = iter->key();
     meta->smallest.DecodeFrom(key);
 
+    MergeHelper merge(user_comparator, options.merge_operator,
+                      options.info_log.get(),
+                      true /* internal key corruption is not ok */);
+
     if (purge) {
+      ParsedInternalKey ikey;
+      // Ugly walkaround to avoid compiler error for release build
+      // TODO: find a clean way to treat in memory key corruption
+      ikey.type = kTypeValue;
       ParsedInternalKey prev_ikey;
       std::string prev_value;
       std::string prev_key;
 
-      // store first key-value
-      prev_key.assign(key.data(), key.size());
-      prev_value.assign(iter->value().data(), iter->value().size());
-      ParseInternalKey(Slice(prev_key), &prev_ikey);
-      assert(prev_ikey.sequence >= earliest_seqno_in_memtable);
+      // Ugly walkaround to avoid compiler error for release build
+      // TODO: find a clean way to treat in memory key corruption
+      auto ok __attribute__((unused)) = ParseInternalKey(key, &ikey);
+      // in-memory key corruption is not ok;
+      assert(ok);
 
-      for (iter->Next(); iter->Valid(); iter->Next()) {
+      if (ikey.type == kTypeMerge) {
+        // merge values if the first entry is of merge type
+        merge.MergeUntil(iter,  0 /* don't worry about snapshot */);
+        prev_key.assign(merge.key().data(), merge.key().size());
+        ok = ParseInternalKey(Slice(prev_key), &prev_ikey);
+        assert(ok);
+        prev_value.assign(merge.value().data(), merge.value().size());
+      } else {
+        // store first key-value
+        prev_key.assign(key.data(), key.size());
+        prev_value.assign(iter->value().data(), iter->value().size());
+        ok = ParseInternalKey(Slice(prev_key), &prev_ikey);
+        assert(ok);
+        assert(prev_ikey.sequence >= earliest_seqno_in_memtable);
+        iter->Next();
+      }
+
+      while (iter->Valid()) {
+        bool iterator_at_next = false;
         ParsedInternalKey this_ikey;
         Slice key = iter->key();
-        ParseInternalKey(key, &this_ikey);
+        ok = ParseInternalKey(key, &this_ikey);
+        assert(ok);
         assert(this_ikey.sequence >= earliest_seqno_in_memtable);
 
         if (user_comparator->Compare(prev_ikey.user_key, this_ikey.user_key)) {
           // This key is different from previous key.
           // Output prev key and remember current key
           builder->Add(Slice(prev_key), Slice(prev_value));
-          prev_key.assign(key.data(), key.size());
-          prev_value.assign(iter->value().data(), iter->value().size());
-          ParseInternalKey(Slice(prev_key), &prev_ikey);
+          if (this_ikey.type == kTypeMerge) {
+            merge.MergeUntil(iter, 0 /* don't worry about snapshot */);
+            iterator_at_next = true;
+            prev_key.assign(merge.key().data(), merge.key().size());
+            ok = ParseInternalKey(Slice(prev_key), &prev_ikey);
+            assert(ok);
+            prev_value.assign(merge.value().data(), merge.value().size());
+          } else {
+            prev_key.assign(key.data(), key.size());
+            prev_value.assign(iter->value().data(), iter->value().size());
+            ok = ParseInternalKey(Slice(prev_key), &prev_ikey);
+            assert(ok);
+          }
         } else {
           // seqno within the same key are in decreasing order
           assert(this_ikey.sequence < prev_ikey.sequence);
           // This key is an earlier version of the same key in prev_key.
           // Skip current key.
         }
+
+        if (!iterator_at_next) iter->Next();
       }
       // output last key
       builder->Add(Slice(prev_key), Slice(prev_value));
@@ -106,8 +148,10 @@ Status BuildTable(const std::string& dbname,
     // Finish and check for file errors
     if (s.ok() && !options.disableDataSync) {
       if (options.use_fsync) {
+        StopWatch sw(env, options.statistics, TABLE_SYNC_MICROS);
         s = file->Fsync();
       } else {
+        StopWatch sw(env, options.statistics, TABLE_SYNC_MICROS);
         s = file->Sync();
       }
     }
@@ -118,6 +162,7 @@ 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);
       s = it->status();

db/builder.h

@@ -15,6 +15,7 @@ struct Options;
 struct FileMetaData;
 
 class Env;
+class EnvOptions;
 class Iterator;
 class TableCache;
 class VersionEdit;
@@ -27,6 +28,7 @@ class VersionEdit;
 extern Status BuildTable(const std::string& dbname,
                          Env* env,
                          const Options& options,
+                         const EnvOptions& soptions,
                          TableCache* table_cache,
                          Iterator* iter,
                          FileMetaData* meta,

db/corruption_test.cc

@@ -10,12 +10,12 @@
 #include <sys/types.h>
 #include "leveldb/cache.h"
 #include "leveldb/env.h"
-#include "leveldb/table.h"
 #include "leveldb/write_batch.h"
 #include "db/db_impl.h"
 #include "db/filename.h"
 #include "db/log_format.h"
 #include "db/version_set.h"
+#include "table/table.h"
 #include "util/logging.h"
 #include "util/testharness.h"
 #include "util/testutil.h"
@@ -40,6 +40,7 @@ class CorruptionTest {
 
     db_ = nullptr;
     options_.create_if_missing = true;
+    options_.block_size_deviation = 0; // make unit test pass for now
     Reopen();
     options_.create_if_missing = false;
   }
@@ -55,6 +56,7 @@ class CorruptionTest {
     Options opt = (options ? *options : options_);
     opt.env = &env_;
     opt.block_cache = tiny_cache_;
+    opt.block_size_deviation = 0;
     return DB::Open(opt, dbname_, &db_);
   }
 

db/db_filesnapshot.cc

@@ -17,12 +17,14 @@ namespace leveldb {
 Status DBImpl::DisableFileDeletions() {
   MutexLock l(&mutex_);
   disable_delete_obsolete_files_ = true;
+  Log(options_.info_log, "File Deletions Disabled");
   return Status::OK();
 }
 
 Status DBImpl::EnableFileDeletions() {
   MutexLock l(&mutex_);
   disable_delete_obsolete_files_ = false;
+  Log(options_.info_log, "File Deletions Enabled");
   return Status::OK();
 }
 

db/db_impl.h

@@ -5,8 +5,10 @@
 #ifndef STORAGE_LEVELDB_DB_DB_IMPL_H_
 #define STORAGE_LEVELDB_DB_DB_IMPL_H_
 
+#include <atomic>
 #include <deque>
 #include <set>
+#include <vector>
 #include "db/dbformat.h"
 #include "db/log_file.h"
 #include "db/log_writer.h"
@@ -36,11 +38,16 @@ class DBImpl : public DB {
 
   // Implementations of the DB interface
   virtual Status Put(const WriteOptions&, const Slice& key, const Slice& value);
+  virtual Status Merge(const WriteOptions&, const Slice& key,
+                       const Slice& value);
   virtual Status Delete(const WriteOptions&, const Slice& key);
   virtual Status Write(const WriteOptions& options, WriteBatch* updates);
   virtual Status Get(const ReadOptions& options,
                      const Slice& key,
                      std::string* value);
+  virtual std::vector<Status> MultiGet(const ReadOptions& options,
+                                       const std::vector<Slice>& keys,
+                                       std::vector<std::string>* values);
   virtual Iterator* NewIterator(const ReadOptions&);
   virtual const Snapshot* GetSnapshot();
   virtual void ReleaseSnapshot(const Snapshot* snapshot);
@@ -61,7 +68,7 @@ class DBImpl : public DB {
 
   // Extra methods (for testing) that are not in the public DB interface
 
-  // Compact any files in the named level that overlap [*begin,*end]
+  // Compact any files in the named level that overlap [*begin, *end]
   void TEST_CompactRange(int level, const Slice* begin, const Slice* end);
 
   // Force current memtable contents to be compacted.
@@ -87,6 +94,10 @@ class DBImpl : public DB {
 
   // Return the current manifest file no.
   uint64_t TEST_Current_Manifest_FileNo();
+
+  // Trigger's a background call for testing.
+  void TEST_PurgeObsoleteteWAL();
+
  protected:
   Env* const env_;
   const std::string dbname_;
@@ -141,7 +152,7 @@ class DBImpl : public DB {
   // for the entire period. The second method WriteLevel0Table supports
   // concurrent flush memtables to storage.
   Status WriteLevel0TableForRecovery(MemTable* mem, VersionEdit* edit);
-  Status WriteLevel0Table(MemTable* mem, VersionEdit* edit,
+  Status WriteLevel0Table(std::vector<MemTable*> &mems, VersionEdit* edit,
                                 uint64_t* filenumber);
 
   Status MakeRoomForWrite(bool force /* compact even if there is room? */);
@@ -194,12 +205,18 @@ class DBImpl : public DB {
   Status FindProbableWALFiles(std::vector<LogFile>* const allLogs,
                               std::vector<LogFile>* const result,
                               const SequenceNumber target);
-
+  //  return true if
+  bool CheckFileExistsAndEmpty(const LogFile& file);
 
   Status ReadFirstRecord(const LogFile& file, WriteBatch* const result);
 
-
   Status ReadFirstLine(const std::string& fname, WriteBatch* const batch);
+
+  void PrintStatistics();
+
+  // dump leveldb.stats to LOG
+  void MaybeDumpStats();
+
   // Constant after construction
   const InternalFilterPolicy internal_filter_policy_;
   bool owns_info_log_;
@@ -223,7 +240,7 @@ class DBImpl : public DB {
 
   // Queue of writers.
   std::deque<Writer*> writers_;
-  WriteBatch* tmp_batch_;
+  WriteBatch tmp_batch_;
 
   SnapshotList snapshots_;
 
@@ -251,7 +268,7 @@ class DBImpl : public DB {
   // Have we encountered a background error in paranoid mode?
   Status bg_error_;
 
-  StatsLogger* logger_;
+  std::unique_ptr<StatsLogger> logger_;
 
   int64_t volatile last_log_ts;
 
@@ -261,6 +278,12 @@ class DBImpl : public DB {
   // last time when DeleteObsoleteFiles was invoked
   uint64_t delete_obsolete_files_last_run_;
 
+  // last time when PurgeObsoleteWALFiles ran.
+  uint64_t purge_wal_files_last_run_;
+
+  // last time stats were dumped to LOG
+  std::atomic<uint64_t> last_stats_dump_time_microsec_;
+
   // These count the number of microseconds for which MakeRoomForWrite stalls.
   uint64_t stall_level0_slowdown_;
   uint64_t stall_memtable_compaction_;
@@ -315,7 +338,20 @@ class DBImpl : public DB {
     }
   };
 
-  CompactionStats* stats_;
+  std::vector<CompactionStats> stats_;
+
+  // Used to compute per-interval statistics
+  struct StatsSnapshot {
+    uint64_t bytes_read_;
+    uint64_t bytes_written_;
+    uint64_t bytes_new_;
+    double   seconds_up_;
+
+    StatsSnapshot() : bytes_read_(0), bytes_written_(0),
+                      bytes_new_(0), seconds_up_(0) {}
+  };
+
+  StatsSnapshot last_stats_;
 
   static const int KEEP_LOG_FILE_NUM = 1000;
   std::string db_absolute_path_;
@@ -327,6 +363,9 @@ class DBImpl : public DB {
   // Used by transaction log iterator.
   SequenceNumber last_flushed_sequence_;
 
+  // The options to access storage files
+  const EnvOptions storage_options_;
+
   // No copying allowed
   DBImpl(const DBImpl&);
   void operator=(const DBImpl&);
@@ -334,9 +373,12 @@ class DBImpl : public DB {
   // dump the delayed_writes_ to the log file and reset counter.
   void DelayLoggingAndReset();
 
-  // find the earliest snapshot where seqno is visible
-  inline SequenceNumber findEarliestVisibleSnapshot(SequenceNumber in,
-    std::vector<SequenceNumber>& snapshots);
+  // Return the earliest snapshot where seqno is visible.
+  // Store the snapshot right before that, if any, in prev_snapshot
+  inline SequenceNumber findEarliestVisibleSnapshot(
+    SequenceNumber in,
+    std::vector<SequenceNumber>& snapshots,
+    SequenceNumber* prev_snapshot);
 };
 
 // Sanitize db options.  The caller should delete result.info_log if

db/db_impl_readonly.cc

@@ -24,11 +24,11 @@
 #include "leveldb/db.h"
 #include "leveldb/env.h"
 #include "leveldb/status.h"
-#include "leveldb/table.h"
 #include "leveldb/table_builder.h"
 #include "port/port.h"
 #include "table/block.h"
 #include "table/merger.h"
+#include "table/table.h"
 #include "table/two_level_iterator.h"
 #include "util/coding.h"
 #include "util/logging.h"
@@ -54,10 +54,10 @@ Status DBImplReadOnly::Get(const ReadOptions& options,
   Version* current = versions_->current();
   SequenceNumber snapshot = versions_->LastSequence();
   LookupKey lkey(key, snapshot);
-  if (mem->Get(lkey, value, &s)) {
+  if (mem->Get(lkey, value, &s, options_)) {
   } else {
     Version::GetStats stats;
-    s = current->Get(options, lkey, value, &stats);
+    current->Get(options, lkey, value, &s, &stats, options_);
   }
   return s;
 }
@@ -66,7 +66,7 @@ Iterator* DBImplReadOnly::NewIterator(const ReadOptions& options) {
   SequenceNumber latest_snapshot;
   Iterator* internal_iter = NewInternalIterator(options, &latest_snapshot);
   return NewDBIterator(
-      &dbname_, env_, user_comparator(), internal_iter,
+    &dbname_, env_, options_,  user_comparator(),internal_iter,
       (options.snapshot != nullptr
       ? reinterpret_cast<const SnapshotImpl*>(options.snapshot)->number_
       : latest_snapshot));

db/db_impl_readonly.h

@@ -37,6 +37,10 @@ public:
  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.");
  }

db/db_iter.cc

@@ -3,11 +3,14 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #include "db/db_iter.h"
+#include <stdexcept>
 
 #include "db/filename.h"
 #include "db/dbformat.h"
 #include "leveldb/env.h"
+#include "leveldb/options.h"
 #include "leveldb/iterator.h"
+#include "leveldb/merge_operator.h"
 #include "port/port.h"
 #include "util/logging.h"
 #include "util/mutexlock.h"
@@ -36,6 +39,7 @@ namespace {
 // numbers, deletion markers, overwrites, etc.
 class DBIter: public Iterator {
  public:
+  // The following is grossly complicated. TODO: clean it up
   // Which direction is the iterator currently moving?
   // (1) When moving forward, the internal iterator is positioned at
   //     the exact entry that yields this->key(), this->value()
@@ -46,27 +50,34 @@ class DBIter: public Iterator {
     kReverse
   };
 
-  DBIter(const std::string* dbname, Env* env,
+  DBIter(const std::string* dbname, Env* env, const Options& options,
          const Comparator* cmp, Iterator* iter, SequenceNumber s)
       : dbname_(dbname),
         env_(env),
+        logger_(options.info_log),
         user_comparator_(cmp),
+        user_merge_operator_(options.merge_operator),
         iter_(iter),
         sequence_(s),
         direction_(kForward),
-        valid_(false) {
+        valid_(false),
+        current_entry_is_merged_(false),
+        statistics_(options.statistics) {
+    RecordTick(statistics_, NO_ITERATORS, 1);
   }
   virtual ~DBIter() {
+    RecordTick(statistics_, NO_ITERATORS, -1);
     delete iter_;
   }
   virtual bool Valid() const { return valid_; }
   virtual Slice key() const {
     assert(valid_);
-    return (direction_ == kForward) ? ExtractUserKey(iter_->key()) : saved_key_;
+    return saved_key_;
   }
   virtual Slice value() const {
     assert(valid_);
-    return (direction_ == kForward) ? iter_->value() : saved_value_;
+    return (direction_ == kForward && !current_entry_is_merged_) ?
+      iter_->value() : saved_value_;
   }
   virtual Status status() const {
     if (status_.ok()) {
@@ -83,9 +94,10 @@ class DBIter: public Iterator {
   virtual void SeekToLast();
 
  private:
-  void FindNextUserEntry(bool skipping, std::string* skip);
+  void FindNextUserEntry(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());
@@ -102,15 +114,20 @@ class DBIter: public Iterator {
 
   const std::string* const dbname_;
   Env* const env_;
+  shared_ptr<Logger> logger_;
   const Comparator* const user_comparator_;
+  const MergeOperator* const user_merge_operator_;
   Iterator* const iter_;
   SequenceNumber const sequence_;
 
   Status status_;
   std::string saved_key_;     // == current key when direction_==kReverse
   std::string saved_value_;   // == current raw value when direction_==kReverse
+  std::string skip_key_;
   Direction direction_;
   bool valid_;
+  bool current_entry_is_merged_;
+  std::shared_ptr<Statistics> statistics_;
 
   // No copying allowed
   DBIter(const DBIter&);
@@ -120,6 +137,8 @@ class DBIter: public Iterator {
 inline bool DBIter::ParseKey(ParsedInternalKey* ikey) {
   if (!ParseInternalKey(iter_->key(), ikey)) {
     status_ = Status::Corruption("corrupted internal key in DBIter");
+    Log(logger_, "corrupted internal key in DBIter: %s",
+        iter_->key().ToString(true).c_str());
     return false;
   } else {
     return true;
@@ -146,47 +165,136 @@ void DBIter::Next() {
     }
   }
 
-  // Temporarily use saved_key_ as storage for key to skip.
-  std::string* skip = &saved_key_;
-  SaveKey(ExtractUserKey(iter_->key()), skip);
-  FindNextUserEntry(true, skip);
+  // If the current value is merged, we might already hit end of iter_
+  if (!iter_->Valid()) {
+    valid_ = false;
+    return;
+  }
+  FindNextUserEntry(true /* skipping the current user key */);
 }
 
-void DBIter::FindNextUserEntry(bool skipping, std::string* skip) {
+
+// PRE: saved_key_ has the current user key if skipping
+// POST: saved_key_ should have the next user key if valid_,
+//       if the current entry is a result of merge
+//           current_entry_is_merged_ => true
+//           saved_value_             => the merged value
+//
+// NOTE: In between, saved_key_ can point to a user key that has
+//       a delete marker
+void DBIter::FindNextUserEntry(bool skipping) {
   // Loop until we hit an acceptable entry to yield
   assert(iter_->Valid());
   assert(direction_ == kForward);
+  current_entry_is_merged_ = false;
   do {
     ParsedInternalKey ikey;
     if (ParseKey(&ikey) && ikey.sequence <= sequence_) {
-      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, skip);
-          skipping = true;
-          break;
-        case kTypeValue:
-          if (skipping &&
-              user_comparator_->Compare(ikey.user_key, *skip) <= 0) {
-            // Entry hidden
-          } else {
+      if (skipping &&
+          user_comparator_->Compare(ikey.user_key, saved_key_) <= 0) {
+        // skip this entry
+      } 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_);
+            skipping = true;
+            break;
+          case kTypeValue:
             valid_ = true;
-            saved_key_.clear();
+            SaveKey(ikey.user_key, &saved_key_);
             return;
-          }
-          break;
+          case kTypeMerge:
+            // By now, we are sure the current ikey is going to yield a value
+            SaveKey(ikey.user_key, &saved_key_);
+            current_entry_is_merged_ = true;
+            valid_ = true;
+            // Go to a different state machine
+            MergeValuesNewToOld();
+            // TODO: what if !iter_->Valid()
+            return;
+            break;
+        }
       }
     }
     iter_->Next();
   } while (iter_->Valid());
-  saved_key_.clear();
   valid_ = false;
 }
 
+// Merge values of the same user key starting from the current iter_ position
+// Scan from the newer entries to older entries.
+// PRE: iter_->key() points to the first merge type entry
+//      saved_key_ stores the user key
+// POST: saved_value_ has the merged value for the user key
+//       iter_ points to the next entry (or invalid)
+void DBIter::MergeValuesNewToOld() {
+
+  const Slice value = iter_->value();
+  std::string operand(value.data(), value.size());
+
+  ParsedInternalKey ikey;
+  for (iter_->Next(); iter_->Valid(); iter_->Next()) {
+    if (!ParseKey(&ikey)) {
+      // skip corrupted key
+      continue;
+    }
+
+    if (user_comparator_->Compare(ikey.user_key, saved_key_) != 0) {
+      // hit the next user key, stop right here
+      break;
+    }
+
+    if (kTypeDeletion == ikey.type) {
+      // hit a delete with the same user key, stop right here
+      // iter_ is positioned after delete
+      iter_->Next();
+      break;
+    }
+
+    if (kTypeValue == ikey.type) {
+      // hit a put, merge the put value with operand and store it in the
+      // final result saved_value_. We are done!
+      const Slice value = iter_->value();
+      user_merge_operator_->Merge(ikey.user_key, &value, Slice(operand),
+                                  &saved_value_, logger_.get());
+      // iter_ is positioned after put
+      iter_->Next();
+      return;
+    }
+
+    if (kTypeMerge == ikey.type) {
+      // hit a merge, merge the value with operand and continue.
+      // saved_value_ is used as a scratch area. The result is put
+      // back in operand
+      const Slice value = iter_->value();
+      user_merge_operator_->Merge(ikey.user_key, &value, operand,
+                                  &saved_value_, logger_.get());
+      swap(saved_value_, operand);
+    }
+  }
+
+  // we either exhausted all internal keys under this user key, or hit
+  // a deletion marker.
+  // feed null as the existing value to the merge opexrator, such that
+  // client can differentiate this scenario and do things accordingly.
+  user_merge_operator_->Merge(ikey.user_key, nullptr, operand,
+                              &saved_value_, logger_.get());
+}
+
 void DBIter::Prev() {
   assert(valid_);
 
+  // TODO: support backward iteration
+  // Throw an exception now if merge_operator is provided
+  if (user_merge_operator_) {
+    Log(logger_, "Prev not supported yet if merge_operator is provided");
+    throw std::logic_error("DBIter::Prev backward iteration not supported"
+                           " if merge_operator is provided");
+  }
+
   if (direction_ == kForward) {  // Switch directions?
     // iter_ is pointing at the current entry.  Scan backwards until
     // the key changes so we can use the normal reverse scanning code.
@@ -261,7 +369,7 @@ void DBIter::Seek(const Slice& target) {
       &saved_key_, ParsedInternalKey(target, sequence_, kValueTypeForSeek));
   iter_->Seek(saved_key_);
   if (iter_->Valid()) {
-    FindNextUserEntry(false, &saved_key_ /* temporary storage */);
+    FindNextUserEntry(false /*not skipping */);
   } else {
     valid_ = false;
   }
@@ -272,13 +380,21 @@ void DBIter::SeekToFirst() {
   ClearSavedValue();
   iter_->SeekToFirst();
   if (iter_->Valid()) {
-    FindNextUserEntry(false, &saved_key_ /* temporary storage */);
+    FindNextUserEntry(false /* not skipping */);
   } else {
     valid_ = false;
   }
 }
 
 void DBIter::SeekToLast() {
+  // TODO: support backward iteration
+  // throw an exception for now if merge_operator is provided
+  if (user_merge_operator_) {
+    Log(logger_, "SeekToLast not supported yet if merge_operator is provided");
+    throw std::logic_error("DBIter::SeekToLast: backward iteration not"
+                           " supported if merge_operator is provided");
+  }
+
   direction_ = kReverse;
   ClearSavedValue();
   iter_->SeekToLast();
@@ -290,10 +406,12 @@ void DBIter::SeekToLast() {
 Iterator* NewDBIterator(
     const std::string* dbname,
     Env* env,
-    const Comparator* user_key_comparator,
+    const Options& options,
+    const Comparator *user_key_comparator,
     Iterator* internal_iter,
     const SequenceNumber& sequence) {
-  return new DBIter(dbname, env, user_key_comparator, internal_iter, sequence);
+  return new DBIter(dbname, env, options, user_key_comparator,
+                    internal_iter, sequence);
 }
 
 }  // namespace leveldb

db/db_iter.h

@@ -17,7 +17,8 @@ namespace leveldb {
 extern Iterator* NewDBIterator(
     const std::string* dbname,
     Env* env,
-    const Comparator* user_key_comparator,
+    const Options& options,
+    const Comparator *user_key_comparator,
     Iterator* internal_iter,
     const SequenceNumber& sequence);
 

db/db_statistics.h

@@ -8,6 +8,7 @@
 #include <cassert>
 #include <stdlib.h>
 #include <vector>
+#include <memory>
 
 #include "leveldb/statistics.h"
 #include "util/histogram.h"
@@ -39,11 +40,6 @@ class DBStatistics: public Statistics {
     allHistograms_[histogramType].Add(value);
   }
 
-  virtual void measureTime(Histograms histogramType, double value) {
-    assert(histogramType < HISTOGRAM_ENUM_MAX);
-    allHistograms_[histogramType].Add(value);
-  }
-
   virtual void histogramData(Histograms histogramType,
                              HistogramData * const data) {
     assert(histogramType < HISTOGRAM_ENUM_MAX);
@@ -53,6 +49,11 @@ class DBStatistics: public Statistics {
   std::vector<Ticker> allTickers_;
   std::vector<HistogramImpl> allHistograms_;
 };
+
+std::shared_ptr<Statistics> CreateDBStatistics() {
+  return std::make_shared<DBStatistics>();
+}
+
 } // namespace leveldb
 
 #endif // LEVELDB_STORAGE_DB_DB_STATISTICS_H_

db/db_test.cc

@@ -12,13 +12,15 @@
 #include "db/version_set.h"
 #include "db/write_batch_internal.h"
 #include "leveldb/cache.h"
+#include "leveldb/compaction_filter.h"
 #include "leveldb/env.h"
-#include "leveldb/table.h"
+#include "table/table.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"
 
 namespace leveldb {
 
@@ -100,7 +102,8 @@ class SpecialEnv : public EnvWrapper {
     manifest_write_error_.Release_Store(nullptr);
    }
 
-  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r) {
+  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r,
+                         const EnvOptions& soptions) {
     class SSTableFile : public WritableFile {
      private:
       SpecialEnv* env_;
@@ -157,7 +160,7 @@ class SpecialEnv : public EnvWrapper {
       return Status::IOError("simulated write error");
     }
 
-    Status s = target()->NewWritableFile(f, r);
+    Status s = target()->NewWritableFile(f, r, soptions);
     if (s.ok()) {
       if (strstr(f.c_str(), ".sst") != nullptr) {
         r->reset(new SSTableFile(this, std::move(*r)));
@@ -169,7 +172,8 @@ class SpecialEnv : public EnvWrapper {
   }
 
   Status NewRandomAccessFile(const std::string& f,
-                             unique_ptr<RandomAccessFile>* r) {
+                             unique_ptr<RandomAccessFile>* r,
+                             const EnvOptions& soptions) {
     class CountingFile : public RandomAccessFile {
      private:
       unique_ptr<RandomAccessFile> target_;
@@ -186,7 +190,7 @@ class SpecialEnv : public EnvWrapper {
       }
     };
 
-    Status s = target()->NewRandomAccessFile(f, r);
+    Status s = target()->NewRandomAccessFile(f, r, soptions);
     if (s.ok() && count_random_reads_) {
       r->reset(new CountingFile(std::move(*r), &random_read_counter_));
     }
@@ -206,6 +210,7 @@ class DBTest {
   // Sequence of option configurations to try
   enum OptionConfig {
     kDefault,
+    kMergePut,
     kFilter,
     kUncompressed,
     kNumLevel_3,
@@ -217,6 +222,8 @@ class DBTest {
   };
   int option_config_;
 
+  std::shared_ptr<MergeOperator> merge_operator_;
+
  public:
   std::string dbname_;
   SpecialEnv* env_;
@@ -225,6 +232,7 @@ class DBTest {
   Options last_options_;
 
   DBTest() : option_config_(kDefault),
+             merge_operator_(MergeOperators::CreatePutOperator()),
              env_(new SpecialEnv(Env::Default())) {
     filter_policy_ = NewBloomFilterPolicy(10);
     dbname_ = test::TmpDir() + "/db_test";
@@ -256,6 +264,9 @@ class DBTest {
   Options CurrentOptions() {
     Options options;
     switch (option_config_) {
+      case kMergePut:
+        options.merge_operator = merge_operator_.get();
+        break;
       case kFilter:
         options.filter_policy = filter_policy_;
         break;
@@ -323,8 +334,12 @@ class DBTest {
     return DB::Open(opts, dbname_, &db_);
   }
 
-  Status Put(const std::string& k, const std::string& v) {
-    return db_->Put(WriteOptions(), k, v);
+  Status Put(const Slice& k, const Slice& v) {
+    if (kMergePut == option_config_ ) {
+      return db_->Merge(WriteOptions(), k, v);
+    } else {
+      return db_->Put(WriteOptions(), k, v);
+    }
   }
 
   Status Delete(const std::string& k) {
@@ -397,6 +412,10 @@ class DBTest {
             case kTypeValue:
               result += iter->value().ToString();
               break;
+            case kTypeMerge:
+              // keep it the same as kTypeValue for testing kMergePut
+              result += iter->value().ToString();
+              break;
             case kTypeDeletion:
               result += "DEL";
               break;
@@ -514,6 +533,26 @@ class DBTest {
     }
     return result;
   }
+
+  Options OptionsForLogIterTest() {
+    Options options = CurrentOptions();
+    options.create_if_missing = true;
+    options.WAL_ttl_seconds = 1000;
+    return options;
+  }
+
+  std::unique_ptr<TransactionLogIterator> OpenTransactionLogIter(
+    const SequenceNumber seq) {
+    unique_ptr<TransactionLogIterator> iter;
+    Status status = dbfull()->GetUpdatesSince(seq, &iter);
+    ASSERT_TRUE(status.ok());
+    ASSERT_TRUE(iter->Valid());
+    return std::move(iter);
+  }
+
+  std::string DummyString(size_t len, char c = 'a') {
+    return std::string(len, c);
+  }
 };
 
 TEST(DBTest, Empty) {
@@ -551,6 +590,7 @@ TEST(DBTest, LevelLimitReopen) {
   }
 
   options.num_levels = 1;
+  options.max_bytes_for_level_multiplier_additional.resize(1, 1);
   Status s = TryReopen(&options);
   ASSERT_EQ(s.IsCorruption(), true);
   ASSERT_EQ(s.ToString(),
@@ -558,13 +598,15 @@ TEST(DBTest, LevelLimitReopen) {
             "more levels than options.num_levels");
 
   options.num_levels = 10;
+  options.max_bytes_for_level_multiplier_additional.resize(10, 1);
   ASSERT_OK(TryReopen(&options));
 }
 
 TEST(DBTest, Preallocation) {
   const std::string src = dbname_ + "/alloc_test";
   unique_ptr<WritableFile> srcfile;
-  ASSERT_OK(env_->NewWritableFile(src, &srcfile));
+  const EnvOptions soptions;
+  ASSERT_OK(env_->NewWritableFile(src, &srcfile, soptions));
   srcfile->SetPreallocationBlockSize(1024 * 1024);
 
   // No writes should mean no preallocation
@@ -911,8 +953,11 @@ TEST(DBTest, IterMultiWithDelete) {
     Iterator* iter = db_->NewIterator(ReadOptions());
     iter->Seek("c");
     ASSERT_EQ(IterStatus(iter), "c->vc");
-    iter->Prev();
-    ASSERT_EQ(IterStatus(iter), "a->va");
+    if (!CurrentOptions().merge_operator) {
+      // TODO: merge operator does not support backward iteration yet
+      iter->Prev();
+      ASSERT_EQ(IterStatus(iter), "a->va");
+    }
     delete iter;
   } while (ChangeOptions());
 }
@@ -1326,40 +1371,69 @@ TEST(DBTest, RepeatedWritesToSameKey) {
 // kvs during the compaction process.
 static int cfilter_count;
 static std::string NEW_VALUE = "NewValue";
-static bool keep_filter(void* arg, int level, const Slice& key,
-  const Slice& value, Slice** new_value) {
-  assert(arg == nullptr);
-  cfilter_count++;
-  return false;
-}
-static bool delete_filter(void*argv, int level, const Slice& key,
-  const Slice& value, Slice** new_value) {
-  assert(argv == nullptr);
-  cfilter_count++;
-  return true;
-}
-static bool change_filter(void*argv, int level, const Slice& key,
-  const Slice& value, Slice** new_value) {
-  assert(argv == (void*)100);
-  assert(new_value != nullptr);
-  *new_value = new Slice(NEW_VALUE);
-  return false;
-}
+
+class KeepFilter : public CompactionFilter {
+ public:
+  virtual bool Filter(int level, const Slice& key,
+                      const Slice& value, std::string* new_value,
+                      bool* value_changed) const override {
+    cfilter_count++;
+    return false;
+  }
+
+  virtual const char* Name() const override {
+    return "KeepFilter";
+  }
+
+};
+
+class DeleteFilter : public CompactionFilter {
+ public:
+  virtual bool Filter(int level, const Slice& key,
+                      const Slice& value, std::string* new_value,
+                      bool* value_changed) const override {
+    cfilter_count++;
+    return true;
+  }
+
+  virtual const char* Name() const override {
+    return "DeleteFilter";
+  }
+};
+
+class ChangeFilter : public CompactionFilter {
+ public:
+  ChangeFilter(int argv) : argv_(argv) {}
+
+  virtual bool Filter(int level, const Slice& key,
+                      const Slice& value, std::string* new_value,
+                      bool* value_changed) const override {
+    assert(argv_ == 100);
+    assert(new_value != nullptr);
+    *new_value = NEW_VALUE;
+    *value_changed = true;
+    return false;
+  }
+
+  virtual const char* Name() const override {
+    return "ChangeFilter";
+  }
+
+ private:
+  const int argv_;
+};
 
 TEST(DBTest, CompactionFilter) {
   Options options = CurrentOptions();
   options.num_levels = 3;
   options.max_mem_compaction_level = 0;
-  options.CompactionFilter = keep_filter;
+  auto keep_filter = std::make_shared<KeepFilter>();
+  options.compaction_filter = keep_filter.get();
   Reopen(&options);
 
-  // Write 100K+1 keys, these are written to a few files
-  // in L0. We do this so that the current snapshot points
-  // to the 100001 key.The compaction filter is  not invoked
-  // on keys that are visible via a snapshot because we
-  // anyways cannot delete it.
+  // Write 100K keys, these are written to a few files in L0.
   const std::string value(10, 'x');
-  for (int i = 0; i < 100001; i++) {
+  for (int i = 0; i < 100000; i++) {
     char key[100];
     snprintf(key, sizeof(key), "B%010d", i);
     Put(key, value);
@@ -1386,6 +1460,7 @@ TEST(DBTest, CompactionFilter) {
   // has sequence number zero. The 100001st record
   // is at the tip of this snapshot and cannot
   // be zeroed out.
+  // TODO: figure out sequence number squashtoo
   int count = 0;
   int total = 0;
   Iterator* iter = dbfull()->TEST_NewInternalIterator();
@@ -1401,12 +1476,12 @@ TEST(DBTest, CompactionFilter) {
     }
     iter->Next();
   }
-  ASSERT_EQ(total, 100001);
+  ASSERT_EQ(total, 100000);
   ASSERT_EQ(count, 1);
   delete iter;
 
-  // overwrite all the 100K+1 keys once again.
-  for (int i = 0; i < 100001; i++) {
+  // overwrite all the 100K keys once again.
+  for (int i = 0; i < 100000; i++) {
     char key[100];
     snprintf(key, sizeof(key), "B%010d", i);
     Put(key, value);
@@ -1428,12 +1503,13 @@ TEST(DBTest, CompactionFilter) {
 
   // create a new database with the compaction
   // filter in such a way that it deletes all keys
-  options.CompactionFilter = delete_filter;
+  auto delete_filter = std::make_shared<DeleteFilter>();
+  options.compaction_filter = delete_filter.get();
   options.create_if_missing = true;
   DestroyAndReopen(&options);
 
   // write all the keys once again.
-  for (int i = 0; i < 100001; i++) {
+  for (int i = 0; i < 100000; i++) {
     char key[100];
     snprintf(key, sizeof(key), "B%010d", i);
     Put(key, value);
@@ -1456,10 +1532,7 @@ TEST(DBTest, CompactionFilter) {
   ASSERT_EQ(NumTableFilesAtLevel(0), 0);
   ASSERT_EQ(NumTableFilesAtLevel(1), 0);
 
-  // Scan the entire database to ensure that only the
-  // 100001th key is left in the db. The 100001th key
-  // is part of the default-most-current snapshot and
-  // cannot be deleted.
+  // Scan the entire database to ensure that nothing is left
   iter = db_->NewIterator(ReadOptions());
   iter->SeekToFirst();
   count = 0;
@@ -1467,12 +1540,14 @@ TEST(DBTest, CompactionFilter) {
     count++;
     iter->Next();
   }
-  ASSERT_EQ(count, 1);
+  ASSERT_EQ(count, 0);
   delete iter;
 
   // The sequence number of the remaining record
   // is not zeroed out even though it is at the
   // level Lmax because this record is at the tip
+  // TODO: remove the following or design a different
+  // test
   count = 0;
   iter = dbfull()->TEST_NewInternalIterator();
   iter->SeekToFirst();
@@ -1480,11 +1555,11 @@ TEST(DBTest, CompactionFilter) {
   while (iter->Valid()) {
     ParsedInternalKey ikey;
     ASSERT_EQ(ParseInternalKey(iter->key(), &ikey), true);
-    ASSERT_NE(ikey.sequence, 0);
+    ASSERT_NE(ikey.sequence, (unsigned)0);
     count++;
     iter->Next();
   }
-  ASSERT_EQ(count, 1);
+  ASSERT_EQ(count, 0);
   delete iter;
 }
 
@@ -1492,8 +1567,8 @@ TEST(DBTest, CompactionFilterWithValueChange) {
   Options options = CurrentOptions();
   options.num_levels = 3;
   options.max_mem_compaction_level = 0;
-  options.compaction_filter_args = (void *)100;
-  options.CompactionFilter = change_filter;
+  auto change_filter = std::make_shared<ChangeFilter>(100);
+  options.compaction_filter = change_filter.get();
   Reopen(&options);
 
   // Write 100K+1 keys, these are written to a few files
@@ -2307,6 +2382,7 @@ TEST(DBTest, BloomFilter) {
 
 TEST(DBTest, SnapshotFiles) {
   Options options = CurrentOptions();
+  const EnvOptions soptions;
   options.write_buffer_size = 100000000;        // Large write buffer
   Reopen(&options);
 
@@ -2360,9 +2436,9 @@ TEST(DBTest, SnapshotFiles) {
       }
     }
     unique_ptr<SequentialFile> srcfile;
-    ASSERT_OK(env_->NewSequentialFile(src, &srcfile));
+    ASSERT_OK(env_->NewSequentialFile(src, &srcfile, soptions));
     unique_ptr<WritableFile> destfile;
-    ASSERT_OK(env_->NewWritableFile(dest, &destfile));
+    ASSERT_OK(env_->NewWritableFile(dest, &destfile, soptions));
 
     char buffer[4096];
     Slice slice;
@@ -2518,20 +2594,20 @@ TEST(DBTest, CompactOnFlush) {
   db_->ReleaseSnapshot(snapshot1);
 }
 
-void ListLogFiles(Env* env,
-                  const std::string& path,
-                  std::vector<uint64_t>* logFiles) {
+std::vector<std::uint64_t> ListLogFiles(Env* env, const std::string& path) {
   std::vector<std::string> files;
+  std::vector<uint64_t> log_files;
   env->GetChildren(path, &files);
   uint64_t number;
   FileType type;
   for (size_t i = 0; i < files.size(); ++i) {
     if (ParseFileName(files[i], &number, &type)) {
       if (type == kLogFile) {
-        logFiles->push_back(number);
+        log_files.push_back(number);
       }
     }
   }
+  return std::move(log_files);
 }
 
 TEST(DBTest, WALArchival) {
@@ -2550,94 +2626,157 @@ TEST(DBTest, WALArchival) {
   std::string archiveDir = ArchivalDirectory(dbname_);
 
   for (int i = 0; i < 10; ++i) {
-
     for (int j = 0; j < 10; ++j) {
       ASSERT_OK(Put(Key(10*i+j), value));
     }
 
-    std::vector<uint64_t> logFiles;
-    ListLogFiles(env_, dbname_, &logFiles);
+    std::vector<uint64_t> logFiles = ListLogFiles(env_, dbname_);
 
     options.create_if_missing = false;
     Reopen(&options);
 
-    std::vector<uint64_t> logs;
-    ListLogFiles(env_, archiveDir, &logs);
+    std::vector<uint64_t> logs = ListLogFiles(env_, archiveDir);
     std::set<uint64_t> archivedFiles(logs.begin(), logs.end());
 
-    for (std::vector<uint64_t>::iterator it = logFiles.begin();
-         it != logFiles.end();
-         ++it) {
-      ASSERT_TRUE(archivedFiles.find(*it) != archivedFiles.end());
+    for (auto& log : logFiles) {
+      ASSERT_TRUE(archivedFiles.find(log) != archivedFiles.end());
     }
   }
 
-  // REOPEN database with 0 TTL. all files should have been deleted.
-  std::vector<uint64_t> logFiles;
-  ListLogFiles(env_, archiveDir, &logFiles);
+  std::vector<uint64_t> logFiles = ListLogFiles(env_, archiveDir);
   ASSERT_TRUE(logFiles.size() > 0);
   options.WAL_ttl_seconds = 1;
   env_->SleepForMicroseconds(2*1000*1000);
   Reopen(&options);
 
-  logFiles.clear();
-  ListLogFiles(env_, archiveDir, &logFiles);
+  logFiles = ListLogFiles(env_, archiveDir);
   ASSERT_TRUE(logFiles.size() == 0);
 
 }
 
-TEST(DBTest, TransactionLogIterator) {
-  std::string value(1024, '1');
+TEST(DBTest, WALClear) {
   Options options = CurrentOptions();
   options.create_if_missing = true;
-  options.WAL_ttl_seconds = 1000;
+  options.WAL_ttl_seconds = 1;
+
+  for (int j = 0; j < 10; ++j)
+  for (int i = 0; i < 10; ++i)
+    ASSERT_OK(Put(Key(10*i+j), DummyString(1024)));
+  Reopen(&options);
+  std::string archive_dir = ArchivalDirectory(dbname_);
+  std::vector<std::uint64_t> log_files = ListLogFiles(env_, archive_dir);
+  ASSERT_TRUE(!log_files.empty());
+  env_->SleepForMicroseconds(2 * 1000 * 1000);
+  dbfull()->TEST_PurgeObsoleteteWAL();
+  log_files = ListLogFiles(env_, archive_dir);
+  ASSERT_TRUE(log_files.empty());
+}
+
+void ExpectRecords(
+  const int expected_no_records,
+  std::unique_ptr<TransactionLogIterator>& iter) {
+  int i = 0;
+  SequenceNumber lastSequence = 0;
+  while (iter->Valid()) {
+    BatchResult res = iter->GetBatch();
+    ASSERT_TRUE(res.sequence > lastSequence);
+    ++i;
+    lastSequence = res.sequence;
+    ASSERT_TRUE(iter->status().ok());
+    iter->Next();
+  }
+  ASSERT_EQ(i, expected_no_records);
+}
+
+TEST(DBTest, TransactionLogIterator) {
+  Options options = OptionsForLogIterTest();
   DestroyAndReopen(&options);
-  Put("key1", value);
-  Put("key2", value);
-  Put("key2", value);
+  Put("key1", DummyString(1024));
+  Put("key2", DummyString(1024));
+  Put("key2", DummyString(1024));
   ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3U);
   {
-    unique_ptr<TransactionLogIterator> iter;
-    Status status = dbfull()->GetUpdatesSince(0, &iter);
-    ASSERT_TRUE(status.ok());
-    ASSERT_TRUE(iter->Valid());
-    int i = 0;
-    SequenceNumber lastSequence = 0;
-    while (iter->Valid()) {
-      BatchResult res = iter->GetBatch();
-      ASSERT_TRUE(res.sequence > lastSequence);
-      ++i;
-      lastSequence = res.sequence;
-      ASSERT_TRUE(iter->status().ok());
-      iter->Next();
-    }
-    ASSERT_EQ(i, 3);
+    auto iter = OpenTransactionLogIter(0);
+    ExpectRecords(3, iter);
   }
   Reopen(&options);
-  {
-    Put("key4", value);
-    Put("key5", value);
-    Put("key6", value);
+    env_->SleepForMicroseconds(2 * 1000 * 1000);{
+    Put("key4", DummyString(1024));
+    Put("key5", DummyString(1024));
+    Put("key6", DummyString(1024));
   }
   {
-    unique_ptr<TransactionLogIterator> iter;
-    Status status = dbfull()->GetUpdatesSince(0, &iter);
-    ASSERT_TRUE(status.ok());
-    ASSERT_TRUE(iter->Valid());
-    int i = 0;
-    SequenceNumber lastSequence = 0;
-    while (iter->Valid()) {
-      BatchResult res = iter->GetBatch();
-      ASSERT_TRUE(res.sequence > lastSequence);
-      lastSequence = res.sequence;
-      ASSERT_TRUE(iter->status().ok());
-      iter->Next();
-      ++i;
-    }
-    ASSERT_EQ(i, 6);
+    auto iter = OpenTransactionLogIter(0);
+    ExpectRecords(6, iter);
   }
 }
 
+TEST(DBTest, TransactionLogIteratorMoveOverZeroFiles) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  // Do a plain Reopen.
+  Put("key1", DummyString(1024));
+  // Two reopens should create a zero record WAL file.
+  Reopen(&options);
+  Reopen(&options);
+
+  Put("key2", DummyString(1024));
+
+  auto iter = OpenTransactionLogIter(0);
+  ExpectRecords(2, iter);
+}
+
+TEST(DBTest, TransactionLogIteratorStallAtLastRecord) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  Put("key1", DummyString(1024));
+  auto iter = OpenTransactionLogIter(0);
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+  iter->Next();
+  ASSERT_TRUE(!iter->Valid());
+  ASSERT_OK(iter->status());
+  Put("key2", DummyString(1024));
+  iter->Next();
+  ASSERT_OK(iter->status());
+  ASSERT_TRUE(iter->Valid());
+}
+
+TEST(DBTest, TransactionLogIteratorJustEmptyFile) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  unique_ptr<TransactionLogIterator> iter;
+  Status status = dbfull()->GetUpdatesSince(0, &iter);
+  ASSERT_TRUE(!status.ok());
+}
+
+TEST(DBTest, TransactionLogIteratorCheckAfterRestart) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  Put("key1", DummyString(1024));
+  Put("key2", DummyString(1023));
+  dbfull()->Flush(FlushOptions());
+  Reopen(&options);
+  auto iter = OpenTransactionLogIter(0);
+  ExpectRecords(2, iter);
+}
+
+TEST(DBTest, TransactionLogIteratorBatchOperations) {
+  Options options = OptionsForLogIterTest();
+  DestroyAndReopen(&options);
+  WriteBatch batch;
+  batch.Put("key1", DummyString(1024));
+  batch.Put("key2", DummyString(1024));
+  batch.Put("key3", DummyString(1024));
+  batch.Delete("key2");
+  dbfull()->Write(WriteOptions(), &batch);
+  dbfull()->Flush(FlushOptions());
+  Reopen(&options);
+  Put("key4", DummyString(1024));
+  auto iter = OpenTransactionLogIter(3);
+  ExpectRecords(1, iter);
+}
+
 TEST(DBTest, ReadCompaction) {
   std::string value(4096, '4'); // a string of size 4K
   {
@@ -2740,7 +2879,7 @@ static void MTThreadBody(void* arg) {
       // We add some padding for force compactions.
       snprintf(valbuf, sizeof(valbuf), "%d.%d.%-1000d",
                key, id, static_cast<int>(counter));
-      ASSERT_OK(db->Put(WriteOptions(), Slice(keybuf), Slice(valbuf)));
+      ASSERT_OK(t->state->test->Put(Slice(keybuf), Slice(valbuf)));
     } else {
       // Read a value and verify that it matches the pattern written above.
       Status s = db->Get(ReadOptions(), Slice(keybuf), &value);
@@ -2813,13 +2952,23 @@ class ModelDB: public DB {
   virtual Status Put(const WriteOptions& o, const Slice& k, const Slice& v) {
     return DB::Put(o, k, v);
   }
+  virtual Status Merge(const WriteOptions& o, const Slice& k, const Slice& v) {
+    return DB::Merge(o, k, v);
+  }
   virtual Status Delete(const WriteOptions& o, const Slice& key) {
     return DB::Delete(o, key);
   }
   virtual Status Get(const ReadOptions& options,
                      const Slice& key, std::string* value) {
-    assert(false);      // Not implemented
-    return Status::NotFound(key);
+    return Status::NotSupported(key);
+  }
+
+  virtual std::vector<Status> MultiGet(const ReadOptions& options,
+                                       const std::vector<Slice>& keys,
+                                       std::vector<std::string>* values) {
+    std::vector<Status> s(keys.size(),
+                          Status::NotSupported("Not implemented."));
+    return s;
   }
   virtual Iterator* NewIterator(const ReadOptions& options) {
     if (options.snapshot == nullptr) {
@@ -2848,6 +2997,10 @@ class ModelDB: public DB {
       virtual void Put(const Slice& key, const Slice& value) {
         (*map_)[key.ToString()] = value.ToString();
       }
+      virtual void Merge(const Slice& key, const Slice& value) {
+        // ignore merge for now
+        //(*map_)[key.ToString()] = value.ToString();
+      }
       virtual void Delete(const Slice& key) {
         map_->erase(key.ToString());
       }
@@ -3063,6 +3216,61 @@ TEST(DBTest, Randomized) {
   } while (ChangeOptions());
 }
 
+TEST(DBTest, MultiGetSimple) {
+  ASSERT_OK(db_->Put(WriteOptions(),"k1","v1"));
+  ASSERT_OK(db_->Put(WriteOptions(),"k2","v2"));
+  ASSERT_OK(db_->Put(WriteOptions(),"k3","v3"));
+  ASSERT_OK(db_->Put(WriteOptions(),"k4","v4"));
+  ASSERT_OK(db_->Delete(WriteOptions(),"k4"));
+  ASSERT_OK(db_->Put(WriteOptions(),"k5","v5"));
+  ASSERT_OK(db_->Delete(WriteOptions(),"no_key"));
+
+  std::vector<Slice> keys(6);
+  keys[0] = "k1";
+  keys[1] = "k2";
+  keys[2] = "k3";
+  keys[3] = "k4";
+  keys[4] = "k5";
+  keys[5] = "no_key";
+
+  std::vector<std::string> values(20,"Temporary data to be overwritten");
+
+  std::vector<Status> s =  db_->MultiGet(ReadOptions(),keys,&values);
+  ASSERT_EQ(values.size(),keys.size());
+  ASSERT_EQ(values[0], "v1");
+  ASSERT_EQ(values[1], "v2");
+  ASSERT_EQ(values[2], "v3");
+  ASSERT_EQ(values[4], "v5");
+
+  ASSERT_OK(s[0]);
+  ASSERT_OK(s[1]);
+  ASSERT_OK(s[2]);
+  ASSERT_TRUE(s[3].IsNotFound());
+  ASSERT_OK(s[4]);
+  ASSERT_TRUE(s[5].IsNotFound());
+}
+
+TEST(DBTest, MultiGetEmpty) {
+  // Empty Key Set
+  std::vector<Slice> keys;
+  std::vector<std::string> values;
+  std::vector<Status> s = db_->MultiGet(ReadOptions(),keys,&values);
+  ASSERT_EQ((int)s.size(),0);
+
+  // Empty Database, Empty Key Set
+  DestroyAndReopen();
+  s = db_->MultiGet(ReadOptions(), keys, &values);
+  ASSERT_EQ((int)s.size(),0);
+
+  // Empty Database, Search for Keys
+  keys.resize(2);
+  keys[0] = "a";
+  keys[1] = "b";
+  s = db_->MultiGet(ReadOptions(),keys,&values);
+  ASSERT_EQ((int)s.size(), 2);
+  ASSERT_TRUE(s[0].IsNotFound() && s[1].IsNotFound());
+}
+
 std::string MakeKey(unsigned int num) {
   char buf[30];
   snprintf(buf, sizeof(buf), "%016u", num);
@@ -3090,7 +3298,8 @@ void BM_LogAndApply(int iters, int num_base_files) {
 
   InternalKeyComparator cmp(BytewiseComparator());
   Options options;
-  VersionSet vset(dbname, &options, nullptr, &cmp);
+  EnvOptions sopt;
+  VersionSet vset(dbname, &options, sopt, nullptr, &cmp);
   ASSERT_OK(vset.Recover());
   VersionEdit vbase(vset.NumberLevels());
   uint64_t fnum = 1;

db/dbformat.cc

@@ -44,7 +44,7 @@ std::string InternalKey::DebugString(bool hex) const {
 }
 
 const char* InternalKeyComparator::Name() const {
-  return "leveldb.InternalKeyComparator";
+  return name_.c_str();
 }
 
 int InternalKeyComparator::Compare(const Slice& akey, const Slice& bkey) const {

db/dbformat.h

@@ -24,7 +24,8 @@ class InternalKey;
 // data structures.
 enum ValueType {
   kTypeDeletion = 0x0,
-  kTypeValue = 0x1
+  kTypeValue = 0x1,
+  kTypeMerge = 0x2
 };
 // kValueTypeForSeek defines the ValueType that should be passed when
 // constructing a ParsedInternalKey object for seeking to a particular
@@ -32,7 +33,7 @@ enum ValueType {
 // and the value type is embedded as the low 8 bits in the sequence
 // number in internal keys, we need to use the highest-numbered
 // ValueType, not the lowest).
-static const ValueType kValueTypeForSeek = kTypeValue;
+static const ValueType kValueTypeForSeek = kTypeMerge;
 
 // We leave eight bits empty at the bottom so a type and sequence#
 // can be packed together into 64-bits.
@@ -85,8 +86,13 @@ inline ValueType ExtractValueType(const Slice& internal_key) {
 class InternalKeyComparator : public Comparator {
  private:
   const Comparator* user_comparator_;
+  std::string name_;
  public:
-  explicit InternalKeyComparator(const Comparator* c) : user_comparator_(c) { }
+  explicit InternalKeyComparator(const Comparator* c) : user_comparator_(c),
+    name_("leveldb.InternalKeyComparator:" +
+          std::string(user_comparator_->Name())) {
+  }
+
   virtual const char* Name() const;
   virtual int Compare(const Slice& a, const Slice& b) const;
   virtual void FindShortestSeparator(
@@ -154,7 +160,7 @@ inline bool ParseInternalKey(const Slice& internal_key,
   result->sequence = num >> 8;
   result->type = static_cast<ValueType>(c);
   result->user_key = Slice(internal_key.data(), n - 8);
-  return (c <= static_cast<unsigned char>(kTypeValue));
+  return (c <= static_cast<unsigned char>(kValueTypeForSeek));
 }
 
 // Update the sequence number in the internal key

db/log_format.h

@@ -24,7 +24,7 @@ enum RecordType {
 };
 static const int kMaxRecordType = kLastType;
 
-static const int kBlockSize = 32768;
+static const unsigned int kBlockSize = 32768;
 
 // Header is checksum (4 bytes), type (1 byte), length (2 bytes).
 static const int kHeaderSize = 4 + 1 + 2;

db/log_reader.h

@@ -59,6 +59,17 @@ class Reader {
   // Undefined before the first call to ReadRecord.
   uint64_t LastRecordOffset();
 
+  // returns true if the reader has encountered an eof condition.
+  bool IsEOF() {
+    return eof_;
+  }
+
+  // when we know more data has been written to the file. we can use this
+  // function to force the reader to look again in the file.
+  void UnmarkEOF() {
+    eof_ = false;
+  }
+
   SequentialFile* file() { return file_.get(); }
 
  private:

db/log_test.cc

@@ -450,13 +450,13 @@ TEST(LogTest, ErrorJoinsRecords) {
   Write("correct");
 
   // Wipe the middle block
-  for (int offset = kBlockSize; offset < 2*kBlockSize; offset++) {
+  for (unsigned int offset = kBlockSize; offset < 2*kBlockSize; offset++) {
     SetByte(offset, 'x');
   }
 
   ASSERT_EQ("correct", Read());
   ASSERT_EQ("EOF", Read());
-  const int dropped = DroppedBytes();
+  const unsigned int dropped = DroppedBytes();
   ASSERT_LE(dropped, 2*kBlockSize + 100);
   ASSERT_GE(dropped, 2*kBlockSize);
 }

db/memtable.cc

@@ -7,6 +7,7 @@
 #include "leveldb/comparator.h"
 #include "leveldb/env.h"
 #include "leveldb/iterator.h"
+#include "leveldb/merge_operator.h"
 #include "util/coding.h"
 
 namespace leveldb {
@@ -26,7 +27,8 @@ MemTable::MemTable(const InternalKeyComparator& cmp, int numlevel)
       flush_completed_(false),
       file_number_(0),
       edit_(numlevel),
-      first_seqno_(0) {
+      first_seqno_(0),
+      mem_logfile_number_(0) {
 }
 
 MemTable::~MemTable() {
@@ -116,11 +118,23 @@ void MemTable::Add(SequenceNumber s, ValueType type,
   }
 }
 
-bool MemTable::Get(const LookupKey& key, std::string* value, Status* s) {
+bool MemTable::Get(const LookupKey& key, std::string* value, Status* s,
+                  const Options& options) {
   Slice memkey = key.memtable_key();
   Table::Iterator iter(&table_);
   iter.Seek(memkey.data());
-  if (iter.Valid()) {
+
+  bool merge_in_progress = false;
+  std::string operand;
+  if (s->IsMergeInProgress()) {
+    swap(*value, operand);
+    merge_in_progress = true;
+  }
+
+
+  auto merge_operator = options.merge_operator;
+  auto logger = options.info_log;
+  for (; iter.Valid(); iter.Next()) {
     // entry format is:
     //    klength  varint32
     //    userkey  char[klength-8]
@@ -141,15 +155,47 @@ bool MemTable::Get(const LookupKey& key, std::string* value, Status* s) {
       switch (static_cast<ValueType>(tag & 0xff)) {
         case kTypeValue: {
           Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
-          value->assign(v.data(), v.size());
+          if (merge_in_progress) {
+            merge_operator->Merge(key.user_key(), &v, operand,
+                                   value, logger.get());
+          } else {
+            value->assign(v.data(), v.size());
+          }
           return true;
         }
-        case kTypeDeletion:
-          *s = Status::NotFound(Slice());
+        case kTypeMerge: {
+          Slice v = GetLengthPrefixedSlice(key_ptr + key_length);
+          if (merge_in_progress) {
+            merge_operator->Merge(key.user_key(), &v, operand,
+                                  value, logger.get());
+            swap(*value, operand);
+          } else {
+            assert(merge_operator);
+            merge_in_progress = true;
+            operand.assign(v.data(), v.size());
+          }
+          break;
+        }
+        case kTypeDeletion: {
+          if (merge_in_progress) {
+            merge_operator->Merge(key.user_key(), nullptr, operand,
+                                   value, logger.get());
+          } else {
+            *s = Status::NotFound(Slice());
+          }
           return true;
+        }
       }
+    } else {
+      // exit loop if user key does not match
+      break;
     }
   }
+
+  if (merge_in_progress) {
+    swap(*value, operand);
+    *s = Status::MergeInProgress("");
+  }
   return false;
 }
 

db/memtable.h

@@ -62,8 +62,13 @@ class MemTable {
   // If memtable contains a value for key, store it in *value and return true.
   // If memtable contains a deletion for key, store a NotFound() error
   // in *status and return true.
+  // If memtable contains Merge operation as the most recent entry for a key,
+  //   and the merge process does not stop (not reaching a value or delete),
+  //   store the current merged result in value and MergeInProgress in s.
+  //   return false
   // Else, return false.
-  bool Get(const LookupKey& key, std::string* value, Status* s);
+  bool Get(const LookupKey& key, std::string* value, Status* s,
+          const Options& options);
 
   // Returns the edits area that is needed for flushing the memtable
   VersionEdit* GetEdits() { return &edit_; }
@@ -72,6 +77,14 @@ class MemTable {
   // into the memtable
   SequenceNumber GetFirstSequenceNumber() { return first_seqno_; }
 
+  // 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; }
+
  private:
   ~MemTable();  // Private since only Unref() should be used to delete it
 
@@ -103,6 +116,9 @@ class MemTable {
   // The sequence number of the kv that was inserted first
   SequenceNumber first_seqno_;
 
+  // The log files earlier than this number can be deleted.
+  uint64_t mem_logfile_number_;
+
   // No copying allowed
   MemTable(const MemTable&);
   void operator=(const MemTable&);

db/memtablelist.cc

@@ -43,17 +43,16 @@ int MemTableList::size() {
 
 // Returns true if there is at least one memtable on which flush has
 // not yet started.
-bool MemTableList::IsFlushPending() {
-  if (num_flush_not_started_ > 0) {
+bool MemTableList::IsFlushPending(int min_write_buffer_number_to_merge) {
+  if (num_flush_not_started_ >= min_write_buffer_number_to_merge) {
     assert(imm_flush_needed.NoBarrier_Load() != nullptr);
     return true;
   }
   return false;
 }
 
-// Returns the earliest memtable that needs to be flushed.
-// Returns null, if no such memtable exist.
-MemTable* MemTableList::PickMemtableToFlush() {
+// Returns the memtables that need to be flushed.
+void MemTableList::PickMemtablesToFlush(std::vector<MemTable*>* ret) {
   for (list<MemTable*>::reverse_iterator it = memlist_.rbegin();
        it != memlist_.rend(); it++) {
     MemTable* m = *it;
@@ -64,37 +63,48 @@ MemTable* MemTableList::PickMemtableToFlush() {
         imm_flush_needed.Release_Store(nullptr);
       }
       m->flush_in_progress_ = true; // flushing will start very soon
-      return m;
+      ret->push_back(m);
     }
   }
-  return nullptr;
 }
 
 // Record a successful flush in the manifest file
-Status MemTableList::InstallMemtableFlushResults(MemTable* m,
+Status MemTableList::InstallMemtableFlushResults(
+                      const std::vector<MemTable*> &mems,
                       VersionSet* vset, Status flushStatus,
                       port::Mutex* mu, Logger* info_log,
                       uint64_t file_number,
                       std::set<uint64_t>& pending_outputs) {
   mu->AssertHeld();
-  assert(m->flush_in_progress_);
-  assert(m->file_number_ == 0);
 
   // If the flush was not successful, then just reset state.
   // Maybe a suceeding attempt to flush will be successful.
   if (!flushStatus.ok()) {
-    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);
+    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
-  m->flush_completed_ = true;
-  m->file_number_ = file_number;
+  bool first = true;
+  for (MemTable* m : mems) {
+
+    // All the edits are associated with the first memtable of this batch.
+    assert(first || m->GetEdits()->NumEntries() == 0);
+    first = false;
+
+    m->flush_completed_ = true;
+    m->file_number_ = file_number;
+  }
 
   // if some other thread is already commiting, then return
   Status s;
@@ -106,12 +116,15 @@ Status MemTableList::InstallMemtableFlushResults(MemTable* m,
   commit_in_progress_ = true;
 
   // scan all memtables from the earliest, and commit those
-  // (in that order) that have finished flushing.
-  while (!memlist_.empty()) {
-    m = memlist_.back(); // get the last element
+  // (in that order) that have finished flushing. Memetables
+  // are always committed in the order that they were created.
+  while (!memlist_.empty() && s.ok()) {
+    MemTable* m = memlist_.back(); // get the last element
     if (!m->flush_completed_) {
       break;
     }
+    first = true;
+
     Log(info_log,
         "Level-0 commit table #%llu: started",
         (unsigned long long)m->file_number_);
@@ -119,33 +132,39 @@ Status MemTableList::InstallMemtableFlushResults(MemTable* m,
     // this can release and reacquire the mutex.
     s = vset->LogAndApply(&m->edit_, mu);
 
-    if (s.ok()) { // commit new state
-      Log(info_log, "Level-0 commit table #%llu: done",
-                     (unsigned long long)m->file_number_);
-      memlist_.remove(m);
-      assert(m->file_number_ > 0);
+    // All the later memtables that have the same filenum
+    // are part of the same batch. They can be committed now.
+    do {
+      if (s.ok()) { // commit new state
+        Log(info_log, "Level-0 commit table #%llu: done %s",
+                       (unsigned long long)m->file_number_,
+                       first ? "": "bulk");
+        memlist_.remove(m);
+        assert(m->file_number_ > 0);
 
-      // pending_outputs can be cleared only after the newly created file
-      // has been written to a committed version so that other concurrently
-      // executing compaction threads do not mistakenly assume that this
-      // file is not live.
-      pending_outputs.erase(m->file_number_);
-      m->Unref();
-      size_--;
-    } else {
-      //commit failed. setup state so that we can flush again.
-      Log(info_log, "Level-0 commit table #%llu: failed",
-                     (unsigned long long)m->file_number_);
-      m->flush_completed_ = false;
-      m->flush_in_progress_ = false;
-      m->edit_.Clear();
-      num_flush_not_started_++;
-      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");
-      break;
-    }
+        // pending_outputs can be cleared only after the newly created file
+        // has been written to a committed version so that other concurrently
+        // executing compaction threads do not mistakenly assume that this
+        // file is not live.
+        pending_outputs.erase(m->file_number_);
+        m->Unref();
+        size_--;
+      } else {
+        //commit failed. setup state so that we can flush again.
+        Log(info_log, "Level-0 commit table #%llu: failed",
+                       (unsigned long long)m->file_number_);
+        m->flush_completed_ = false;
+        m->flush_in_progress_ = false;
+        m->edit_.Clear();
+        num_flush_not_started_++;
+        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");
+      }
+      first = false;
+    } while (!memlist_.empty() && (m = memlist_.back()) &&
+             m->file_number_ == file_number);
   }
   commit_in_progress_ = false;
   return s;
@@ -174,10 +193,11 @@ size_t MemTableList::ApproximateMemoryUsage() {
 
 // Search all the memtables starting from the most recent one.
 // Return the most recent value found, if any.
-bool MemTableList::Get(const LookupKey& key, std::string* value, Status* s) {
+bool MemTableList::Get(const LookupKey& key, std::string* value, Status* s,
+                      const Options& options ) {
   for (list<MemTable*>::iterator it = memlist_.begin();
        it != memlist_.end(); ++it) {
-    if ((*it)->Get(key, value, s)) {
+    if ((*it)->Get(key, value, s, options)) {
       return true;
     }
   }

db/memtablelist.h

@@ -18,7 +18,7 @@ class Mutex;
 class MemTableListIterator;
 
 //
-// This class stores refeernces to all the immutable memtables.
+// This class stores references to all the immutable memtables.
 // The memtables are flushed to L0 as soon as possible and in
 // any order. If there are more than one immutable memtable, their
 // flushes can occur concurrently.  However, they are 'committed'
@@ -49,14 +49,13 @@ class MemTableList {
 
   // Returns true if there is at least one memtable on which flush has
   // not yet started.
-  bool IsFlushPending();
+  bool IsFlushPending(int min_write_buffer_number_to_merge);
 
-  // Returns the earliest memtable that needs to be flushed.
-  // Returns null, if no such memtable exist.
-  MemTable* PickMemtableToFlush();
+  // Returns the earliest memtables that needs to be flushed.
+  void PickMemtablesToFlush(std::vector<MemTable*>* mems);
 
   // Commit a successful flush in the manifest file
-  Status InstallMemtableFlushResults(MemTable* m,
+  Status InstallMemtableFlushResults(const std::vector<MemTable*> &m,
                       VersionSet* vset, Status flushStatus,
                       port::Mutex* mu, Logger* info_log,
                       uint64_t file_number,
@@ -71,7 +70,8 @@ class MemTableList {
 
   // Search all the memtables starting from the most recent one.
   // Return the most recent value found, if any.
-  bool Get(const LookupKey& key, std::string* value, Status* s);
+  bool Get(const LookupKey& key, std::string* value, Status* s,
+           const Options& options);
 
   // Returns the list of underlying memtables.
   void GetMemTables(std::vector<MemTable*>* list);

db/merge_helper.cc

@@ -0,0 +1,114 @@
+#include "merge_helper.h"
+#include "db/dbformat.h"
+#include "leveldb/comparator.h"
+#include "leveldb/db.h"
+#include "leveldb/merge_operator.h"
+#include <string>
+#include <stdio.h>
+
+namespace leveldb {
+
+// PRE:  iter points to the first merge type entry
+// POST: iter points to the first entry beyond the merge process (or the end)
+//       key_, value_ are updated to reflect the merge result
+void MergeHelper::MergeUntil(Iterator* iter, SequenceNumber stop_before,
+                             bool at_bottom) {
+  // get a copy of the internal key, before it's invalidated by iter->Next()
+  key_.assign(iter->key().data(), iter->key().size());
+  // we need to parse the internal key again as the parsed key is
+  // backed by the internal key!
+  ParsedInternalKey orig_ikey;
+  // Assume no internal key corruption as it has been successfully parsed
+  // by the caller.
+  // TODO: determine a good alternative of assert (exception?)
+  ParseInternalKey(key_, &orig_ikey);
+  std::string operand(iter->value().data(), iter->value().size());
+
+  bool hit_the_next_user_key = false;
+  ParsedInternalKey ikey;
+  for (iter->Next(); iter->Valid(); iter->Next()) {
+    if (!ParseInternalKey(iter->key(), &ikey)) {
+      // stop at corrupted key
+      if (assert_valid_internal_key_) {
+        assert(!"corrupted internal key is not expected");
+      }
+      break;
+    }
+
+    if (user_comparator_->Compare(ikey.user_key, orig_ikey.user_key) != 0) {
+      // hit a different user key, stop right here
+      hit_the_next_user_key = true;
+      break;
+    }
+
+    if (stop_before && ikey.sequence <= stop_before) {
+      // hit an entry that's visible by the previous snapshot, can't touch that
+      break;
+    }
+
+    if (kTypeDeletion == ikey.type) {
+      // hit a delete
+      //   => merge nullptr with operand
+      //   => change the entry type to kTypeValue
+      // We are done!
+      user_merge_operator_->Merge(ikey.user_key, nullptr, operand,
+                                  &value_, logger_);
+      orig_ikey.type = kTypeValue;
+      UpdateInternalKey(&key_[0], key_.size(),
+                        orig_ikey.sequence, orig_ikey.type);
+      // move iter to the next entry
+      iter->Next();
+      return;
+    }
+
+    if (kTypeValue == ikey.type) {
+      // hit a put
+      //   => merge the put value with operand
+      //   => change the entry type to kTypeValue
+      // We are done!
+      const Slice value = iter->value();
+      user_merge_operator_->Merge(ikey.user_key, &value, Slice(operand),
+                                  &value_, logger_);
+      orig_ikey.type = kTypeValue;
+      UpdateInternalKey(&key_[0], key_.size(),
+                        orig_ikey.sequence, orig_ikey.type);
+      // move iter to the next entry
+      iter->Next();
+      return;
+    }
+
+    if (kTypeMerge == ikey.type) {
+      // hit a merge
+      //   => merge the value with operand.
+      //   => put the result back to operand and continue
+      const Slice value = iter->value();
+      user_merge_operator_->Merge(ikey.user_key, &value, operand,
+                                  &value_, logger_);
+      swap(value_, operand);
+      continue;
+    }
+  }
+
+  // We have seen the root history of this key if we are at the
+  // bottem level and exhausted all internal keys of this user key
+  // NOTE: !iter->Valid() does not necessarily mean we hit the
+  // beginning of a user key, as versions of a user key might be
+  // split into multiple files and some files might not be included
+  // in the merge.
+  bool seen_the_beginning = hit_the_next_user_key && at_bottom;
+
+  if (seen_the_beginning) {
+    // do a final merge with nullptr as the existing value and say
+    // bye to the merge type (it's now converted to a Put)
+    assert(kTypeMerge == orig_ikey.type);
+    user_merge_operator_->Merge(orig_ikey.user_key, nullptr, operand,
+                                &value_, logger_);
+    orig_ikey.type = kTypeValue;
+    UpdateInternalKey(&key_[0], key_.size(),
+                      orig_ikey.sequence, orig_ikey.type);
+  } else {
+    swap(value_, operand);
+  }
+}
+
+} // namespace leveldb

db/merge_helper.h

@@ -0,0 +1,64 @@
+#ifndef MERGE_HELPER_H
+#define MERGE_HELPER_H
+
+#include "db/dbformat.h"
+#include "leveldb/slice.h"
+#include <string>
+
+namespace leveldb {
+
+class Comparator;
+class Iterator;
+class Logger;
+class MergeOperator;
+
+class MergeHelper {
+ public:
+  MergeHelper(const Comparator* user_comparator,
+              const MergeOperator* user_merge_operator,
+              Logger* logger,
+              bool assert_valid_internal_key)
+      : user_comparator_(user_comparator),
+        user_merge_operator_(user_merge_operator),
+        logger_(logger),
+        assert_valid_internal_key_(assert_valid_internal_key) {}
+
+  // Merge entries until we hit
+  //     - a corrupted key
+  //     - a Put/Delete,
+  //     - a different user key,
+  //     - a specific sequence number (snapshot boundary),
+  //  or - the end of iteration
+  // iter: (IN)  points to the first merge type entry
+  //       (OUT) points to the first entry not included in the merge process
+  // stop_before: (IN) a sequence number that merge should not cross.
+  //                   0 means no restriction
+  // 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);
+
+  // Query the merge result
+  // These are valid until the next MergeUtil call
+  // IMPORTANT: the key type could change after the MergeUntil call.
+  //            Put/Delete + Merge + ... + Merge => Put
+  //            Merge + ... + Merge => Merge
+  Slice key() { return Slice(key_); }
+  Slice value() { return Slice(value_); }
+
+ private:
+  const Comparator* user_comparator_;
+  const MergeOperator* user_merge_operator_;
+  Logger* logger_;
+  Iterator* iter_; // in: the internal iterator, positioned at the first merge entry
+  bool assert_valid_internal_key_; // enforce no internal key corruption?
+
+  // the scratch area that holds the result of MergeUntil
+  // valid up to the next MergeUntil call
+  std::string key_;
+  std::string value_;
+};
+
+} // namespace leveldb
+
+#endif

db/merge_test.cc

@@ -0,0 +1,253 @@
+#include <assert.h>
+#include <memory>
+#include <iostream>
+
+#include "leveldb/cache.h"
+#include "leveldb/comparator.h"
+#include "leveldb/db.h"
+#include "leveldb/env.h"
+#include "leveldb/merge_operator.h"
+#include "db/dbformat.h"
+#include "utilities/merge_operators.h"
+
+using namespace std;
+using namespace leveldb;
+
+auto mergeOperator = MergeOperators::CreateUInt64AddOperator();
+
+std::shared_ptr<DB> OpenDb() {
+  DB* db;
+  Options options;
+  options.create_if_missing = true;
+  options.merge_operator = mergeOperator.get();
+  Status s = DB::Open(options, "/tmp/testdb", &db);
+  if (!s.ok()) {
+    cerr << s.ToString() << endl;
+    assert(false);
+  }
+  return std::shared_ptr<DB>(db);
+}
+
+// Imagine we are maintaining a set of uint64 counters.
+// Each counter has a distinct name. And we would like
+// to support four high level operations:
+// set, add, get and remove
+// This is a quick implementation without a Merge operation.
+class Counters {
+
+ protected:
+  std::shared_ptr<DB> db_;
+
+  WriteOptions put_option_;
+  ReadOptions get_option_;
+  WriteOptions delete_option_;
+
+  uint64_t default_;
+
+ public:
+  Counters(std::shared_ptr<DB> db, uint64_t defaultCount = 0)
+      : db_(db),
+        put_option_(),
+        get_option_(),
+        delete_option_(),
+        default_(defaultCount) {
+    assert(db_);
+  }
+
+  virtual ~Counters() {}
+
+  // public interface of Counters.
+  // All four functions return false
+  // if the underlying level db operation failed.
+
+  // mapped to a levedb Put
+  bool set(const string& key, uint64_t value) {
+    // just treat the internal rep of int64 as the string
+    Slice slice((char *)&value, sizeof(value));
+    auto s = db_->Put(put_option_, key, slice);
+
+    if (s.ok()) {
+      return true;
+    } else {
+      cerr << s.ToString() << endl;
+      return false;
+    }
+  }
+
+  // mapped to a leveldb Delete
+  bool remove(const string& key) {
+    auto s = db_->Delete(delete_option_, key);
+
+    if (s.ok()) {
+      return true;
+    } else {
+      cerr << s.ToString() << std::endl;
+      return false;
+    }
+  }
+
+  // mapped to a leveldb Get
+  bool get(const string& key, uint64_t *value) {
+    string str;
+    auto s = db_->Get(get_option_, key, &str);
+
+    if (s.IsNotFound()) {
+      // return default value if not found;
+      *value = default_;
+      return true;
+    } else if (s.ok()) {
+      // deserialization
+      if (str.size() != sizeof(uint64_t)) {
+        cerr << "value corruption\n";
+        return false;
+      }
+      *value = DecodeFixed64(&str[0]);
+      return true;
+    } else {
+      cerr << s.ToString() << std::endl;
+      return false;
+    }
+  }
+
+  // 'add' is implemented as get -> modify -> set
+  // An alternative is a single merge operation, see MergeBasedCounters
+  virtual bool add(const string& key, uint64_t value) {
+    uint64_t base = default_;
+    return get(key, &base) && set(key, base + value);
+  }
+
+
+  // convenience functions for testing
+  void assert_set(const string& key, uint64_t value) {
+    assert(set(key, value));
+  }
+
+  void assert_remove(const string& key) {
+    assert(remove(key));
+  }
+
+  uint64_t assert_get(const string& key) {
+    uint64_t value = default_;
+    assert(get(key, &value));
+    return value;
+  }
+
+  void assert_add(const string& key, uint64_t value) {
+    assert(add(key, value));
+  }
+};
+
+// Implement 'add' directly with the new Merge operation
+class MergeBasedCounters : public Counters {
+ private:
+  WriteOptions merge_option_; // for merge
+
+ public:
+  MergeBasedCounters(std::shared_ptr<DB> db, uint64_t defaultCount = 0)
+      : Counters(db, defaultCount),
+        merge_option_() {
+  }
+
+  // mapped to a leveldb Merge operation
+  virtual bool add(const string& key, uint64_t value) override {
+    char encoded[sizeof(uint64_t)];
+    EncodeFixed64(encoded, value);
+    Slice slice(encoded, sizeof(uint64_t));
+    auto s = db_->Merge(merge_option_, key, slice);
+
+    if (s.ok()) {
+      return true;
+    } else {
+      cerr << s.ToString() << endl;
+      return false;
+    }
+  }
+};
+
+void dumpDb(DB* db) {
+  auto it = unique_ptr<Iterator>(db->NewIterator(ReadOptions()));
+  for (it->SeekToFirst(); it->Valid(); it->Next()) {
+    uint64_t value = DecodeFixed64(it->value().data());
+    cout << it->key().ToString() << ": "  << value << endl;
+  }
+  assert(it->status().ok());  // Check for any errors found during the scan
+}
+
+void testCounters(Counters& counters, DB* db, bool test_compaction) {
+
+  FlushOptions o;
+  o.wait = true;
+
+  counters.assert_set("a", 1);
+
+  if (test_compaction) db->Flush(o);
+
+  assert(counters.assert_get("a") == 1);
+
+  counters.assert_remove("b");
+
+  // defaut value is 0 if non-existent
+  assert(counters.assert_get("b") == 0);
+
+  counters.assert_add("a", 2);
+
+  if (test_compaction) db->Flush(o);
+
+  // 1+2 = 3
+  assert(counters.assert_get("a")== 3);
+
+  dumpDb(db);
+
+  std::cout << "1\n";
+
+  // 1+...+49 = ?
+  uint64_t sum = 0;
+  for (int i = 1; i < 50; i++) {
+    counters.assert_add("b", i);
+    sum += i;
+  }
+  assert(counters.assert_get("b") == sum);
+
+  std::cout << "2\n";
+  dumpDb(db);
+
+  std::cout << "3\n";
+
+  if (test_compaction) {
+    db->Flush(o);
+
+    cout << "Compaction started ...\n";
+    db->CompactRange(nullptr, nullptr);
+    cout << "Compaction ended\n";
+
+    dumpDb(db);
+
+    assert(counters.assert_get("a")== 3);
+    assert(counters.assert_get("b") == sum);
+  }
+}
+
+int main(int argc, char *argv[]) {
+
+  auto db = OpenDb();
+
+  {
+    cout << "Test read-modify-write counters... \n";
+    Counters counters(db, 0);
+    testCounters(counters, db.get(), true);
+  }
+
+  bool compact = false;
+  if (argc > 1) {
+    compact = true;
+    cout << "Turn on Compaction\n";
+  }
+
+  {
+    cout << "Test merge-based counters... \n";
+    MergeBasedCounters counters(db, 0);
+    testCounters(counters, db.get(), compact);
+  }
+
+  return 0;
+}

db/repair.cc

@@ -52,7 +52,7 @@ class Repairer {
         options_(SanitizeOptions(dbname, &icmp_, &ipolicy_, options)),
         next_file_number_(1) {
     // TableCache can be small since we expect each table to be opened once.
-    table_cache_ = new TableCache(dbname_, &options_, 10);
+    table_cache_ = new TableCache(dbname_, &options_, storage_options_, 10);
     edit_ = new VersionEdit(options.num_levels);
   }
 
@@ -105,6 +105,7 @@ class Repairer {
   std::vector<uint64_t> logs_;
   std::vector<TableInfo> tables_;
   uint64_t next_file_number_;
+  const EnvOptions storage_options_;
 
   Status FindFiles() {
     std::vector<std::string> filenames;
@@ -169,7 +170,7 @@ class Repairer {
     // Open the log file
     std::string logname = LogFileName(dbname_, log);
     unique_ptr<SequentialFile> lfile;
-    Status status = env_->NewSequentialFile(logname, &lfile);
+    Status status = env_->NewSequentialFile(logname, &lfile, storage_options_);
     if (!status.ok()) {
       return status;
     }
@@ -216,7 +217,8 @@ class Repairer {
     FileMetaData meta;
     meta.number = next_file_number_++;
     Iterator* iter = mem->NewIterator();
-    status = BuildTable(dbname_, env_, options_, table_cache_, iter, &meta,
+    status = BuildTable(dbname_, env_, options_, storage_options_,
+                        table_cache_, iter, &meta,
                         icmp_.user_comparator(), 0, 0);
     delete iter;
     mem->Unref();
@@ -258,7 +260,7 @@ class Repairer {
     Status status = env_->GetFileSize(fname, &t->meta.file_size);
     if (status.ok()) {
       Iterator* iter = table_cache_->NewIterator(
-          ReadOptions(), t->meta.number, t->meta.file_size);
+          ReadOptions(), storage_options_, t->meta.number, t->meta.file_size);
       bool empty = true;
       ParsedInternalKey parsed;
       t->max_sequence = 0;
@@ -296,7 +298,7 @@ class Repairer {
   Status WriteDescriptor() {
     std::string tmp = TempFileName(dbname_, 1);
     unique_ptr<WritableFile> file;
-    Status status = env_->NewWritableFile(tmp, &file);
+    Status status = env_->NewWritableFile(tmp, &file, storage_options_);
     if (!status.ok()) {
       return status;
     }

db/snapshot.h

@@ -59,12 +59,9 @@ class SnapshotList {
 
   // retrieve all snapshot numbers. They are sorted in ascending order.
   void getAll(std::vector<SequenceNumber>& ret) {
-    SnapshotImpl* s = &list_;
     if (empty()) return;
-    SequenceNumber prev __attribute__((unused)) = 0;
+    SnapshotImpl* s = &list_;
     while (s->next_ != &list_) {
-      assert(prev <= s->next_->number_);
-      assert(prev = s->next_->number_); // assignment
       ret.push_back(s->next_->number_);
       s = s ->next_;
     }

db/table_cache.cc

@@ -6,23 +6,16 @@
 
 #include "db/filename.h"
 
-#include "leveldb/env.h"
-#include "leveldb/table.h"
 #include "leveldb/statistics.h"
+#include "table/table.h"
 #include "util/coding.h"
+#include "util/stop_watch.h"
 
 namespace leveldb {
 
-struct TableAndFile {
-  unique_ptr<Table> table;
-};
-
-static class Statistics* dbstatistics;
-
 static void DeleteEntry(const Slice& key, void* value) {
-  TableAndFile* tf = reinterpret_cast<TableAndFile*>(value);
-  RecordTick(dbstatistics, NO_FILE_CLOSES);
-  delete tf;
+  Table* table = reinterpret_cast<Table*>(value);
+  delete table;
 }
 
 static void UnrefEntry(void* arg1, void* arg2) {
@@ -33,18 +26,19 @@ static void UnrefEntry(void* arg1, void* arg2) {
 
 TableCache::TableCache(const std::string& dbname,
                        const Options* options,
+                       const EnvOptions& storage_options,
                        int entries)
     : env_(options->env),
       dbname_(dbname),
       options_(options),
-      cache_(NewLRUCache(entries, options->table_cache_numshardbits)) {
-  dbstatistics = options->statistics;
-}
+      storage_options_(storage_options),
+      cache_(NewLRUCache(entries, options->table_cache_numshardbits)) {}
 
 TableCache::~TableCache() {
 }
 
-Status TableCache::FindTable(uint64_t file_number, uint64_t file_size,
+Status TableCache::FindTable(const EnvOptions& toptions,
+                             uint64_t file_number, uint64_t file_size,
                              Cache::Handle** handle, bool* tableIO) {
   Status s;
   char buf[sizeof(file_number)];
@@ -58,10 +52,14 @@ Status TableCache::FindTable(uint64_t file_number, uint64_t file_size,
     std::string fname = TableFileName(dbname_, file_number);
     unique_ptr<RandomAccessFile> file;
     unique_ptr<Table> table;
-    s = env_->NewRandomAccessFile(fname, &file);
+    s = env_->NewRandomAccessFile(fname, &file, toptions);
     RecordTick(options_->statistics, NO_FILE_OPENS);
     if (s.ok()) {
-      s = Table::Open(*options_, std::move(file), file_size, &table);
+      if (options_->advise_random_on_open) {
+        file->Hint(RandomAccessFile::RANDOM);
+      }
+      StopWatch sw(env_, options_->statistics, TABLE_OPEN_IO_MICROS);
+      s = Table::Open(*options_, toptions, std::move(file), file_size, &table);
     }
 
     if (!s.ok()) {
@@ -70,36 +68,41 @@ Status TableCache::FindTable(uint64_t file_number, uint64_t file_size,
       // We do not cache error results so that if the error is transient,
       // or somebody repairs the file, we recover automatically.
     } else {
-      TableAndFile* tf = new TableAndFile;
-      tf->table = std::move(table);
       assert(file.get() == nullptr);
-      *handle = cache_->Insert(key, tf, 1, &DeleteEntry);
+      *handle = cache_->Insert(key, table.release(), 1, &DeleteEntry);
     }
   }
   return s;
 }
 
 Iterator* TableCache::NewIterator(const ReadOptions& options,
+                                  const EnvOptions& toptions,
                                   uint64_t file_number,
                                   uint64_t file_size,
-                                  Table** tableptr) {
+                                  Table** tableptr,
+                                  bool for_compaction) {
   if (tableptr != nullptr) {
     *tableptr = nullptr;
   }
 
   Cache::Handle* handle = nullptr;
-  Status s = FindTable(file_number, file_size, &handle);
+  Status s = FindTable(toptions, file_number, file_size, &handle);
   if (!s.ok()) {
     return NewErrorIterator(s);
   }
 
   Table* table =
-    reinterpret_cast<TableAndFile*>(cache_->Value(handle))->table.get();
+    reinterpret_cast<Table*>(cache_->Value(handle));
   Iterator* result = table->NewIterator(options);
   result->RegisterCleanup(&UnrefEntry, cache_.get(), handle);
   if (tableptr != nullptr) {
     *tableptr = table;
   }
+
+  if (for_compaction) {
+    table->SetupForCompaction();
+  }
+
   return result;
 }
 
@@ -108,13 +111,14 @@ Status TableCache::Get(const ReadOptions& options,
                        uint64_t file_size,
                        const Slice& k,
                        void* arg,
-                       void (*saver)(void*, const Slice&, const Slice&, bool),
+                       bool (*saver)(void*, const Slice&, const Slice&, bool),
                        bool* tableIO) {
   Cache::Handle* handle = nullptr;
-  Status s = FindTable(file_number, file_size, &handle, tableIO);
+  Status s = FindTable(storage_options_, file_number, file_size,
+                       &handle, tableIO);
   if (s.ok()) {
     Table* t =
-      reinterpret_cast<TableAndFile*>(cache_->Value(handle))->table.get();
+      reinterpret_cast<Table*>(cache_->Value(handle));
     s = t->InternalGet(options, k, arg, saver);
     cache_->Release(handle);
   }

db/table_cache.h

@@ -10,9 +10,10 @@
 #include <string>
 #include <stdint.h>
 #include "db/dbformat.h"
+#include "leveldb/env.h"
 #include "leveldb/cache.h"
-#include "leveldb/table.h"
 #include "port/port.h"
+#include "table/table.h"
 
 namespace leveldb {
 
@@ -20,7 +21,8 @@ class Env;
 
 class TableCache {
  public:
-  TableCache(const std::string& dbname, const Options* options, int entries);
+  TableCache(const std::string& dbname, const Options* options,
+             const EnvOptions& storage_options, int entries);
   ~TableCache();
 
   // Return an iterator for the specified file number (the corresponding
@@ -31,18 +33,21 @@ class TableCache {
   // 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,
-                        Table** tableptr = nullptr);
+                        Table** tableptr = nullptr,
+                        bool for_compaction = false);
 
   // If a seek to internal key "k" in specified file finds an entry,
-  // call (*handle_result)(arg, found_key, found_value).
+  // 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,
-             void (*handle_result)(void*, const Slice&, const Slice&, bool),
+             bool (*handle_result)(void*, const Slice&, const Slice&, bool),
              bool* tableIO);
 
   // Evict any entry for the specified file number
@@ -52,9 +57,11 @@ class TableCache {
   Env* const env_;
   const std::string dbname_;
   const Options* options_;
+  const EnvOptions& storage_options_;
   std::shared_ptr<Cache> cache_;
 
-  Status FindTable(uint64_t file_number, uint64_t file_size, Cache::Handle**,
+  Status FindTable(const EnvOptions& toptions,
+                   uint64_t file_number, uint64_t file_size, Cache::Handle**,
                    bool* tableIO = nullptr);
 };
 

db/transaction_log_iterator_impl.cc

@@ -1,33 +1,30 @@
 #include "db/transaction_log_iterator_impl.h"
 #include "db/write_batch_internal.h"
 #include "db/filename.h"
+
 namespace leveldb {
 
 TransactionLogIteratorImpl::TransactionLogIteratorImpl(
                            const std::string& dbname,
                            const Options* options,
+                           const EnvOptions& soptions,
                            SequenceNumber& seq,
-                           std::vector<LogFile>* files,
+                           std::unique_ptr<std::vector<LogFile>> files,
                            SequenceNumber const * const lastFlushedSequence) :
   dbname_(dbname),
   options_(options),
-  sequenceNumber_(seq),
-  files_(files),
+  soptions_(soptions),
+  startingSequenceNumber_(seq),
+  files_(std::move(files)),
   started_(false),
-  isValid_(true),
+  isValid_(false),
   currentFileIndex_(0),
   lastFlushedSequence_(lastFlushedSequence) {
-  assert(files_ != nullptr);
+  assert(files_.get() != nullptr);
   assert(lastFlushedSequence_);
-}
 
-LogReporter
-TransactionLogIteratorImpl::NewLogReporter(const uint64_t logNumber) {
-  LogReporter reporter;
-  reporter.env = options_->env;
-  reporter.info_log = options_->info_log.get();
-  reporter.log_number = logNumber;
-  return reporter;
+  reporter_.env = options_->env;
+  reporter_.info_log = options_->info_log.get();
 }
 
 Status TransactionLogIteratorImpl::OpenLogFile(
@@ -36,15 +33,15 @@ Status TransactionLogIteratorImpl::OpenLogFile(
   Env* env = options_->env;
   if (logFile.type == kArchivedLogFile) {
     std::string fname = ArchivedLogFileName(dbname_, logFile.logNumber);
-    return env->NewSequentialFile(fname, file);
+    return env->NewSequentialFile(fname, file, soptions_);
   } else {
     std::string fname = LogFileName(dbname_, logFile.logNumber);
-    Status status = env->NewSequentialFile(fname, file);
+    Status status = env->NewSequentialFile(fname, file, soptions_);
     if (!status.ok()) {
       //  If cannot open file in DB directory.
       //  Try the archive dir, as it could have moved in the meanwhile.
       fname = ArchivedLogFileName(dbname_, logFile.logNumber);
-      status = env->NewSequentialFile(fname, file);
+      status = env->NewSequentialFile(fname, file, soptions_);
       if (!status.ok()) {
         return Status::IOError(" Requested file not present in the dir");
       }
@@ -70,95 +67,84 @@ bool TransactionLogIteratorImpl::Valid() {
 }
 
 void TransactionLogIteratorImpl::Next() {
+  LogFile currentLogFile = files_.get()->at(currentFileIndex_);
+
 //  First seek to the given seqNo. in the current file.
-  LogFile currentLogFile = files_->at(currentFileIndex_);
-  LogReporter reporter = NewLogReporter(currentLogFile.logNumber);
   std::string scratch;
   Slice record;
-
   if (!started_) {
+    started_ = true;  // this piece only runs onced.
     isValid_ = false;
-    if (sequenceNumber_ > *lastFlushedSequence_) {
+    if (startingSequenceNumber_ > *lastFlushedSequence_) {
       currentStatus_ = Status::IOError("Looking for a sequence, "
                                         "which is not flushed yet.");
       return;
     }
-    unique_ptr<SequentialFile> file;
-    Status status = OpenLogFile(currentLogFile, &file);
-    if (!status.ok()) {
-      currentStatus_ = status;
+    Status s = OpenLogReader(currentLogFile);
+    if (!s.ok()) {
+      currentStatus_ = s;
+      isValid_ = false;
       return;
     }
-    assert(file);
-    unique_ptr<log::Reader> reader(
-      new log::Reader(std::move(file), &reporter, true, 0));
-    assert(reader);
-    while (reader->ReadRecord(&record, &scratch)) {
+    while (currentLogReader_->ReadRecord(&record, &scratch)) {
       if (record.size() < 12) {
-        reporter.Corruption(
+        reporter_.Corruption(
           record.size(), Status::Corruption("log record too small"));
         continue;
       }
       UpdateCurrentWriteBatch(record);
-      if (currentSequence_ >= sequenceNumber_) {
-        assert(currentSequence_ < *lastFlushedSequence_);
+      if (currentSequence_ >= startingSequenceNumber_) {
+        assert(currentSequence_ <= *lastFlushedSequence_);
         isValid_ = true;
-        currentLogReader_ = std::move(reader);
         break;
+      } else {
+        isValid_ = false;
       }
     }
-    if (!isValid_) {
-      //  TODO read the entire first file. and did not find the seq number.
-      //  Error out.
-      currentStatus_ =
-        Status::NotFound("Did not find the Seq no. in first file");
+    if (isValid_) {
+      // Done for this iteration
+      return;
     }
-    started_ = true;
-  } else {
-LOOK_NEXT_FILE:
+  }
+  bool openNextFile = true;
+  while(openNextFile) {
     assert(currentLogReader_);
-    bool openNextFile = true;
-
-    if (currentSequence_ == *lastFlushedSequence_) {
-      // The last update has been read. and next is being called.
-      isValid_ = false;
-      currentStatus_ = Status::OK();
-    }
-
-    while (currentLogReader_->ReadRecord(&record, &scratch) &&
-           currentSequence_ < *lastFlushedSequence_) {
-      if (record.size() < 12) {
-        reporter.Corruption(
-          record.size(), Status::Corruption("log record too small"));
-        continue;
-      } else {
-        UpdateCurrentWriteBatch(record);
-        openNextFile = false;
-        break;
+    if (currentSequence_ < *lastFlushedSequence_) {
+      if (currentLogReader_->IsEOF()) {
+        currentLogReader_->UnmarkEOF();
+      }
+      while (currentLogReader_->ReadRecord(&record, &scratch)) {
+        if (record.size() < 12) {
+          reporter_.Corruption(
+            record.size(), Status::Corruption("log record too small"));
+          continue;
+        } else {
+          UpdateCurrentWriteBatch(record);
+          openNextFile = false;
+          break;
+        }
       }
     }
 
     if (openNextFile) {
-      if (currentFileIndex_ < files_->size() - 1) {
+      if (currentFileIndex_ < files_.get()->size() - 1) {
         ++currentFileIndex_;
-        currentLogReader_.reset();
-        unique_ptr<SequentialFile> file;
-        Status status = OpenLogFile(files_->at(currentFileIndex_), &file);
+        Status status = OpenLogReader(files_.get()->at(currentFileIndex_));
         if (!status.ok()) {
           isValid_ = false;
           currentStatus_ = status;
           return;
         }
-        currentLogReader_.reset(
-          new log::Reader(std::move(file), &reporter, true, 0));
-        goto LOOK_NEXT_FILE;
       } else {
-        //  LOOKED AT FILES. WE ARE DONE HERE.
         isValid_ = false;
-        currentStatus_ = Status::IOError(" NO MORE DATA LEFT");
+        openNextFile = false;
+        if (currentSequence_ == *lastFlushedSequence_) {
+          currentStatus_ = Status::OK();
+        } else {
+          currentStatus_ = Status::IOError(" NO MORE DATA LEFT");
+        }
       }
     }
-
   }
 }
 
@@ -167,6 +153,20 @@ void TransactionLogIteratorImpl::UpdateCurrentWriteBatch(const Slice& record) {
   WriteBatchInternal::SetContents(batch, record);
   currentSequence_ = WriteBatchInternal::Sequence(batch);
   currentBatch_.reset(batch);
+  isValid_ = true;
+  currentStatus_ = Status::OK();
 }
 
+Status TransactionLogIteratorImpl::OpenLogReader(const LogFile& logFile) {
+  unique_ptr<SequentialFile> file;
+  Status status = OpenLogFile(logFile, &file);
+  if (!status.ok()) {
+    return status;
+  }
+  assert(file);
+  currentLogReader_.reset(
+    new log::Reader(std::move(file), &reporter_, true, 0)
+  );
+  return Status::OK();
+}
 }  //  namespace leveldb

db/transaction_log_iterator_impl.h

@@ -16,10 +16,8 @@ namespace leveldb {
 struct LogReporter : public log::Reader::Reporter {
   Env* env;
   Logger* info_log;
-  uint64_t log_number;
   virtual void Corruption(size_t bytes, const Status& s) {
-    Log(info_log, "%ld: dropping %d bytes; %s",
-        log_number, static_cast<int>(bytes), s.ToString().c_str());
+    Log(info_log, "dropping %zu bytes; %s", bytes, s.ToString().c_str());
   }
 };
 
@@ -27,15 +25,11 @@ class TransactionLogIteratorImpl : public TransactionLogIterator {
  public:
   TransactionLogIteratorImpl(const std::string& dbname,
                              const Options* options,
+                             const EnvOptions& soptions,
                              SequenceNumber& seqNum,
-                             std::vector<LogFile>* files,
+                             std::unique_ptr<std::vector<LogFile>> files,
                              SequenceNumber const * const lastFlushedSequence);
 
-  virtual ~TransactionLogIteratorImpl() {
-    //  TODO move to cc file.
-    delete files_;
-  }
-
   virtual bool Valid();
 
   virtual void Next();
@@ -47,8 +41,9 @@ class TransactionLogIteratorImpl : public TransactionLogIterator {
  private:
   const std::string& dbname_;
   const Options* options_;
-  const uint64_t sequenceNumber_;
-  const std::vector<LogFile>* files_;
+  const EnvOptions& soptions_;
+  const uint64_t startingSequenceNumber_;
+  std::unique_ptr<std::vector<LogFile>> files_;
   bool started_;
   bool isValid_;  // not valid when it starts of.
   Status currentStatus_;
@@ -56,12 +51,13 @@ class TransactionLogIteratorImpl : public TransactionLogIterator {
   std::unique_ptr<WriteBatch> currentBatch_;
   unique_ptr<log::Reader> currentLogReader_;
   Status OpenLogFile(const LogFile& logFile, unique_ptr<SequentialFile>* file);
-  LogReporter NewLogReporter(uint64_t logNumber);
+  LogReporter reporter_;
   SequenceNumber const * const lastFlushedSequence_;
   // represents the sequence number being read currently.
   SequenceNumber currentSequence_;
 
   void UpdateCurrentWriteBatch(const Slice& record);
+  Status OpenLogReader(const LogFile& file);
 };
 
 

db/version_edit.h

@@ -81,6 +81,11 @@ class VersionEdit {
     deleted_files_.insert(std::make_pair(level, file));
   }
 
+  // Number of edits
+  int NumEntries() {
+    return new_files_.size() + deleted_files_.size();
+  }
+
   void EncodeTo(std::string* dst) const;
   Status DecodeFrom(const Slice& src);
 

db/version_set.cc

@@ -12,11 +12,13 @@
 #include "db/memtable.h"
 #include "db/table_cache.h"
 #include "leveldb/env.h"
+#include "leveldb/merge_operator.h"
 #include "leveldb/table_builder.h"
 #include "table/merger.h"
 #include "table/two_level_iterator.h"
 #include "util/coding.h"
 #include "util/logging.h"
+#include "util/stop_watch.h"
 
 namespace leveldb {
 
@@ -178,32 +180,39 @@ class Version::LevelFileNumIterator : public Iterator {
 
 static Iterator* GetFileIterator(void* arg,
                                  const ReadOptions& options,
-                                 const Slice& file_value) {
+                                 const EnvOptions& soptions,
+                                 const Slice& file_value,
+                                 bool for_compaction) {
   TableCache* cache = reinterpret_cast<TableCache*>(arg);
   if (file_value.size() != 16) {
     return NewErrorIterator(
         Status::Corruption("FileReader invoked with unexpected value"));
   } else {
     return cache->NewIterator(options,
+                              soptions,
                               DecodeFixed64(file_value.data()),
-                              DecodeFixed64(file_value.data() + 8));
+                              DecodeFixed64(file_value.data() + 8),
+                              nullptr /* don't need reference to table*/,
+                              for_compaction);
   }
 }
 
 Iterator* Version::NewConcatenatingIterator(const ReadOptions& options,
+                                            const EnvOptions& soptions,
                                             int level) const {
   return NewTwoLevelIterator(
       new LevelFileNumIterator(vset_->icmp_, &files_[level]),
-      &GetFileIterator, vset_->table_cache_, options);
+      &GetFileIterator, vset_->table_cache_, options, soptions);
 }
 
 void Version::AddIterators(const ReadOptions& options,
+                           const EnvOptions& soptions,
                            std::vector<Iterator*>* iters) {
   // Merge all level zero files together since they may overlap
   for (size_t i = 0; i < files_[0].size(); i++) {
     iters->push_back(
         vset_->table_cache_->NewIterator(
-            options, files_[0][i]->number, files_[0][i]->file_size));
+            options, soptions, files_[0][i]->number, files_[0][i]->file_size));
   }
 
   // For levels > 0, we can use a concatenating iterator that sequentially
@@ -211,7 +220,7 @@ void Version::AddIterators(const ReadOptions& options,
   // lazily.
   for (int level = 1; level < vset_->NumberLevels(); level++) {
     if (!files_[level].empty()) {
-      iters->push_back(NewConcatenatingIterator(options, level));
+      iters->push_back(NewConcatenatingIterator(options, soptions, level));
     }
   }
 }
@@ -223,29 +232,78 @@ enum SaverState {
   kFound,
   kDeleted,
   kCorrupt,
+  kMerge // value contains the current merge result (the operand)
 };
 struct Saver {
   SaverState state;
   const Comparator* ucmp;
   Slice user_key;
   std::string* value;
+  const MergeOperator* merge_operator;
+  Logger* logger;
   bool didIO;    // did we do any disk io?
 };
 }
-static void SaveValue(void* arg, const Slice& ikey, const Slice& v, bool didIO){
+static bool SaveValue(void* arg, const Slice& ikey, const Slice& v, bool didIO){
   Saver* s = reinterpret_cast<Saver*>(arg);
   ParsedInternalKey parsed_key;
+  // TODO: didIO and Merge?
   s->didIO = didIO;
   if (!ParseInternalKey(ikey, &parsed_key)) {
+    // TODO: what about corrupt during Merge?
     s->state = kCorrupt;
   } else {
     if (s->ucmp->Compare(parsed_key.user_key, s->user_key) == 0) {
-      s->state = (parsed_key.type == kTypeValue) ? kFound : kDeleted;
-      if (s->state == kFound) {
-        s->value->assign(v.data(), v.size());
+      switch (parsed_key.type) {
+        case kTypeValue:
+          if (kNotFound == s->state) {
+            s->value->assign(v.data(), v.size());
+          } else if (kMerge == s->state) {
+            std::string operand;
+            swap(operand, *s->value);
+            s->merge_operator->Merge(s->user_key, &v, operand,
+                                     s->value, s->logger);
+          } else {
+            assert(false);
+          }
+          s->state = kFound;
+          return false;
+
+        case kTypeMerge:
+          if (kNotFound == s->state) {
+            s->state = kMerge;
+            s->value->assign(v.data(), v.size());
+          } else if (kMerge == s->state) {
+            std::string operand;
+            swap(operand, *s->value);
+            s->merge_operator->Merge(s->user_key, &v, operand,
+                                     s->value, s->logger);
+          } else {
+            assert(false);
+          }
+          return true;
+
+        case kTypeDeletion:
+          if (kNotFound == s->state) {
+            s->state = kDeleted;
+          } else if (kMerge == s->state) {
+            std::string operand;
+            swap(operand, *s->value);
+            s->merge_operator->Merge(s->user_key, nullptr, operand,
+                                     s->value, s->logger);
+            s->state = kFound;
+          } else {
+            assert(false);
+          }
+          return false;
+
       }
     }
   }
+
+  // s->state could be Corrupt, merge or notfound
+
+  return false;
 }
 
 static bool NewestFirst(FileMetaData* a, FileMetaData* b) {
@@ -265,14 +323,28 @@ Version::Version(VersionSet* vset, uint64_t version_number)
   files_ = new std::vector<FileMetaData*>[vset->NumberLevels()];
 }
 
-Status Version::Get(const ReadOptions& options,
-                    const LookupKey& k,
-                    std::string* value,
-                    GetStats* stats) {
+void Version::Get(const ReadOptions& options,
+                  const LookupKey& k,
+                  std::string* value,
+                  Status *status,
+                  GetStats* stats,
+                  const Options& db_options) {
   Slice ikey = k.internal_key();
   Slice user_key = k.user_key();
   const Comparator* ucmp = vset_->icmp_.user_comparator();
-  Status s;
+
+  auto merge_operator = db_options.merge_operator;
+  auto logger = db_options.info_log;
+
+  assert(status->ok() || status->IsMergeInProgress());
+  Saver saver;
+  saver.state = status->ok()? kNotFound : kMerge;
+  saver.ucmp = ucmp;
+  saver.user_key = user_key;
+  saver.value = value;
+  saver.merge_operator = merge_operator;
+  saver.logger = logger.get();
+  saver.didIO = false;
 
   stats->seek_file = nullptr;
   stats->seek_file_level = -1;
@@ -321,24 +393,21 @@ Status Version::Get(const ReadOptions& options,
         } else {
           files = &tmp2;
           num_files = 1;
+          // TODO, is level 1-n files all disjoint in user key space?
         }
       }
     }
 
-    for (uint32_t i = 0; i < num_files; ++i) {
 
+
+    for (uint32_t i = 0; i < num_files; ++i) {
       FileMetaData* f = files[i];
-      Saver saver;
-      saver.state = kNotFound;
-      saver.ucmp = ucmp;
-      saver.user_key = user_key;
-      saver.value = value;
-      saver.didIO = false;
       bool tableIO = false;
-      s = vset_->table_cache_->Get(options, f->number, f->file_size,
-                                   ikey, &saver, SaveValue, &tableIO);
-      if (!s.ok()) {
-        return s;
+      *status = vset_->table_cache_->Get(options, f->number, f->file_size,
+                                         ikey, &saver, SaveValue, &tableIO);
+      // TODO: examine the behavior for corrupted key
+      if (!status->ok()) {
+        return;
       }
 
       if (last_file_read != nullptr && stats->seek_file == nullptr) {
@@ -363,18 +432,33 @@ Status Version::Get(const ReadOptions& options,
         case kNotFound:
           break;      // Keep searching in other files
         case kFound:
-          return s;
+          return;
         case kDeleted:
-          s = Status::NotFound(Slice());  // Use empty error message for speed
-          return s;
+          *status = Status::NotFound(Slice());  // Use empty error message for speed
+          return;
         case kCorrupt:
-          s = Status::Corruption("corrupted key for ", user_key);
-          return s;
+          *status = Status::Corruption("corrupted key for ", user_key);
+          return;
+        case kMerge:
+          break;
       }
     }
   }
 
-  return Status::NotFound(Slice());  // Use an empty error message for speed
+
+  if (kMerge == saver.state) {
+    // merge operand is in *value and we hit the beginning of the key history
+    // do a final merge of nullptr and operand;
+    std::string operand;
+    swap(operand, *value);
+    merge_operator->Merge(user_key, nullptr, operand,
+                          value, logger.get());
+    *status = Status::OK();
+    return;
+  } else {
+    *status = Status::NotFound(Slice()); // Use an empty error message for speed
+    return;
+  }
 }
 
 bool Version::UpdateStats(const GetStats& stats) {
@@ -559,7 +643,7 @@ void Version::ExtendOverlappingInputs(
     const Slice& user_begin,
     const Slice& user_end,
     std::vector<FileMetaData*>* inputs,
-    int midIndex) {
+    unsigned int midIndex) {
 
   const Comparator* user_cmp = vset_->icmp_.user_comparator();
 #ifndef NDEBUG
@@ -736,13 +820,16 @@ class VersionSet::Builder {
 #endif
   }
 
-  void CheckConsistencyForDeletes(VersionEdit* edit, int number, int level) {
+  void CheckConsistencyForDeletes(
+    VersionEdit* edit,
+    unsigned int number,
+    int level) {
 #ifndef NDEBUG
       // a file to be deleted better exist in the previous version
       bool found = false;
       for (int l = 0; !found && l < edit->number_levels_; l++) {
         const std::vector<FileMetaData*>& base_files = base_->files_[l];
-        for (int i = 0; i < base_files.size(); i++) {
+        for (unsigned int i = 0; i < base_files.size(); i++) {
           FileMetaData* f = base_files[i];
           if (f->number == number) {
             found =  true;
@@ -884,6 +971,7 @@ class VersionSet::Builder {
 
 VersionSet::VersionSet(const std::string& dbname,
                        const Options* options,
+                       const EnvOptions& storage_options,
                        TableCache* table_cache,
                        const InternalKeyComparator* cmp)
     : env_(options->env),
@@ -901,7 +989,9 @@ VersionSet::VersionSet(const std::string& dbname,
       current_(nullptr),
       compactions_in_progress_(options_->num_levels),
       current_version_number_(0),
-      last_observed_manifest_size_(0) {
+      last_observed_manifest_size_(0),
+      storage_options_(storage_options),
+      storage_options_compactions_(storage_options_)  {
   compact_pointer_ = new std::string[options_->num_levels];
   Init(options_->num_levels);
   AppendVersion(new Version(this, current_version_number_++));
@@ -923,7 +1013,8 @@ void VersionSet::Init(int num_levels) {
   for (int i = 0; i < num_levels; i++) {
     if (i > 1) {
       max_file_size_[i] = max_file_size_[i-1] * target_file_size_multiplier;
-      level_max_bytes_[i] = level_max_bytes_[i-1] * max_bytes_multiplier;
+      level_max_bytes_[i] = level_max_bytes_[i-1] * max_bytes_multiplier *
+        options_->max_bytes_for_level_multiplier_additional[i-1];
     } else {
       max_file_size_[i] = options_->target_file_size_base;
       level_max_bytes_[i] = options_->max_bytes_for_level_base;
@@ -999,7 +1090,8 @@ Status VersionSet::LogAndApply(VersionEdit* edit, port::Mutex* mu,
     new_manifest_file = DescriptorFileName(dbname_, manifest_file_number_);
     edit->SetNextFile(next_file_number_);
     unique_ptr<WritableFile> descriptor_file;
-    s = env_->NewWritableFile(new_manifest_file, &descriptor_file);
+    s = env_->NewWritableFile(new_manifest_file, &descriptor_file,
+                              storage_options_);
     if (s.ok()) {
       descriptor_log_.reset(new log::Writer(std::move(descriptor_file)));
       s = WriteSnapshot(descriptor_log_.get());
@@ -1032,8 +1124,10 @@ Status VersionSet::LogAndApply(VersionEdit* edit, port::Mutex* mu,
       }
       if (s.ok()) {
         if (options_->use_fsync) {
+          StopWatch sw(env_, options_->statistics, MANIFEST_FILE_SYNC_MICROS);
           s = descriptor_log_->file()->Fsync();
         } else {
+          StopWatch sw(env_, options_->statistics, MANIFEST_FILE_SYNC_MICROS);
           s = descriptor_log_->file()->Sync();
         }
       }
@@ -1144,7 +1238,7 @@ Status VersionSet::Recover() {
 
   std::string dscname = dbname_ + "/" + current;
   unique_ptr<SequentialFile> file;
-  s = env_->NewSequentialFile(dscname, &file);
+  s = env_->NewSequentialFile(dscname, &file, storage_options_);
   if (!s.ok()) {
     return s;
   }
@@ -1266,7 +1360,7 @@ Status VersionSet::DumpManifest(Options& options, std::string& dscname,
 
   // Open the specified manifest file.
   unique_ptr<SequentialFile> file;
-  Status s = options.env->NewSequentialFile(dscname, &file);
+  Status s = options.env->NewSequentialFile(dscname, &file, storage_options_);
   if (!s.ok()) {
     return s;
   }
@@ -1576,7 +1670,7 @@ bool VersionSet::ManifestContains(const std::string& record) const {
   std::string fname = DescriptorFileName(dbname_, manifest_file_number_);
   Log(options_->info_log, "ManifestContains: checking %s\n", fname.c_str());
   unique_ptr<SequentialFile> file;
-  Status s = env_->NewSequentialFile(fname, &file);
+  Status s = env_->NewSequentialFile(fname, &file, storage_options_);
   if (!s.ok()) {
     Log(options_->info_log, "ManifestContains: %s\n", s.ToString().c_str());
     Log(options_->info_log,
@@ -1620,7 +1714,8 @@ uint64_t VersionSet::ApproximateOffsetOf(Version* v, const InternalKey& ikey) {
         // approximate offset of "ikey" within the table.
         Table* tableptr;
         Iterator* iter = table_cache_->NewIterator(
-            ReadOptions(), files[i]->number, files[i]->file_size, &tableptr);
+            ReadOptions(), storage_options_, files[i]->number,
+            files[i]->file_size, &tableptr);
         if (tableptr != nullptr) {
           result += tableptr->ApproximateOffsetOf(ikey.Encode());
         }
@@ -1631,14 +1726,26 @@ uint64_t VersionSet::ApproximateOffsetOf(Version* v, const InternalKey& ikey) {
   return result;
 }
 
-void VersionSet::AddLiveFiles(std::set<uint64_t>* live) {
+void VersionSet::AddLiveFiles(std::vector<uint64_t>* live_list) {
+  // pre-calculate space requirement
+  int64_t total_files = 0;
   for (Version* v = dummy_versions_.next_;
        v != &dummy_versions_;
        v = v->next_) {
     for (int level = 0; level < NumberLevels(); level++) {
-      const std::vector<FileMetaData*>& files = v->files_[level];
-      for (size_t i = 0; i < files.size(); i++) {
-        live->insert(files[i]->number);
+      total_files += v->files_[level].size();
+    }
+  }
+
+  // just one time extension to the right size
+  live_list->reserve(live_list->size() + total_files);
+
+  for (Version* v = dummy_versions_.next_;
+       v != &dummy_versions_;
+       v = v->next_) {
+    for (int level = 0; level < NumberLevels(); level++) {
+      for (const auto& f : v->files_[level]) {
+        live_list->push_back(f->number);
       }
     }
   }
@@ -1657,10 +1764,8 @@ void VersionSet::AddLiveFilesCurrentVersion(std::set<uint64_t>* live) {
 int64_t VersionSet::NumLevelBytes(int level) const {
   assert(level >= 0);
   assert(level < NumberLevels());
-  if(current_ && level < (int)current_->files_->size())
-    return TotalFileSize(current_->files_[level]);
-  else
-    return 0;
+  assert(current_);
+  return TotalFileSize(current_->files_[level]);
 }
 
 int64_t VersionSet::MaxNextLevelOverlappingBytes() {
@@ -1734,13 +1839,16 @@ Iterator* VersionSet::MakeInputIterator(Compaction* c) {
         const std::vector<FileMetaData*>& files = c->inputs_[which];
         for (size_t i = 0; i < files.size(); i++) {
           list[num++] = table_cache_->NewIterator(
-              options, files[i]->number, files[i]->file_size);
+              options, storage_options_compactions_,
+              files[i]->number, files[i]->file_size, nullptr,
+              true /* for compaction */);
         }
       } else {
         // Create concatenating iterator for the files from this level
         list[num++] = NewTwoLevelIterator(
             new Version::LevelFileNumIterator(icmp_, &c->inputs_[which]),
-            &GetFileIterator, table_cache_, options);
+            &GetFileIterator, table_cache_, options, storage_options_,
+            true /* for compaction */);
       }
     }
   }
@@ -1953,14 +2061,23 @@ Compaction* VersionSet::PickCompaction() {
   }
 
   // Find compactions needed by seeks
-  if (c == nullptr && (current_->file_to_compact_ != nullptr)) {
+  FileMetaData* f = current_->file_to_compact_;
+  if (c == nullptr && f != nullptr && !f->being_compacted) {
+
     level = current_->file_to_compact_level_;
+    int parent_index = -1;
 
     // Only allow one level 0 compaction at a time.
+    // Do not pick this file if its parents at level+1 are being compacted.
     if (level != 0 || compactions_in_progress_[0].empty()) {
-      c = new Compaction(level, MaxFileSizeForLevel(level),
-      MaxGrandParentOverlapBytes(level), NumberLevels(), true);
-      c->inputs_[0].push_back(current_->file_to_compact_);
+      if(!ParentRangeInCompaction(&f->smallest, &f->largest, level,
+                                  &parent_index)) {
+        c = new Compaction(level, MaxFileSizeForLevel(level),
+                MaxGrandParentOverlapBytes(level), NumberLevels(), true);
+        c->inputs_[0].push_back(f);
+        c->parent_index_ = parent_index;
+        current_->file_to_compact_ = nullptr;
+      }
     }
   }
 
@@ -1983,6 +2100,11 @@ Compaction* VersionSet::PickCompaction() {
     // which will include the picked file.
     c->inputs_[0].clear();
     current_->GetOverlappingInputs(0, &smallest, &largest, &c->inputs_[0]);
+
+    // If we include more L0 files in the same compaction run it can
+    // cause the 'smallest' and 'largest' key to get extended to a
+    // larger range. So, re-invoke GetRange to get the new key range
+    GetRange(c->inputs_[0], &smallest, &largest);
     if (ParentRangeInCompaction(&smallest, &largest,
                                 level, &c->parent_index_)) {
       delete c;
@@ -2105,16 +2227,23 @@ Compaction* VersionSet::CompactRange(
     return nullptr;
   }
 
+
   // Avoid compacting too much in one shot in case the range is large.
-  const uint64_t limit = MaxFileSizeForLevel(level) *
+  // But we cannot do this for level-0 since level-0 files can overlap
+  // and we must not pick one file and drop another older file if the
+  // two files overlap.
+  if (level > 0) {
+
+    const uint64_t limit = MaxFileSizeForLevel(level) *
                          options_->source_compaction_factor;
-  uint64_t total = 0;
-  for (size_t i = 0; i < inputs.size(); i++) {
-    uint64_t s = inputs[i]->file_size;
-    total += s;
-    if (total >= limit) {
-      inputs.resize(i + 1);
-      break;
+    uint64_t total = 0;
+    for (size_t i = 0; i < inputs.size(); ++i) {
+      uint64_t s = inputs[i]->file_size;
+      total += s;
+      if (total >= limit) {
+        inputs.resize(i + 1);
+        break;
+      }
     }
   }
 
@@ -2146,16 +2275,15 @@ Compaction::Compaction(int level, uint64_t target_file_size,
       overlapped_bytes_(0),
       base_index_(-1),
       parent_index_(-1),
-      score_(0) {
+      score_(0),
+      level_ptrs_(std::vector<size_t>(number_levels)) {
   edit_ = new VersionEdit(number_levels_);
-  level_ptrs_ = new size_t[number_levels_];
   for (int i = 0; i < number_levels_; i++) {
     level_ptrs_[i] = 0;
   }
 }
 
 Compaction::~Compaction() {
-  delete[] level_ptrs_;
   delete edit_;
   if (input_version_ != nullptr) {
     input_version_->Unref();

db/version_set.h

@@ -63,7 +63,8 @@ class Version {
   // Append to *iters a sequence of iterators that will
   // yield the contents of this Version when merged together.
   // REQUIRES: This version has been saved (see VersionSet::SaveTo)
-  void AddIterators(const ReadOptions&, std::vector<Iterator*>* iters);
+  void AddIterators(const ReadOptions&, const EnvOptions& soptions,
+                    std::vector<Iterator*>* iters);
 
   // Lookup the value for key.  If found, store it in *val and
   // return OK.  Else return a non-OK status.  Fills *stats.
@@ -72,8 +73,8 @@ class Version {
     FileMetaData* seek_file;
     int seek_file_level;
   };
-  Status Get(const ReadOptions&, const LookupKey& key, std::string* val,
-             GetStats* stats);
+  void Get(const ReadOptions&, const LookupKey& key, std::string* val,
+           Status* status, GetStats* stats,  const Options& db_option);
 
   // Adds "stats" into the current state.  Returns true if a new
   // compaction may need to be triggered, false otherwise.
@@ -106,7 +107,7 @@ class Version {
       const Slice& begin,         // nullptr means before all keys
       const Slice& end,           // nullptr means after all keys
       std::vector<FileMetaData*>* inputs,
-      int index);                 // start extending from this index
+      unsigned int index);                 // start extending from this index
 
   // Returns true iff some file in the specified level overlaps
   // some part of [*smallest_user_key,*largest_user_key].
@@ -136,7 +137,9 @@ class Version {
   friend class VersionSet;
 
   class LevelFileNumIterator;
-  Iterator* NewConcatenatingIterator(const ReadOptions&, int level) const;
+  Iterator* NewConcatenatingIterator(const ReadOptions&,
+                                     const EnvOptions& soptions,
+                                     int level) const;
 
   VersionSet* vset_;            // VersionSet to which this Version belongs
   Version* next_;               // Next version in linked list
@@ -204,6 +207,7 @@ class VersionSet {
  public:
   VersionSet(const std::string& dbname,
              const Options* options,
+             const EnvOptions& storage_options,
              TableCache* table_cache,
              const InternalKeyComparator*);
   ~VersionSet();
@@ -322,8 +326,7 @@ class VersionSet {
   }
 
   // Add all files listed in any live version to *live.
-  // May also mutate some internal state.
-  void AddLiveFiles(std::set<uint64_t>* live);
+  void AddLiveFiles(std::vector<uint64_t>* live_list);
 
   // Add all files listed in the current version to *live.
   void AddLiveFilesCurrentVersion(std::set<uint64_t>* live);
@@ -378,6 +381,8 @@ class VersionSet {
   // Get the max file size in a given level.
   uint64_t MaxFileSizeForLevel(int level);
 
+  double MaxBytesForLevel(int level);
+
  private:
   class Builder;
   struct ManifestWriter;
@@ -407,8 +412,6 @@ class VersionSet {
 
   bool ManifestContains(const std::string& record) const;
 
-  double MaxBytesForLevel(int level);
-
   int64_t ExpandedCompactionByteSizeLimit(int level);
 
   int64_t MaxGrandParentOverlapBytes(int level);
@@ -454,6 +457,13 @@ class VersionSet {
   // Save us the cost of checking file size twice in LogAndApply
   uint64_t last_observed_manifest_size_;
 
+  // storage options for all reads and writes except compactions
+  const EnvOptions& storage_options_;
+
+  // storage options used for compactions. This is a copy of
+  // storage_options_ but with readaheads set to readahead_compactions_.
+  const EnvOptions storage_options_compactions_;
+
   // No copying allowed
   VersionSet(const VersionSet&);
   void operator=(const VersionSet&);
@@ -557,7 +567,7 @@ class Compaction {
   // is that we are positioned at one of the file ranges for each
   // higher level than the ones involved in this compaction (i.e. for
   // all L >= level_ + 2).
-  size_t* level_ptrs_;
+  std::vector<size_t> level_ptrs_;
 
   // mark (or clear) all files that are being compacted
   void MarkFilesBeingCompacted(bool);

db/write_batch.cc

@@ -7,7 +7,8 @@
 //    count: fixed32
 //    data: record[count]
 // record :=
-//    kTypeValue varstring varstring         |
+//    kTypeValue varstring varstring
+//    kTypeMerge varstring varstring
 //    kTypeDeletion varstring
 // varstring :=
 //    len: varint32
@@ -20,6 +21,7 @@
 #include "db/memtable.h"
 #include "db/write_batch_internal.h"
 #include "util/coding.h"
+#include <stdexcept>
 
 namespace leveldb {
 
@@ -34,6 +36,10 @@ WriteBatch::~WriteBatch() { }
 
 WriteBatch::Handler::~Handler() { }
 
+void WriteBatch::Handler::Merge(const Slice& key, const Slice& value) {
+  throw std::runtime_error("Handler::Merge not implemented!");
+}
+
 void WriteBatch::Clear() {
   rep_.clear();
   rep_.resize(kHeader);
@@ -68,6 +74,14 @@ Status WriteBatch::Iterate(Handler* handler) const {
           return Status::Corruption("bad WriteBatch Delete");
         }
         break;
+      case kTypeMerge:
+        if (GetLengthPrefixedSlice(&input, &key) &&
+            GetLengthPrefixedSlice(&input, &value)) {
+          handler->Merge(key, value);
+        } else {
+          return Status::Corruption("bad WriteBatch Merge");
+        }
+        break;
       default:
         return Status::Corruption("unknown WriteBatch tag");
     }
@@ -108,6 +122,14 @@ void WriteBatch::Delete(const Slice& key) {
   PutLengthPrefixedSlice(&rep_, key);
 }
 
+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);
+}
+
+
 namespace {
 class MemTableInserter : public WriteBatch::Handler {
  public:
@@ -118,6 +140,10 @@ class MemTableInserter : public WriteBatch::Handler {
     mem_->Add(sequence_, kTypeValue, key, value);
     sequence_++;
   }
+  virtual void Merge(const Slice& key, const Slice& value) {
+    mem_->Add(sequence_, kTypeMerge, key, value);
+    sequence_++;
+  }
   virtual void Delete(const Slice& key) {
     mem_->Add(sequence_, kTypeDeletion, key, Slice());
     sequence_++;

db/write_batch_test.cc

@@ -33,6 +33,14 @@ static std::string PrintContents(WriteBatch* b) {
         state.append(")");
         count++;
         break;
+      case kTypeMerge:
+        state.append("Merge(");
+        state.append(ikey.user_key.ToString());
+        state.append(", ");
+        state.append(iter->value().ToString());
+        state.append(")");
+        count++;
+        break;
       case kTypeDeletion:
         state.append("Delete(");
         state.append(ikey.user_key.ToString());

e

@@ -0,0 +1,2 @@
+ar: creating libmemenv.a
+ar: creating libleveldb.a

fbcode.gcc471.sh

@@ -4,10 +4,10 @@
 # fbcode settings.  It uses the latest g++ compiler and also
 # uses jemalloc
 
-TOOLCHAIN_REV=3748fabb2c5e033009597bae1f9ef8bf4b218581
+TOOLCHAIN_REV=83eb773e262fa705eaebbf3de40db71d53fabf2e
 TOOLCHAIN_EXECUTABLES="/mnt/gvfs/third-party/$TOOLCHAIN_REV/centos5.2-native"
 TOOLCHAIN_LIB_BASE="/mnt/gvfs/third-party/$TOOLCHAIN_REV/gcc-4.7.1-glibc-2.14.1"
-TOOL_JEMALLOC=jemalloc-3.0.0/2f45f3a
+TOOL_JEMALLOC=jemalloc-3.3.1/2f45f3a
 
 # location of libhdfs libraries
 if test "$USE_HDFS"; then

hdfs/env_hdfs.h

@@ -235,15 +235,18 @@ class HdfsEnv : public Env {
   }
 
   virtual Status NewSequentialFile(const std::string& fname,
-                                   unique_ptr<SequentialFile>* result);
+                                   unique_ptr<SequentialFile>* result,
+                                   const EnvOptions& options);
 
   virtual Status NewRandomAccessFile(const std::string& fname,
-                                     unique_ptr<RandomAccessFile>* result) {
+                                     unique_ptr<RandomAccessFile>* result,
+                                     const EnvOptions& options) {
     return notsup;
   }
 
   virtual Status NewWritableFile(const std::string& fname,
-                                 unique_ptr<WritableFile>* result) {
+                                 unique_ptr<WritableFile>* result,
+                                 const EnvOptions& options) {
     return notsup;
   }
 

helpers/memenv/memenv.cc

@@ -233,7 +233,8 @@ class InMemoryEnv : public EnvWrapper {
 
   // Partial implementation of the Env interface.
   virtual Status NewSequentialFile(const std::string& fname,
-                                   unique_ptr<SequentialFile>* result) {
+                                   unique_ptr<SequentialFile>* result,
+                                   const EnvOptions& soptions) {
     MutexLock lock(&mutex_);
     if (file_map_.find(fname) == file_map_.end()) {
       *result = NULL;
@@ -245,7 +246,8 @@ class InMemoryEnv : public EnvWrapper {
   }
 
   virtual Status NewRandomAccessFile(const std::string& fname,
-                                     unique_ptr<RandomAccessFile>* result) {
+                                     unique_ptr<RandomAccessFile>* result,
+                                     const EnvOptions& soptions) {
     MutexLock lock(&mutex_);
     if (file_map_.find(fname) == file_map_.end()) {
       *result = NULL;
@@ -257,7 +259,8 @@ class InMemoryEnv : public EnvWrapper {
   }
 
   virtual Status NewWritableFile(const std::string& fname,
-                                 unique_ptr<WritableFile>* result) {
+                                 unique_ptr<WritableFile>* result,
+                                 const EnvOptions& soptions) {
     MutexLock lock(&mutex_);
     if (file_map_.find(fname) != file_map_.end()) {
       DeleteFileInternal(fname);

helpers/memenv/memenv_test.cc

@@ -17,6 +17,7 @@ namespace leveldb {
 class MemEnvTest {
  public:
   Env* env_;
+  const EnvOptions soptions_;
 
   MemEnvTest()
       : env_(NewMemEnv(Env::Default())) {
@@ -40,7 +41,7 @@ TEST(MemEnvTest, Basics) {
   ASSERT_EQ(0U, children.size());
 
   // Create a file.
-  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file));
+  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file, soptions_));
   writable_file.reset();
 
   // Check that the file exists.
@@ -52,7 +53,7 @@ TEST(MemEnvTest, Basics) {
   ASSERT_EQ("f", children[0]);
 
   // Write to the file.
-  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file));
+  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file, soptions_));
   ASSERT_OK(writable_file->Append("abc"));
   writable_file.reset();
 
@@ -71,9 +72,11 @@ TEST(MemEnvTest, Basics) {
   // Check that opening non-existent file fails.
   unique_ptr<SequentialFile> seq_file;
   unique_ptr<RandomAccessFile> rand_file;
-  ASSERT_TRUE(!env_->NewSequentialFile("/dir/non_existent", &seq_file).ok());
+  ASSERT_TRUE(!env_->NewSequentialFile("/dir/non_existent", &seq_file,
+                                       soptions_).ok());
   ASSERT_TRUE(!seq_file);
-  ASSERT_TRUE(!env_->NewRandomAccessFile("/dir/non_existent", &rand_file).ok());
+  ASSERT_TRUE(!env_->NewRandomAccessFile("/dir/non_existent", &rand_file,
+                                         soptions_).ok());
   ASSERT_TRUE(!rand_file);
 
   // Check that deleting works.
@@ -94,13 +97,13 @@ TEST(MemEnvTest, ReadWrite) {
 
   ASSERT_OK(env_->CreateDir("/dir"));
 
-  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file));
+  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file, soptions_));
   ASSERT_OK(writable_file->Append("hello "));
   ASSERT_OK(writable_file->Append("world"));
   writable_file.reset();
 
   // Read sequentially.
-  ASSERT_OK(env_->NewSequentialFile("/dir/f", &seq_file));
+  ASSERT_OK(env_->NewSequentialFile("/dir/f", &seq_file, soptions_));
   ASSERT_OK(seq_file->Read(5, &result, scratch)); // Read "hello".
   ASSERT_EQ(0, result.compare("hello"));
   ASSERT_OK(seq_file->Skip(1));
@@ -113,7 +116,7 @@ TEST(MemEnvTest, ReadWrite) {
   ASSERT_EQ(0U, result.size());
 
   // Random reads.
-  ASSERT_OK(env_->NewRandomAccessFile("/dir/f", &rand_file));
+  ASSERT_OK(env_->NewRandomAccessFile("/dir/f", &rand_file, soptions_));
   ASSERT_OK(rand_file->Read(6, 5, &result, scratch)); // Read "world".
   ASSERT_EQ(0, result.compare("world"));
   ASSERT_OK(rand_file->Read(0, 5, &result, scratch)); // Read "hello".
@@ -139,7 +142,7 @@ TEST(MemEnvTest, Misc) {
   ASSERT_TRUE(!test_dir.empty());
 
   unique_ptr<WritableFile> writable_file;
-  ASSERT_OK(env_->NewWritableFile("/a/b", &writable_file));
+  ASSERT_OK(env_->NewWritableFile("/a/b", &writable_file, soptions_));
 
   // These are no-ops, but we test they return success.
   ASSERT_OK(writable_file->Sync());
@@ -158,14 +161,14 @@ TEST(MemEnvTest, LargeWrite) {
   }
 
   unique_ptr<WritableFile> writable_file;
-  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file));
+  ASSERT_OK(env_->NewWritableFile("/dir/f", &writable_file, soptions_));
   ASSERT_OK(writable_file->Append("foo"));
   ASSERT_OK(writable_file->Append(write_data));
   writable_file.reset();
 
   unique_ptr<SequentialFile> seq_file;
   Slice result;
-  ASSERT_OK(env_->NewSequentialFile("/dir/f", &seq_file));
+  ASSERT_OK(env_->NewSequentialFile("/dir/f", &seq_file, soptions_));
   ASSERT_OK(seq_file->Read(3, &result, scratch)); // Read "foo".
   ASSERT_EQ(0, result.compare("foo"));
 

include/leveldb/compaction_filter.h

@@ -0,0 +1,46 @@
+// Copyright (c) 2013 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_LEVELDB_INCLUDE_COMPACTION_FILTER_H_
+#define STORAGE_LEVELDB_INCLUDE_COMPACTION_FILTER_H_
+
+#include <string>
+
+namespace leveldb {
+
+class Slice;
+
+// CompactionFilter allows an application to modify/delete a key-value at
+// the time of compaction.
+
+class CompactionFilter {
+ public:
+  virtual ~CompactionFilter() {}
+
+
+  // The compaction process invokes this
+  // method for kv that is being compacted. A return value
+  // of false indicates that the kv should be preserved in the
+  // output of this compaction run and a return value of true
+  // indicates that this key-value should be removed from the
+  // output of the compaction.  The application can inspect
+  // the existing value of the key and make decision based on it.
+  //
+  // When the value is to be preserved, the application has the option
+  // to modify the existing_value and pass it back through new_value.
+  // value_changed needs to be set to true in this case.
+  virtual bool Filter(int level,
+                      const Slice& key,
+                      const Slice& existing_value,
+                      std::string* new_value,
+                      bool* value_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;
+};
+
+}  // namespace leveldb
+
+#endif  // STORAGE_LEVELDB_INCLUDE_COMPACTION_FILTER_H_

include/leveldb/db.h

@@ -83,6 +83,14 @@ class DB {
   // Note: consider setting options.sync = true.
   virtual Status Delete(const WriteOptions& options, const Slice& key) = 0;
 
+  // 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,
+                       const Slice& value) = 0;
+
   // Apply the specified updates to the database.
   // Returns OK on success, non-OK on failure.
   // Note: consider setting options.sync = true.
@@ -98,6 +106,20 @@ class DB {
   virtual Status Get(const ReadOptions& options,
                      const Slice& key, std::string* value) = 0;
 
+  // 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
+  // (*values)[i] will be set to some arbitrary value (often ""). Otherwise,
+  // the i'th returned status will have Status::ok() true, and (*values)[i]
+  // will store the value associated with keys[i].
+  //
+  // (*values) will always be resized to be the same size as (keys).
+  // 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<Slice>& keys,
+                                       std::vector<std::string>* values) = 0;
+
   // Return a heap-allocated iterator over the contents of the database.
   // The result of NewIterator() is initially invalid (caller must
   // call one of the Seek methods on the iterator before using it).
@@ -185,7 +207,7 @@ class DB {
   virtual Status GetLiveFiles(std::vector<std::string>&,
                               uint64_t* manifest_file_size) = 0;
 
-  // The sequence number of the most recent transaction. 
+  // The sequence number of the most recent transaction.
   virtual SequenceNumber GetLatestSequenceNumber() = 0;
 
   // Return's an iterator for all writes since the sequence number

include/leveldb/env.h

@@ -28,10 +28,41 @@ class RandomAccessFile;
 class SequentialFile;
 class Slice;
 class WritableFile;
+class Options;
 
 using std::unique_ptr;
 using std::shared_ptr;
 
+
+// Options while opening a file to read/write
+struct EnvOptions {
+
+  // construct with default Options
+  EnvOptions();
+
+  // construct from Options
+  EnvOptions(const Options& options);
+
+  // If true, then allow caching of data in environment buffers
+  bool use_os_buffer;
+
+   // If true, then use mmap to read data
+  bool use_mmap_reads;
+
+   // If true, then use mmap to write data
+  bool use_mmap_writes;
+
+  // If true, set the FD_CLOEXEC on open fd.
+  bool set_fd_cloexec;
+
+  // 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;
+
+};
+
 class Env {
  public:
   Env() { }
@@ -51,7 +82,9 @@ class Env {
   //
   // The returned file will only be accessed by one thread at a time.
   virtual Status NewSequentialFile(const std::string& fname,
-                                   unique_ptr<SequentialFile>* result) = 0;
+                                   unique_ptr<SequentialFile>* result,
+                                   const EnvOptions& options)
+                                   = 0;
 
   // Create a brand new random access read-only file with the
   // specified name.  On success, stores a pointer to the new file in
@@ -61,7 +94,9 @@ class Env {
   //
   // The returned file may be concurrently accessed by multiple threads.
   virtual Status NewRandomAccessFile(const std::string& fname,
-                                     unique_ptr<RandomAccessFile>* result) = 0;
+                                     unique_ptr<RandomAccessFile>* result,
+                                     const EnvOptions& options)
+                                     = 0;
 
   // Create an object that writes to a new file with the specified
   // name.  Deletes any existing file with the same name and creates a
@@ -71,7 +106,8 @@ class Env {
   //
   // The returned file will only be accessed by one thread at a time.
   virtual Status NewWritableFile(const std::string& fname,
-                                 unique_ptr<WritableFile>* result) = 0;
+                                 unique_ptr<WritableFile>* result,
+                                 const EnvOptions& options) = 0;
 
   // Returns true iff the named file exists.
   virtual bool FileExists(const std::string& fname) = 0;
@@ -228,7 +264,7 @@ class RandomAccessFile {
   // the file is opened (and will stay the same while the file is open).
   // Furthermore, it tries to make this ID at most "max_size" bytes. If such an
   // ID can be created this function returns the length of the ID and places it
-  // in "id"; otherwise, this function returns 0, in which case "id" may more
+  // in "id"; otherwise, this function returns 0, in which case "id"
   // may not have been modified.
   //
   // This function guarantees, for IDs from a given environment, two unique ids
@@ -243,6 +279,12 @@ class RandomAccessFile {
     return 0; // Default implementation to prevent issues with backwards
               // compatibility.
   };
+
+
+  enum AccessPattern { NORMAL, RANDOM, SEQUENTIAL, WILLNEED, DONTNEED };
+
+  virtual void Hint(AccessPattern pattern) {}
+
 };
 
 // A file abstraction for sequential writing.  The implementation
@@ -324,6 +366,16 @@ class WritableFile {
     return Status::OK();
   }
 
+  // Sync a file range with disk.
+  // offset is the starting byte of the file range to be synchronized.
+  // nbytes specifies the length of the range to be synchronized.
+  // This asks the OS to initiate flushing the cached data to disk,
+  // without waiting for completion.
+  // Default implementation does nothing.
+  virtual Status RangeSync(off64_t offset, off64_t nbytes) {
+    return Status::OK();
+  }
+
  private:
   size_t last_preallocated_block_;
   size_t preallocation_block_size_;
@@ -398,15 +450,18 @@ class EnvWrapper : public Env {
 
   // The following text is boilerplate that forwards all methods to target()
   Status NewSequentialFile(const std::string& f,
-                           unique_ptr<SequentialFile>* r) {
-    return target_->NewSequentialFile(f, r);
+                           unique_ptr<SequentialFile>* r,
+                           const EnvOptions& options) {
+    return target_->NewSequentialFile(f, r, options);
   }
   Status NewRandomAccessFile(const std::string& f,
-                             unique_ptr<RandomAccessFile>* r) {
-    return target_->NewRandomAccessFile(f, r);
+                             unique_ptr<RandomAccessFile>* r,
+                             const EnvOptions& options) {
+    return target_->NewRandomAccessFile(f, r, options);
   }
-  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r) {
-    return target_->NewWritableFile(f, r);
+  Status NewWritableFile(const std::string& f, unique_ptr<WritableFile>* r,
+                         const EnvOptions& options) {
+    return target_->NewWritableFile(f, r, options);
   }
   bool FileExists(const std::string& f) { return target_->FileExists(f); }
   Status GetChildren(const std::string& dir, std::vector<std::string>* r) {

include/leveldb/ldb_tool.h

@@ -0,0 +1,14 @@
+// Copyright 2008-present Facebook. All Rights Reserved.
+#ifndef STORAGE_LEVELDB_INCLUDE_LDB_TOOL_H
+#define STORAGE_LEVELDB_INCLUDE_LDB_TOOL_H
+#include "leveldb/options.h"
+
+namespace leveldb {
+
+class LDBTool {
+ public:
+  void Run(int argc, char** argv, Options = Options());
+};
+
+} // namespace leveldb
+#endif // STORAGE_LEVELDB_INCLUDE_LDB_TOOL_H

include/leveldb/merge_operator.h

@@ -0,0 +1,74 @@
+// 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_LEVELDB_INCLUDE_MERGE_OPERATOR_H_
+#define STORAGE_LEVELDB_INCLUDE_MERGE_OPERATOR_H_
+
+#include <string>
+
+namespace leveldb {
+
+class Slice;
+class Logger;
+
+// The Merge Operator interface.
+// Client needs to provide an object implementing this interface if Merge
+// operation is accessed.
+// Essentially, MergeOperator specifies the SEMANTICS of a merge, which only
+// client knows. It could be numeric addition, list append, string
+// concatenation, ... , anything.
+// The library, on the other hand, is concerned with the exercise of this
+// interface, at the right time (during get, iteration, compaction...)
+// Note that, even though in principle we don't require any special property
+// of the merge operator, the current rocksdb compaction order does imply that
+// an associative operator could be exercised more naturally (and more
+// efficiently).
+//
+// Refer to my_test.cc for an example of implementation
+//
+class MergeOperator {
+ public:
+  virtual ~MergeOperator() {}
+
+  // Gives the client a way to express the read -> modify -> write semantics
+  // key:      (IN)    The key that's associated with this merge operation.
+  //                   Client could multiplex the merge operator based on it
+  //                   if the key space is partitioned and different subspaces
+  //                   refer to different types of data which have different
+  //                   merge operation semantics
+  // existing: (IN)    null indicates that the key does not exist before this op
+  // value:    (IN)    The passed-in merge operand value (when Merge is issued)
+  // new_value:(OUT)   Client is responsible for filling the merge result here
+  // logger:   (IN)    Client could use this to log errors during merge.
+  //
+  // Note: Merge does not return anything to indicate if a merge is successful
+  //       or not.
+  // Rationale: If a merge failed due to, say de-serialization error, we still
+  //            need to define a consistent merge result. Should we throw away
+  //            the existing value? the merge operand? Or reset the merged value
+  //            to sth? The rocksdb library is not in a position to make the
+  //            right choice. On the other hand, client knows exactly what
+  //            happened during Merge, thus is able to make the best decision.
+  //            Just save the final decision in new_value. logger is passed in,
+  //            in case client wants to leave a trace of what went wrong.
+  virtual void Merge(const Slice& key,
+                     const Slice* existing_value,
+                     const Slice& value,
+                     std::string* new_value,
+                     Logger* logger) const = 0;
+
+
+  // The name of the MergeOperator.  Used to check for MergeOperator
+  // mismatches (i.e., a DB created with one MergeOperator is
+  // accessed using a different MergeOperator)
+  // TODO: the name is currently not stored persistently and thus
+  //       no checking is enforced. Client is responsible for providing
+  //       consistent MergeOperator between DB opens.
+  virtual const char* Name() const = 0;
+
+};
+
+}  // namespace leveldb
+
+#endif  // STORAGE_LEVELDB_INCLUDE_MERGE_OPERATOR_H_

include/leveldb/options.h

@@ -11,6 +11,7 @@
 #include <vector>
 #include <stdint.h>
 #include "leveldb/slice.h"
+#include "leveldb/statistics.h"
 
 namespace leveldb {
 
@@ -19,8 +20,9 @@ class Comparator;
 class Env;
 class FilterPolicy;
 class Logger;
+class MergeOperator;
 class Snapshot;
-class Statistics;
+class CompactionFilter;
 
 using std::shared_ptr;
 
@@ -63,6 +65,23 @@ struct Options {
   // comparator provided to previous open calls on the same DB.
   const Comparator* comparator;
 
+  // REQUIRES: The client must provide a merge operator if Merge operation
+  // needs to be accessed. Calling Merge on a DB without a merge operator
+  // would result in Status::NotSupported. The client must ensure that the
+  // merge operator supplied here has the same name and *exactly* the same
+  // semantics as the merge operator provided to previous open calls on
+  // the same DB. The only exception is reserved for upgrade, where a DB
+  // previously without a merge operator is introduced to Merge operation
+  // for the first time. It's necessary to specify a merge operator when
+  // openning the DB in this case.
+  // Default: nullptr
+  const MergeOperator* merge_operator;
+
+  // Allows an application to modify/delete a key-value during background
+  // compaction.
+  // Default: nullptr
+  const CompactionFilter* compaction_filter;
+
   // If true, the database will be created if it is missing.
   // Default: false
   bool create_if_missing;
@@ -112,6 +131,15 @@ struct Options {
   // Default: 2
   int max_write_buffer_number;
 
+  // The minimum number of write buffers that will be merged together
+  // before writing to storage.  If set to 1, then 
+  // all write buffers are fushed to L0 as individual files and this increases
+  // read amplification because a get request has to check in all of these
+  // files. Also, an in-memory merge may result in writing lesser
+  // data to storage if there are duplicate records in each of these
+  // 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).
@@ -236,6 +264,12 @@ struct Options {
   // by default 'max_bytes_for_level_base' is 10.
   int max_bytes_for_level_multiplier;
 
+  // Different max-size multipliers for different levels.
+  // These are multiplied by max_bytes_for_level_multiplier to arrive
+  // at the max-size of each level.
+  // Default: 1
+  std::vector<int> max_bytes_for_level_multiplier_additional;
+
   // Maximum number of bytes in all compacted files.  We avoid expanding
   // the lower level file set of a compaction if it would make the
   // total compaction cover more than
@@ -258,7 +292,7 @@ struct Options {
   // 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.
-  Statistics* statistics;
+  shared_ptr<Statistics> statistics;
 
   // If true, then the contents of data files are not synced
   // to stable storage. Their contents remain in the OS buffers till the
@@ -325,7 +359,7 @@ struct Options {
   double rate_limit;
 
   // Max time a put will be stalled when rate_limit is enforced
-  int rate_limit_delay_milliseconds;
+  unsigned int rate_limit_delay_milliseconds;
 
   // manifest file is rolled over on reaching this limit.
   // The older manifest file be deleted.
@@ -354,25 +388,6 @@ struct Options {
   // from the database, because otherwise the read can be very slow.
   Options* PrepareForBulkLoad();
 
-  // This method allows an application to modify/delete a key-value at
-  // the time of compaction. The compaction process invokes this
-  // method for every kv that is being compacted. A return value
-  // of false indicates that the kv should be preserved in the
-  // output of this compaction run and a return value of true
-  // indicates that this key-value should be removed from the
-  // output of the compaction.  The application can inspect
-  // the existing value of the key, modify it if needed and
-  // return back the new value for this key. The application
-  // should allocate memory for the Slice object that is used to
-  // return the new value and the leveldb framework will
-  // free up that memory.
-  // The compaction_filter_args, if specified here, are passed
-  // back to the invocation of the CompactionFilter.
-  void* compaction_filter_args;
-  bool (*CompactionFilter)(void* compaction_filter_args,
-         int level, const Slice& key,
-         const Slice& existing_value, Slice** new_value);
-
   // Disable automatic compactions. Manual compactions can still
   // be issued on this database.
   bool disable_auto_compactions;
@@ -396,6 +411,71 @@ struct Options {
   // 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;
+
+  // Reading a single block from a file can cause the OS/FS to start
+  // readaheads of other blocks from the file. Default: true
+  // Note: Deprecated
+  bool allow_readahead;
+
+  // The reads triggered by compaction allows data to be readahead
+  // by the OS/FS. This overrides the setting of 'allow_readahead'
+  // for compaction-reads. Default: true
+  // Note: Deprecated
+  bool allow_readahead_compactions;
+
+  // 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
+  bool allow_mmap_writes;
+
+  // Disable child process inherit open files. Default: true
+  bool is_fd_close_on_exec;
+
+  // Skip log corruption error on recovery (If client is ok with
+  // losing most recent changes)
+  // Default: false
+  bool skip_log_error_on_recovery;
+
+  // if not zero, dump leveldb.stats to LOG every stats_dump_period_sec
+  // 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
+  bool advise_random_on_open;
+
+  // Specify the file access pattern once a compaction is started.
+  // It will be applied to all input files of a compaction.
+  // Default: NORMAL
+  enum { NONE, NORMAL, SEQUENTIAL, WILLNEED } access_hint_on_compaction_start;
+
+  // Use adaptive mutex, which spins in the user space before resorting
+  // to kernel. This could reduce context switch when the mutex is not
+  // heavily contended. However, if the mutex is hot, we could end up
+  // wasting spin time.
+  // Default: false
+  bool use_adaptive_mutex;
+
+  // Allows OS to incrementally sync files to disk while they are being
+  // written, asynchronously, in the background.
+  // Issue one request for every bytes_per_sync written. 0 turns it off.
+  // Default: 0
+  uint64_t bytes_per_sync;
+
 };
 
 // Options that control read operations

include/leveldb/statistics.h

@@ -5,10 +5,13 @@
 #ifndef STORAGE_LEVELDB_INCLUDE_STATISTICS_H_
 #define STORAGE_LEVELDB_INCLUDE_STATISTICS_H_
 
+#include <atomic>
+#include <cassert>
 #include <cstddef>
 #include <cstdint>
 #include <string>
 #include <memory>
+#include <vector>
 
 namespace leveldb {
 
@@ -16,6 +19,7 @@ namespace leveldb {
  * Keep adding ticker's here.
  * Any ticker should have a value less than TICKER_ENUM_MAX.
  * Add a new ticker by assigning it the current value of TICKER_ENUM_MAX
+ * Add a string representation in TickersNameMap below.
  * And incrementing TICKER_ENUM_MAX.
  */
 enum Tickers {
@@ -39,21 +43,85 @@ enum Tickers {
   NO_FILE_CLOSES = 10,
   NO_FILE_OPENS = 11,
   NO_FILE_ERRORS = 12,
-  TICKER_ENUM_MAX = 13,
+  // Time system had to wait to do LO-L1 compactions
+  STALL_L0_SLOWDOWN_MICROS = 13,
+  // Time system had to wait to move memtable to L1.
+  STALL_MEMTABLE_COMPACTION_MICROS = 14,
+  // write throttle because of too many files in L0
+  STALL_L0_NUM_FILES_MICROS = 15,
+  RATE_LIMIT_DELAY_MILLIS = 16,
+
+  NO_ITERATORS = 17, // number of iterators currently open
+
+  // Number of MultiGet calls, keys read, and bytes read
+  NUMBER_MULTIGET_CALLS = 18,
+  NUMBER_MULTIGET_KEYS_READ = 19,
+  NUMBER_MULTIGET_BYTES_READ = 20,
+
+  TICKER_ENUM_MAX = 21
 };
 
+const std::vector<std::pair<Tickers, std::string>> TickersNameMap = {
+  { BLOCK_CACHE_MISS, "rocksdb.block.cache.miss" },
+  { BLOCK_CACHE_HIT, "rocksdb.block.cache.hit" },
+  { BLOOM_FILTER_USEFUL, "rocksdb.bloom.filter.useful" },
+  { COMPACTION_KEY_DROP_NEWER_ENTRY, "rocksdb.compaction.key.drop.new" },
+  { COMPACTION_KEY_DROP_OBSOLETE, "rocksdb.compaction.key.drop.obsolete" },
+  { COMPACTION_KEY_DROP_USER, "rocksdb.compaction.key.drop.user" },
+  { NUMBER_KEYS_WRITTEN, "rocksdb.number.keys.written" },
+  { NUMBER_KEYS_READ, "rocksdb.number.keys.read" },
+  { BYTES_WRITTEN, "rocksdb.bytes.written" },
+  { BYTES_READ, "rocksdb.bytes.read" },
+  { NO_FILE_CLOSES, "rocksdb.no.file.closes" },
+  { NO_FILE_OPENS, "rocksdb.no.file.opens" },
+  { NO_FILE_ERRORS, "rocksdb.no.file.errors" },
+  { STALL_L0_SLOWDOWN_MICROS, "rocksdb.l0.slowdown.micros" },
+  { STALL_MEMTABLE_COMPACTION_MICROS, "rocksdb.memtable.compaction.micros" },
+  { STALL_L0_NUM_FILES_MICROS, "rocksdb.l0.num.files.stall.micros" },
+  { RATE_LIMIT_DELAY_MILLIS, "rocksdb.rate.limit.dleay.millis" },
+  { NO_ITERATORS, "rocksdb.num.iterators" },
+  { NUMBER_MULTIGET_CALLS, "rocksdb.number.multiget.get" },
+  { NUMBER_MULTIGET_KEYS_READ, "rocksdb.number.multiget.keys.read" },
+  { NUMBER_MULTIGET_BYTES_READ, "rocksdb.number.multiget.bytes.read" }
+};
 
 /**
  * Keep adding histogram's here.
  * Any histogram whould have value less than HISTOGRAM_ENUM_MAX
  * Add a new Histogram by assigning it the current value of HISTOGRAM_ENUM_MAX
+ * Add a string representation in HistogramsNameMap below
  * And increment HISTOGRAM_ENUM_MAX
  */
 enum Histograms {
   DB_GET = 0,
   DB_WRITE = 1,
   COMPACTION_TIME = 2,
-  HISTOGRAM_ENUM_MAX = 3,
+  TABLE_SYNC_MICROS = 3,
+  COMPACTION_OUTFILE_SYNC_MICROS = 4,
+  WAL_FILE_SYNC_MICROS = 5,
+  MANIFEST_FILE_SYNC_MICROS = 6,
+  // TIME SPENT IN IO DURING TABLE OPEN
+  TABLE_OPEN_IO_MICROS = 7,
+  DB_MULTIGET = 8,
+  READ_BLOCK_COMPACTION_MICROS = 9,
+  READ_BLOCK_GET_MICROS = 10,
+  WRITE_RAW_BLOCK_MICROS = 11,
+  HISTOGRAM_ENUM_MAX = 12
+};
+
+const std::vector<std::pair<Histograms, std::string>> HistogramsNameMap = {
+  { DB_GET, "rocksdb.db.get.micros" },
+  { DB_WRITE, "rocksdb.db.write.micros" },
+  { COMPACTION_TIME, "rocksdb.compaction.times.micros" },
+  { TABLE_SYNC_MICROS, "rocksdb.table.sync.micros" },
+  { COMPACTION_OUTFILE_SYNC_MICROS, "rocksdb.compaction.outfile.sync.micros" },
+  { WAL_FILE_SYNC_MICROS, "rocksdb.wal.file.sync.micros" },
+  { MANIFEST_FILE_SYNC_MICROS, "rocksdb.manifest.file.sync.micros" },
+  { TABLE_OPEN_IO_MICROS, "rocksdb.table.open.io.micros" },
+  { DB_MULTIGET, "rocksdb.db.multiget.micros" },
+  { READ_BLOCK_COMPACTION_MICROS, "rocksdb.read.block.compaction.micros" },
+  { READ_BLOCK_GET_MICROS, "rocksdb.read.block.get.micros" },
+  { WRITE_RAW_BLOCK_MICROS, "rocksdb.write.raw.block.micros" }
 };
 
 struct HistogramData {
@@ -72,7 +140,6 @@ class Histogram {
   virtual ~Histogram();
   // Add a value to be recorded in the histogram.
   virtual void Add(uint64_t value) = 0;
-  virtual void Add(double value) = 0;
 
   virtual std::string ToString() const = 0;
 
@@ -93,7 +160,7 @@ class Ticker {
  public:
   Ticker() : count_(0) { }
 
-  inline void recordTick() {
+ inline void recordTick() {
     count_++;
   }
 
@@ -106,8 +173,7 @@ class Ticker {
   }
 
  private:
-  uint64_t count_;
-
+  std::atomic_uint_fast64_t count_;
 };
 
 // Analyze the performance of a db
@@ -116,18 +182,23 @@ class Statistics {
 
   virtual long getTickerCount(Tickers tickerType) = 0;
   virtual void recordTick(Tickers tickerType, uint64_t count = 0) = 0;
-  virtual void measureTime(Histograms histogramType, uint64_t count) = 0;
-  virtual void measureTime(Histograms histogramType, double count) = 0;
+  virtual void measureTime(Histograms histogramType, uint64_t time) = 0;
 
   virtual void histogramData(Histograms type, HistogramData * const data) = 0;
-
+  // String representation of the statistic object.
+  std::string ToString();
 };
 
+// Create a concrete DBStatistics object
+std::shared_ptr<Statistics> CreateDBStatistics();
+
 // Ease of Use functions
-inline void RecordTick(Statistics* const statistics,
+inline void RecordTick(std::shared_ptr<Statistics> statistics,
                        Tickers ticker,
                        uint64_t count = 1) {
-  if (statistics != nullptr) {
+  assert(HistogramsNameMap.size() == HISTOGRAM_ENUM_MAX);
+  assert(TickersNameMap.size() == TICKER_ENUM_MAX);
+  if (statistics) {
     statistics->recordTick(ticker, count);
   }
 }

include/leveldb/status.h

@@ -47,6 +47,9 @@ class Status {
   static Status IOError(const Slice& msg, const Slice& msg2 = Slice()) {
     return Status(kIOError, msg, msg2);
   }
+  static Status MergeInProgress(const Slice& msg, const Slice& msg2 = Slice()) {
+    return Status(kMergeInProgress, msg, msg2);
+  }
 
   // Returns true iff the status indicates success.
   bool ok() const { return (state_ == nullptr); }
@@ -66,6 +69,9 @@ class Status {
   // Returns true iff the status indicates an IOError.
   bool IsIOError() const { return code() == kIOError; }
 
+  // Returns true iff the status indicates an MergeInProgress.
+  bool IsMergeInProgress() const { return code() == kMergeInProgress; }
+
   // Return a string representation of this status suitable for printing.
   // Returns the string "OK" for success.
   std::string ToString() const;
@@ -84,7 +90,8 @@ class Status {
     kCorruption = 2,
     kNotSupported = 3,
     kInvalidArgument = 4,
-    kIOError = 5
+    kIOError = 5,
+    kMergeInProgress = 6
   };
 
   Code code() const {

include/leveldb/write_batch.h

@@ -36,6 +36,10 @@ class WriteBatch {
   // Store the mapping "key->value" in the database.
   void Put(const Slice& key, const Slice& value);
 
+  // Merge "value" with the existing value of "key" in the database.
+  // "key->merge(existing, value)"
+  void Merge(const Slice& key, const Slice& value);
+
   // If the database contains a mapping for "key", erase it.  Else do nothing.
   void Delete(const Slice& key);
 
@@ -47,6 +51,10 @@ class WriteBatch {
    public:
     virtual ~Handler();
     virtual void Put(const Slice& key, const Slice& value) = 0;
+    // Merge is not pure virtual. Otherwise, we would break existing
+    // clients of Handler on a source code level.
+    // The default implementation simply throws a runtime exception.
+    virtual void Merge(const Slice& key, const Slice& value);
     virtual void Delete(const Slice& key) = 0;
   };
   Status Iterate(Handler* handler) const;

include/utilities/utility_db.h

@@ -0,0 +1,55 @@
+// 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 LEVELDB_INCLUDE_UTILITIES_UTILITY_DB_H_
+#define LEVELDB_INCLUDE_UTILITIES_UTILITY_DB_H_
+
+#include "leveldb/db.h"
+
+namespace leveldb {
+
+// This class contains APIs to open leveldb with specific support eg. TTL
+class UtilityDB {
+
+  public:
+    // Open the database with TTL support.
+    //
+    // USE-CASES:
+    // This API should be used to open the db when key-values inserted are
+    //  meant to be removed from the db in a non-strict 'ttl' amount of time
+    //  Therefore, this guarantees that key-values inserted will remain in the
+    //  db for >= ttl amount of time and the db will make efforts to remove the
+    //  key-values as soon as possible after ttl seconds of their insertion.
+    //
+    // BEHAVIOUR:
+    // TTL is accepted in seconds
+    // (int32_t)Timestamp(creation) is suffixed to values in Put internally
+    // Expired TTL values deleted in compaction only:(Timestamp+ttl<time_now)
+    // Get/Iterator may return expired entries(compaction not run on them yet)
+    // Different TTL may be used during different Opens
+    // Example: Open1 at t=0 with ttl=4 and insert k1,k2, close at t=2
+    //          Open2 at t=3 with ttl=5. Now k1,k2 should be deleted at t>=5
+    // read_only=true opens in the usual read-only mode. Compactions will not be
+    //  triggered(neither manual nor automatic), so no expired entries removed
+    //
+    // CONSTRAINTS:
+    // The caller must not specify any compaction-filter in options
+    // Not specifying/passing or non-positive TTL behaves like TTL = infinity
+    //
+    // !!!WARNING!!!:
+    // Calling DB::Open directly to re-open a db created by this API will get
+    //  corrupt values(timestamp suffixed) and no ttl effect will be there
+    //  during the second Open, so use this API consistently to open the db
+    // Be careful when passing ttl with a small positive value because the
+    //  whole database may be deleted in a small amount of time
+    static Status OpenTtlDB(const Options& options,
+                            const std::string& name,
+                            DB** dbptr,
+                            int32_t ttl = 0,
+                            bool read_only = false);
+};
+
+} //  namespace leveldb
+
+#endif  //  LEVELDB_INCLUDE_UTILITIES_UTILITY_DB_H_

port/port_posix.cc

@@ -19,7 +19,24 @@ static void PthreadCall(const char* label, int result) {
   }
 }
 
-Mutex::Mutex() { PthreadCall("init mutex", pthread_mutex_init(&mu_, NULL)); }
+Mutex::Mutex(bool adaptive) {
+#ifdef OS_LINUX
+  if (!adaptive) {
+    PthreadCall("init mutex", pthread_mutex_init(&mu_, NULL));
+  } else {
+    pthread_mutexattr_t mutex_attr;
+    PthreadCall("init mutex attr", pthread_mutexattr_init(&mutex_attr));
+    PthreadCall("set mutex attr",
+                pthread_mutexattr_settype(&mutex_attr,
+                                          PTHREAD_MUTEX_ADAPTIVE_NP));
+    PthreadCall("init mutex", pthread_mutex_init(&mu_, &mutex_attr));
+    PthreadCall("destroy mutex attr",
+                pthread_mutexattr_destroy(&mutex_attr));
+  }
+#else // ignore adaptive for non-linux platform
+  PthreadCall("init mutex", pthread_mutex_init(&mu_, NULL));
+#endif // OS_LINUX
+}
 
 Mutex::~Mutex() { PthreadCall("destroy mutex", pthread_mutex_destroy(&mu_)); }
 

port/port_posix.h

@@ -82,7 +82,7 @@ class CondVar;
 
 class Mutex {
  public:
-  Mutex();
+  /* implicit */ Mutex(bool adaptive = false);
   ~Mutex();
 
   void Lock();
@@ -192,17 +192,20 @@ inline bool Zlib_Compress(const CompressionOptions& opts, const char* input,
   _stream.avail_out = length;
   _stream.next_out = (Bytef *)&(*output)[0];
 
-  int old_sz =0, new_sz =0;
-  while(_stream.next_in != NULL && _stream.avail_in != 0) {
+  int old_sz =0, new_sz =0, new_sz_delta =0;
+  bool done = false;
+  while (!done) {
     int st = deflate(&_stream, Z_FINISH);
     switch (st) {
       case Z_STREAM_END:
+        done = true;
         break;
       case Z_OK:
         // No output space. Increase the output space by 20%.
         // (Should we fail the compression since it expands the size?)
         old_sz = output->size();
-        new_sz = (int)(output->size() * 1.2);
+        new_sz_delta = (int)(output->size() * 0.2);
+        new_sz = output->size() + (new_sz_delta < 10 ? 10 : new_sz_delta);
         output->resize(new_sz);
         // Set more output.
         _stream.next_out = (Bytef *)&(*output)[old_sz];
@@ -228,13 +231,13 @@ inline char* Zlib_Uncompress(const char* input_data, size_t input_length,
   z_stream _stream;
   memset(&_stream, 0, sizeof(z_stream));
 
-  // For raw inflate, the windowBits should be Ð8..Ð15.
+  // For raw inflate, the windowBits should be -8..-15.
   // If windowBits is bigger than zero, it will use either zlib
   // header or gzip header. Adding 32 to it will do automatic detection.
   int st = inflateInit2(&_stream,
       windowBits > 0 ? windowBits + 32 : windowBits);
   if (st != Z_OK) {
-    return NULL;
+    return nullptr;
   }
 
   _stream.next_in = (Bytef *)input_data;
@@ -248,17 +251,22 @@ inline char* Zlib_Uncompress(const char* input_data, size_t input_length,
   _stream.next_out = (Bytef *)output;
   _stream.avail_out = output_len;
 
-  char* tmp = NULL;
+  char* tmp = nullptr;
+  int output_len_delta;
+  bool done = false;
 
-  while(_stream.next_in != NULL && _stream.avail_in != 0) {
+  //while(_stream.next_in != nullptr && _stream.avail_in != 0) {
+  while (!done) {
     int st = inflate(&_stream, Z_SYNC_FLUSH);
     switch (st) {
       case Z_STREAM_END:
+        done = true;
         break;
       case Z_OK:
         // No output space. Increase the output space by 20%.
         old_sz = output_len;
-        output_len = (int)(output_len * 1.2);
+        output_len_delta = (int)(output_len * 0.2);
+        output_len += output_len_delta < 10 ? 10 : output_len_delta;
         tmp = new char[output_len];
         memcpy(tmp, output, old_sz);
         delete[] output;
@@ -272,7 +280,7 @@ inline char* Zlib_Uncompress(const char* input_data, size_t input_length,
       default:
         delete[] output;
         inflateEnd(&_stream);
-        return NULL;
+        return nullptr;
     }
   }
 
@@ -281,7 +289,7 @@ inline char* Zlib_Uncompress(const char* input_data, size_t input_length,
   return output;
 #endif
 
-  return NULL;
+  return nullptr;
 }
 
 inline bool BZip2_Compress(const CompressionOptions& opts, const char* input,
@@ -311,7 +319,7 @@ inline bool BZip2_Compress(const CompressionOptions& opts, const char* input,
   _stream.avail_out = length;
 
   int old_sz =0, new_sz =0;
-  while(_stream.next_in != NULL && _stream.avail_in != 0) {
+  while(_stream.next_in != nullptr && _stream.avail_in != 0) {
     int st = BZ2_bzCompress(&_stream, BZ_FINISH);
     switch (st) {
       case BZ_STREAM_END:
@@ -349,7 +357,7 @@ inline char*  BZip2_Uncompress(const char* input_data, size_t input_length,
 
   int st = BZ2_bzDecompressInit(&_stream, 0, 0);
   if (st != BZ_OK) {
-    return NULL;
+    return nullptr;
   }
 
   _stream.next_in = (char *)input_data;
@@ -363,9 +371,9 @@ inline char*  BZip2_Uncompress(const char* input_data, size_t input_length,
   _stream.next_out = (char *)output;
   _stream.avail_out = output_len;
 
-  char* tmp = NULL;
+  char* tmp = nullptr;
 
-  while(_stream.next_in != NULL && _stream.avail_in != 0) {
+  while(_stream.next_in != nullptr && _stream.avail_in != 0) {
     int st = BZ2_bzDecompress(&_stream);
     switch (st) {
       case BZ_STREAM_END:
@@ -387,7 +395,7 @@ inline char*  BZip2_Uncompress(const char* input_data, size_t input_length,
       default:
         delete[] output;
         BZ2_bzDecompressEnd(&_stream);
-        return NULL;
+        return nullptr;
     }
   }
 
@@ -395,7 +403,7 @@ inline char*  BZip2_Uncompress(const char* input_data, size_t input_length,
   BZ2_bzDecompressEnd(&_stream);
   return output;
 #endif
-  return NULL;
+  return nullptr;
 }
 
 inline bool GetHeapProfile(void (*func)(void*, const char*, int), void* arg) {

port/stack_trace.cc

@@ -0,0 +1,97 @@
+#include "util/stack_trace.h"
+
+#ifdef OS_LINUX
+
+#include <execinfo.h>
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+namespace leveldb {
+
+static const char* GetExecutableName()
+{
+  static char name[1024];
+
+  char link[1024];
+  snprintf(link, sizeof(link), "/proc/%d/exe", getpid());
+  auto read = readlink(link, name, sizeof(name));
+  if (-1 == read) {
+    return nullptr;
+  } else {
+    name[read] = 0;
+    return name;
+  }
+}
+
+static void StackTraceHandler(int sig) {
+  // reset to default handler
+  signal(sig, SIG_DFL);
+
+  fprintf(stderr, "Received signal %d (%s)\n", sig, strsignal(sig));
+
+  const int kMaxFrames = 100;
+  void *frames[kMaxFrames];
+
+  auto num_frames = backtrace(frames, kMaxFrames);
+  auto symbols = backtrace_symbols(frames, num_frames);
+
+  auto executable = GetExecutableName();
+
+  const int kSkip = 2; // skip the top two signal handler related frames
+
+  for (int i = kSkip; i < num_frames; ++i)
+  {
+    fprintf(stderr, "#%-2d %p ", i - kSkip, frames[i]);
+    if (symbols) {
+      fprintf(stderr, "%s ", symbols[i]);
+    }
+    if (executable) {
+      // out source to addr2line, for the address translation
+      const int kLineMax = 256;
+      char cmd[kLineMax];
+      sprintf(cmd,"addr2line %p -e %s 2>&1", frames[i] , executable);
+      auto f = popen(cmd, "r");
+      if (f) {
+        char line[kLineMax];
+        while (fgets(line, sizeof(line), f)) {
+          fprintf(stderr, "%s", line);
+        }
+        pclose(f);
+      } else {
+        fprintf(stderr, "\n");
+      }
+    } else {
+      fprintf(stderr, "\n");
+    }
+  }
+
+  // re-signal to default handler (so we still get core dump if needed...)
+  raise(sig);
+}
+
+void InstallStackTraceHandler() {
+  // just use the plain old signal as it's simple and sufficient
+  // for this use case
+  signal(SIGILL, StackTraceHandler);
+  signal(SIGSEGV, StackTraceHandler);
+  signal(SIGBUS, StackTraceHandler);
+  signal(SIGABRT, StackTraceHandler);
+
+  printf("Installed stack trace handler for SIGILL SIGSEGV SIGBUS SIGABRT\n");
+
+}
+
+}   // namespace leveldb
+
+#else // no-op for non-linux system for now
+
+namespace leveldb {
+
+void InstallStackTraceHandler() {}
+
+}
+
+#endif // OS_LINUX

table/block.cc

@@ -23,7 +23,8 @@ inline uint32_t Block::NumRestarts() const {
 Block::Block(const BlockContents& contents)
     : data_(contents.data.data()),
       size_(contents.data.size()),
-      owned_(contents.heap_allocated) {
+      owned_(contents.heap_allocated),
+      cachable_(contents.cachable) {
   if (size_ < sizeof(uint32_t)) {
     size_ = 0;  // Error marker
   } else {

table/block.h

@@ -22,6 +22,7 @@ class Block {
   ~Block();
 
   size_t size() const { return size_; }
+  bool   isCachable() const { return cachable_; }
   Iterator* NewIterator(const Comparator* comparator);
 
  private:
@@ -31,6 +32,7 @@ class Block {
   size_t size_;
   uint32_t restart_offset_;     // Offset in data_ of restart array
   bool owned_;                  // Block owns data_[]
+  bool cachable_;
 
   // No copying allowed
   Block(const Block&);

table/block_builder.cc

@@ -60,6 +60,21 @@ size_t BlockBuilder::CurrentSizeEstimate() const {
           sizeof(uint32_t));                      // Restart array length
 }
 
+size_t BlockBuilder::EstimateSizeAfterKV(const Slice& key, const Slice& value)
+  const {
+  size_t estimate = CurrentSizeEstimate();
+  estimate += key.size() + value.size();
+  if (counter_ >= options_->block_restart_interval) {
+    estimate += sizeof(uint32_t); // a new restart entry.
+  }
+
+  estimate += sizeof(int32_t); // varint for shared prefix length.
+  estimate += VarintLength(key.size()); // varint for key length.
+  estimate += VarintLength(value.size()); // varint for value length.
+
+  return estimate;
+}
+
 Slice BlockBuilder::Finish() {
   // Append restart array
   for (size_t i = 0; i < restarts_.size(); i++) {

table/block_builder.h

@@ -34,6 +34,9 @@ class BlockBuilder {
   // we are building.
   size_t CurrentSizeEstimate() const;
 
+  // Returns an estimated block size after appending key and value.
+  size_t EstimateSizeAfterKV(const Slice& key, const Slice& value) const;
+
   // Return true iff no entries have been added since the last Reset()
   bool empty() const {
     return buffer_.empty();

table/block_test.cc

@@ -1,9 +1,6 @@
 // Copyright (c) 2012 Facebook.
 // 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 "leveldb/table.h"
-
 #include <string>
 #include "db/dbformat.h"
 #include "db/memtable.h"
@@ -14,6 +11,7 @@
 #include "leveldb/table_builder.h"
 #include "table/block.h"
 #include "table/block_builder.h"
+#include "table/table.h"
 #include "table/format.h"
 #include "util/random.h"
 #include "util/testharness.h"
@@ -60,7 +58,7 @@ TEST(BlockTest, SimpleTest) {
   // read serialized contents of the block
   Slice rawblock = builder.Finish();
 
-  // create block reader 
+  // create block reader
   BlockContents contents;
   contents.data = rawblock;
   contents.cachable = false;
@@ -71,7 +69,7 @@ TEST(BlockTest, SimpleTest) {
   int count = 0;
   Iterator* iter = reader.NewIterator(options.comparator);
   for (iter->SeekToFirst();iter->Valid(); count++, iter->Next()) {
-    
+
     // read kv from block
     Slice k = iter->key();
     Slice v = iter->value();
@@ -81,8 +79,8 @@ TEST(BlockTest, SimpleTest) {
     ASSERT_EQ(v.ToString().compare(values[count]), 0);
   }
   delete iter;
-  
-  // read block contents randomly 
+
+  // read block contents randomly
   iter = reader.NewIterator(options.comparator);
   for (int i = 0; i < num_records; i++) {
 

table/format.cc

@@ -66,10 +66,10 @@ Status Footer::DecodeFrom(Slice* input) {
   return result;
 }
 
-Status ReadBlock(RandomAccessFile* file,
-                 const ReadOptions& options,
-                 const BlockHandle& handle,
-                 BlockContents* result) {
+Status ReadBlockContents(RandomAccessFile* file,
+                         const ReadOptions& options,
+                         const BlockHandle& handle,
+                         BlockContents* result) {
   result->data = Slice();
   result->cachable = false;
   result->heap_allocated = false;

table/format.h

@@ -91,10 +91,10 @@ struct BlockContents {
 
 // Read the block identified by "handle" from "file".  On failure
 // return non-OK.  On success fill *result and return OK.
-extern Status ReadBlock(RandomAccessFile* file,
-                        const ReadOptions& options,
-                        const BlockHandle& handle,
-                        BlockContents* result);
+extern Status ReadBlockContents(RandomAccessFile* file,
+                                const ReadOptions& options,
+                                const BlockHandle& handle,
+                                BlockContents* result);
 
 // Implementation details follow.  Clients should ignore,
 

table/table.cc

@@ -2,7 +2,7 @@
 // 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 "leveldb/table.h"
+#include "table/table.h"
 
 #include "leveldb/cache.h"
 #include "leveldb/comparator.h"
@@ -10,11 +10,14 @@
 #include "leveldb/filter_policy.h"
 #include "leveldb/options.h"
 #include "leveldb/statistics.h"
+
 #include "table/block.h"
 #include "table/filter_block.h"
 #include "table/format.h"
 #include "table/two_level_iterator.h"
+
 #include "util/coding.h"
+#include "util/stop_watch.h"
 
 namespace leveldb {
 
@@ -29,8 +32,12 @@ struct Table::Rep {
     delete [] filter_data;
     delete index_block;
   }
+  Rep(const EnvOptions& storage_options) :
+    soptions(storage_options) {
+  }
 
   Options options;
+  const EnvOptions& soptions;
   Status status;
   unique_ptr<RandomAccessFile> file;
   char cache_key_prefix[kMaxCacheKeyPrefixSize];
@@ -61,7 +68,34 @@ void Table::SetupCacheKeyPrefix(Rep* rep) {
   }
 }
 
+namespace {  // anonymous namespace, not visible externally
+
+// Read the block identified by "handle" from "file".
+// The only relevant option is options.verify_checksums for now.
+// Set *didIO to true if didIO is not null.
+// On failure return non-OK.
+// On success fill *result and return OK - caller owns *result
+Status ReadBlock(RandomAccessFile* file,
+                 const ReadOptions& options,
+                 const BlockHandle& handle,
+                 Block** result,
+                 bool* didIO = nullptr) {
+  BlockContents contents;
+  Status s = ReadBlockContents(file, options, handle, &contents);
+  if (s.ok()) {
+    *result = new Block(contents);
+  }
+
+  if (didIO) {
+    *didIO = true;
+  }
+  return s;
+}
+
+} // end of anonymous namespace
+
 Status Table::Open(const Options& options,
+                   const EnvOptions& soptions,
                    unique_ptr<RandomAccessFile>&& file,
                    uint64_t size,
                    unique_ptr<Table>* table) {
@@ -87,20 +121,14 @@ Status Table::Open(const Options& options,
   s = footer.DecodeFrom(&footer_input);
   if (!s.ok()) return s;
 
-  // Read the index block
-  BlockContents contents;
   Block* index_block = nullptr;
-  if (s.ok()) {
-    s = ReadBlock(file.get(), ReadOptions(), footer.index_handle(), &contents);
-    if (s.ok()) {
-      index_block = new Block(contents);
-    }
-  }
+  // TODO: we never really verify check sum for index block
+  s = ReadBlock(file.get(), ReadOptions(), footer.index_handle(), &index_block);
 
   if (s.ok()) {
     // We've successfully read the footer and the index block: we're
     // ready to serve requests.
-    Rep* rep = new Table::Rep;
+    Rep* rep = new Table::Rep(soptions);
     rep->options = options;
     rep->file = std::move(file);
     rep->metaindex_handle = footer.metaindex_handle();
@@ -117,6 +145,25 @@ Status Table::Open(const Options& options,
   return s;
 }
 
+void Table::SetupForCompaction() {
+  switch (rep_->options.access_hint_on_compaction_start) {
+    case Options::NONE:
+      break;
+    case Options::NORMAL:
+      rep_->file->Hint(RandomAccessFile::NORMAL);
+      break;
+    case Options::SEQUENTIAL:
+      rep_->file->Hint(RandomAccessFile::SEQUENTIAL);
+      break;
+    case Options::WILLNEED:
+      rep_->file->Hint(RandomAccessFile::WILLNEED);
+      break;
+    default:
+      assert(false);
+  }
+  compaction_optimized_ = true;
+}
+
 void Table::ReadMeta(const Footer& footer) {
   if (rep_->options.filter_policy == nullptr) {
     return;  // Do not need any metadata
@@ -124,14 +171,13 @@ void Table::ReadMeta(const Footer& footer) {
 
   // TODO(sanjay): Skip this if footer.metaindex_handle() size indicates
   // it is an empty block.
-  ReadOptions opt;
-  BlockContents contents;
-  if (!ReadBlock(rep_->file.get(), opt, footer.metaindex_handle(),
-                 &contents).ok()) {
+  //  TODO: we never really verify check sum for meta index block
+  Block* meta = nullptr;
+  if (!ReadBlock(rep_->file.get(), ReadOptions(), footer.metaindex_handle(),
+                 &meta).ok()) {
     // Do not propagate errors since meta info is not needed for operation
     return;
   }
-  Block* meta = new Block(contents);
 
   Iterator* iter = meta->NewIterator(BytewiseComparator());
   std::string key = "filter.";
@@ -151,11 +197,11 @@ void Table::ReadFilter(const Slice& filter_handle_value) {
     return;
   }
 
-  // We might want to unify with ReadBlock() if we start
+  // TODO: We might want to unify with ReadBlock() if we start
   // requiring checksum verification in Table::Open.
   ReadOptions opt;
   BlockContents block;
-  if (!ReadBlock(rep_->file.get(), opt, filter_handle, &block).ok()) {
+  if (!ReadBlockContents(rep_->file.get(), opt, filter_handle, &block).ok()) {
     return;
   }
   if (block.heap_allocated) {
@@ -188,10 +234,11 @@ static void ReleaseBlock(void* arg, void* h) {
 Iterator* Table::BlockReader(void* arg,
                              const ReadOptions& options,
                              const Slice& index_value,
-                             bool* didIO) {
+                             bool* didIO,
+                             bool for_compaction) {
   Table* table = reinterpret_cast<Table*>(arg);
   Cache* block_cache = table->rep_->options.block_cache.get();
-  Statistics* const statistics = table->rep_->options.statistics;
+  std::shared_ptr<Statistics> statistics = table->rep_->options.statistics;
   Block* block = nullptr;
   Cache::Handle* cache_handle = nullptr;
 
@@ -202,7 +249,6 @@ Iterator* Table::BlockReader(void* arg,
   // can add more features in the future.
 
   if (s.ok()) {
-    BlockContents contents;
     if (block_cache != nullptr) {
       char cache_key[kMaxCacheKeyPrefixSize + kMaxVarint64Length];
       const size_t cache_key_prefix_size = table->rep_->cache_key_prefix_size;
@@ -219,28 +265,29 @@ Iterator* Table::BlockReader(void* arg,
 
         RecordTick(statistics, BLOCK_CACHE_HIT);
       } else {
-        s = ReadBlock(table->rep_->file.get(), options, handle, &contents);
-        if (s.ok()) {
-          block = new Block(contents);
-          if (contents.cachable && options.fill_cache) {
-            cache_handle = block_cache->Insert(
-                key, block, block->size(), &DeleteCachedBlock);
-          }
+        Histograms histogram = for_compaction ?
+          READ_BLOCK_COMPACTION_MICROS : READ_BLOCK_GET_MICROS;
+        {  // block for stop watch
+          StopWatch sw(table->rep_->options.env, statistics, histogram);
+          s = ReadBlock(
+                table->rep_->file.get(),
+                options,
+                handle,
+                &block,
+                didIO
+              );
         }
-        if (didIO != nullptr) {
-          *didIO = true; // we did some io from storage
+        if (s.ok()) {
+          if (block->isCachable() && options.fill_cache) {
+            cache_handle = block_cache->Insert(
+              key, block, block->size(), &DeleteCachedBlock);
+          }
         }
 
         RecordTick(statistics, BLOCK_CACHE_MISS);
       }
     } else {
-      s = ReadBlock(table->rep_->file.get(), options, handle, &contents);
-      if (s.ok()) {
-        block = new Block(contents);
-      }
-      if (didIO != nullptr) {
-        *didIO = true; // we did some io from storage
-      }
+      s = ReadBlock(table->rep_->file.get(), options, handle, &block, didIO);
     }
   }
 
@@ -260,23 +307,26 @@ Iterator* Table::BlockReader(void* arg,
 
 Iterator* Table::BlockReader(void* arg,
                              const ReadOptions& options,
-                             const Slice& index_value) {
-  return BlockReader(arg, options, index_value, nullptr);
+                             const EnvOptions& soptions,
+                             const Slice& index_value,
+                             bool for_compaction) {
+  return BlockReader(arg, options, index_value, nullptr, for_compaction);
 }
 
 Iterator* Table::NewIterator(const ReadOptions& options) const {
   return NewTwoLevelIterator(
       rep_->index_block->NewIterator(rep_->options.comparator),
-      &Table::BlockReader, const_cast<Table*>(this), options);
+      &Table::BlockReader, const_cast<Table*>(this), options, rep_->soptions);
 }
 
 Status Table::InternalGet(const ReadOptions& options, const Slice& k,
                           void* arg,
-                          void (*saver)(void*, const Slice&, const Slice&, bool)) {
+                          bool (*saver)(void*, const Slice&, const Slice&,
+                                        bool)) {
   Status s;
   Iterator* iiter = rep_->index_block->NewIterator(rep_->options.comparator);
-  iiter->Seek(k);
-  if (iiter->Valid()) {
+  bool done = false;
+  for (iiter->Seek(k); iiter->Valid() && !done; iiter->Next()) {
     Slice handle_value = iiter->value();
     FilterBlockReader* filter = rep_->filter;
     BlockHandle handle;
@@ -284,14 +334,20 @@ Status Table::InternalGet(const ReadOptions& options, const Slice& k,
         handle.DecodeFrom(&handle_value).ok() &&
         !filter->KeyMayMatch(handle.offset(), k)) {
       // Not found
+      // TODO: think about interaction with Merge. If a user key cannot
+      // cross one data block, we should be fine.
       RecordTick(rep_->options.statistics, BLOOM_FILTER_USEFUL);
+      break;
     } else {
       bool didIO = false;
       Iterator* block_iter = BlockReader(this, options, iiter->value(),
                                          &didIO);
-      block_iter->Seek(k);
-      if (block_iter->Valid()) {
-        (*saver)(arg, block_iter->key(), block_iter->value(), didIO);
+
+      for (block_iter->Seek(k); block_iter->Valid(); block_iter->Next()) {
+        if (!(*saver)(arg, block_iter->key(), block_iter->value(), didIO)) {
+          done = true;
+          break;
+        }
       }
       s = block_iter->status();
       delete block_iter;
@@ -304,8 +360,9 @@ Status Table::InternalGet(const ReadOptions& options, const Slice& k,
   return s;
 }
 
-void SaveDidIO(void* arg, const Slice& key, const Slice& value, bool didIO) {
+bool SaveDidIO(void* arg, const Slice& key, const Slice& value, bool didIO) {
   *reinterpret_cast<bool*>(arg) = didIO;
+  return false;
 }
 bool Table::TEST_KeyInCache(const ReadOptions& options, const Slice& key) {
   // We use InternalGet() as it has logic that checks whether we read the

table/table.h

@@ -8,6 +8,7 @@
 #include <memory>
 #include <stdint.h>
 #include "leveldb/iterator.h"
+#include "leveldb/env.h"
 
 namespace leveldb {
 
@@ -39,6 +40,7 @@ class Table {
   //
   // *file must remain live while this Table is in use.
   static Status Open(const Options& options,
+                     const EnvOptions& soptions,
                      unique_ptr<RandomAccessFile>&& file,
                      uint64_t file_size,
                      unique_ptr<Table>* table);
@@ -62,23 +64,29 @@ class Table {
   // REQUIRES: key is in this table.
   bool TEST_KeyInCache(const ReadOptions& options, const Slice& key);
 
+  // Set up the table for Compaction. Might change some parameters with
+  // posix_fadvise
+  void SetupForCompaction();
+
  private:
   struct Rep;
   Rep* rep_;
-
-  explicit Table(Rep* rep) { rep_ = rep; }
-  static Iterator* BlockReader(void*, const ReadOptions&, const Slice&);
+  bool compaction_optimized_;
+  explicit Table(Rep* rep) : compaction_optimized_(false) { rep_ = rep; }
+  static Iterator* BlockReader(void*, const ReadOptions&,
+                               const EnvOptions& soptions, const Slice&,
+                               bool for_compaction);
   static Iterator* BlockReader(void*, const ReadOptions&, const Slice&,
-                               bool* didIO);
+                               bool* didIO, bool for_compaction = false);
 
-  // Calls (*handle_result)(arg, ...) with the entry found after a call
-  // to Seek(key).  May not make such a call if filter policy says
-  // that key is not present.
+  // Calls (*handle_result)(arg, ...) repeatedly, starting with the entry found
+  // after a call to Seek(key), until handle_result returns false.
+  // May not make such a call if filter policy says that key is not present.
   friend class TableCache;
   Status InternalGet(
       const ReadOptions&, const Slice& key,
       void* arg,
-      void (*handle_result)(void* arg, const Slice& k, const Slice& v, bool));
+      bool (*handle_result)(void* arg, const Slice& k, const Slice& v, bool));
 
 
   void ReadMeta(const Footer& footer);

table/table_builder.cc

@@ -14,6 +14,7 @@
 #include "table/format.h"
 #include "util/coding.h"
 #include "util/crc32c.h"
+#include "util/stop_watch.h"
 
 namespace leveldb {
 
@@ -98,6 +99,22 @@ void TableBuilder::Add(const Slice& key, const Slice& value) {
     assert(r->options.comparator->Compare(key, Slice(r->last_key)) > 0);
   }
 
+  const size_t curr_size = r->data_block.CurrentSizeEstimate();
+  const size_t estimated_size_after = r->data_block.EstimateSizeAfterKV(key,
+      value);
+  // Do flush if one of the below two conditions is true:
+  // 1) if the current estimated size already exceeds the block size,
+  // 2) block_size_deviation is set and the estimated size after appending
+  // the kv will exceed the block size and the current size is under the
+  // the deviation.
+  if (curr_size >= r->options.block_size ||
+      (estimated_size_after > r->options.block_size &&
+      r->options.block_size_deviation > 0 &&
+      (curr_size * 100) >
+        r->options.block_size * (100 - r->options.block_size_deviation))) {
+    Flush();
+  }
+
   if (r->pending_index_entry) {
     assert(r->data_block.empty());
     r->options.comparator->FindShortestSeparator(&r->last_key, key);
@@ -114,11 +131,6 @@ void TableBuilder::Add(const Slice& key, const Slice& value) {
   r->last_key.assign(key.data(), key.size());
   r->num_entries++;
   r->data_block.Add(key, value);
-
-  const size_t estimated_block_size = r->data_block.CurrentSizeEstimate();
-  if (estimated_block_size >= r->options.block_size) {
-    Flush();
-  }
 }
 
 void TableBuilder::Flush() {
@@ -220,6 +232,7 @@ void TableBuilder::WriteRawBlock(const Slice& block_contents,
                                  CompressionType type,
                                  BlockHandle* handle) {
   Rep* r = rep_;
+  StopWatch sw(r->options.env, r->options.statistics, WRITE_RAW_BLOCK_MICROS);
   handle->set_offset(r->offset);
   handle->set_size(block_contents.size());
   r->status = r->file->Append(block_contents);

table/table_test.cc

@@ -1,9 +1,6 @@
 // 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 "leveldb/table.h"
-
 #include <map>
 #include <string>
 #include "db/dbformat.h"
@@ -16,6 +13,7 @@
 #include "table/block.h"
 #include "table/block_builder.h"
 #include "table/format.h"
+#include "table/table.h"
 #include "util/random.h"
 #include "util/testharness.h"
 #include "util/testutil.h"
@@ -257,7 +255,7 @@ class TableConstructor: public Constructor {
     // Open the table
     uniq_id_ = cur_uniq_id_++;
     source_.reset(new StringSource(sink_->contents(), uniq_id_));
-    return Table::Open(options, std::move(source_),
+    return Table::Open(options, soptions, std::move(source_),
                        sink_->contents().size(), &table_);
   }
 
@@ -271,7 +269,7 @@ class TableConstructor: public Constructor {
 
   virtual Status Reopen(const Options& options) {
     source_.reset(new StringSource(sink_->contents(), uniq_id_));
-    return Table::Open(options, std::move(source_),
+    return Table::Open(options, soptions, std::move(source_),
                        sink_->contents().size(), &table_);
   }
 
@@ -295,6 +293,7 @@ class TableConstructor: public Constructor {
   TableConstructor();
 
   static uint64_t cur_uniq_id_;
+  const EnvOptions soptions;
 };
 uint64_t TableConstructor::cur_uniq_id_ = 1;
 

table/two_level_iterator.cc

@@ -4,16 +4,18 @@
 
 #include "table/two_level_iterator.h"
 
-#include "leveldb/table.h"
 #include "table/block.h"
 #include "table/format.h"
+#include "table/table.h"
 #include "table/iterator_wrapper.h"
 
 namespace leveldb {
 
 namespace {
 
-typedef Iterator* (*BlockFunction)(void*, const ReadOptions&, const Slice&);
+typedef Iterator* (*BlockFunction)(void*, const ReadOptions&,
+                                   const EnvOptions& soptions, const Slice&,
+                                   bool for_compaction);
 
 class TwoLevelIterator: public Iterator {
  public:
@@ -21,7 +23,9 @@ class TwoLevelIterator: public Iterator {
     Iterator* index_iter,
     BlockFunction block_function,
     void* arg,
-    const ReadOptions& options);
+    const ReadOptions& options,
+    const EnvOptions& soptions,
+    bool for_compaction);
 
   virtual ~TwoLevelIterator();
 
@@ -65,24 +69,30 @@ class TwoLevelIterator: public Iterator {
   BlockFunction block_function_;
   void* arg_;
   const ReadOptions options_;
+  const EnvOptions& soptions_;
   Status status_;
   IteratorWrapper index_iter_;
   IteratorWrapper data_iter_; // May be nullptr
   // If data_iter_ is non-nullptr, then "data_block_handle_" holds the
   // "index_value" passed to block_function_ to create the data_iter_.
   std::string data_block_handle_;
+  bool for_compaction_;
 };
 
 TwoLevelIterator::TwoLevelIterator(
     Iterator* index_iter,
     BlockFunction block_function,
     void* arg,
-    const ReadOptions& options)
+    const ReadOptions& options,
+    const EnvOptions& soptions,
+    bool for_compaction)
     : block_function_(block_function),
       arg_(arg),
       options_(options),
+      soptions_(soptions),
       index_iter_(index_iter),
-      data_iter_(nullptr) {
+      data_iter_(nullptr),
+      for_compaction_(for_compaction) {
 }
 
 TwoLevelIterator::~TwoLevelIterator() {
@@ -163,7 +173,8 @@ void TwoLevelIterator::InitDataBlock() {
       // data_iter_ is already constructed with this iterator, so
       // no need to change anything
     } else {
-      Iterator* iter = (*block_function_)(arg_, options_, handle);
+      Iterator* iter = (*block_function_)(arg_, options_, soptions_, handle,
+                                          for_compaction_);
       data_block_handle_.assign(handle.data(), handle.size());
       SetDataIterator(iter);
     }
@@ -176,8 +187,11 @@ Iterator* NewTwoLevelIterator(
     Iterator* index_iter,
     BlockFunction block_function,
     void* arg,
-    const ReadOptions& options) {
-  return new TwoLevelIterator(index_iter, block_function, arg, options);
+    const ReadOptions& options,
+    const EnvOptions& soptions,
+    bool for_compaction) {
+  return new TwoLevelIterator(index_iter, block_function, arg,
+                              options, soptions, for_compaction);
 }
 
 }  // namespace leveldb

table/two_level_iterator.h

@@ -6,6 +6,7 @@
 #define STORAGE_LEVELDB_TABLE_TWO_LEVEL_ITERATOR_H_
 
 #include "leveldb/iterator.h"
+#include "leveldb/env.h"
 
 namespace leveldb {
 
@@ -25,9 +26,13 @@ extern Iterator* NewTwoLevelIterator(
     Iterator* (*block_function)(
         void* arg,
         const ReadOptions& options,
-        const Slice& index_value),
+        const EnvOptions& soptions,
+        const Slice& index_value,
+        bool for_compaction),
     void* arg,
-    const ReadOptions& options);
+    const ReadOptions& options,
+    const EnvOptions& soptions,
+    bool for_compaction = false);
 
 }  // namespace leveldb
 

tools/db_crashtest.py

@@ -0,0 +1,121 @@
+#! /usr/bin/env python
+import os
+import sys
+import time
+import random
+import getopt
+import logging
+import tempfile
+import subprocess
+
+# This python script runs and kills db_stress multiple times with
+# test-batches-snapshot ON,
+# total operations much less than the total keys, and
+# a high read percentage.
+# This checks consistency in case of unsafe crashes in  Rocksdb
+
+def main(argv):
+    try:
+        opts, args = getopt.getopt(argv, "hd:t:i:o:b:")
+    except getopt.GetoptError:
+        print("db_crashtest.py -d <duration_test> -t <#threads> "
+              "-i <interval for one run> -o <ops_per_thread>\n")
+        sys.exit(2)
+
+    # default values, will be overridden by cmdline args
+    interval = 120  # time for one db_stress instance to run
+    duration = 6000  # total time for this script to test db_stress
+    threads = 32
+    # since we will be killing anyway, use large value for ops_per_thread
+    ops_per_thread = 100000000
+    write_buf_size = 4 * 1024 * 1024
+
+    for opt, arg in opts:
+        if opt == '-h':
+            print("db_crashtest.py -d <duration_test>"
+                  " -t <#threads> -i <interval for one run>"
+                  " -o <ops_per_thread> -b <write_buffer_size>\n")
+            sys.exit()
+        elif opt == ("-d"):
+            duration = int(arg)
+        elif opt == ("-t"):
+            threads = int(arg)
+        elif opt == ("-i"):
+            interval = int(arg)
+        elif opt == ("-o"):
+            ops_per_thread = int(arg)
+        elif opt == ("-b"):
+            write_buf_size = int(arg)
+        else:
+            print("db_crashtest.py -d <duration_test>"
+                  " -t <#threads> -i <interval for one run>"
+                  " -o <ops_per_thread> -b <write_buffer_size>\n")
+            sys.exit(2)
+
+    exit_time = time.time() + duration
+
+    dirpath = tempfile.mkdtemp()
+
+    print("Running blackbox-crash-test with \ninterval_between_crash="
+          + str(interval) + "\ntotal-duration=" + str(duration)
+          + "\nthreads=" + str(threads) + "\nops_per_thread="
+          + str(ops_per_thread) + "\nwrite_buffer_size="
+          + str(write_buf_size) + "\n")
+
+    while time.time() < exit_time:
+        run_had_errors = False
+        additional_opts = ' --disable_seek_compaction=' + \
+                          str(random.randint(0, 1)) + \
+                          ' --mmap_read=' + str(random.randint(0, 1)) + \
+                          ' --block_size=16384 ' + \
+                          ' --cache_size=1048576 ' + \
+                          ' --open_files=500000 ' + \
+                          ' --verify_checksum=1 ' + \
+                          ' --sync=' + str(random.randint(0, 1)) + \
+                          ' --disable_wal=0 ' + \
+                          ' --disable_data_sync=' + \
+                          str(random.randint(0, 1)) + \
+                          ' --target_file_size_base=2097152 ' + \
+                          ' --target_file_size_multiplier=2 ' + \
+                          ' --max_write_buffer_number=3 ' + \
+                          ' --max_background_compactions=20 ' + \
+                          ' --max_bytes_for_level_base=10485760'
+        killtime = time.time() + interval
+        child = subprocess.Popen(['./db_stress \
+                        --test_batches_snapshots=1 \
+                        --ops_per_thread=0' + str(ops_per_thread) + ' \
+                        --threads=0' + str(threads) + ' \
+                        --write_buffer_size=' + str(write_buf_size) + '\
+                        --destroy_db_initially=0 \
+                        --reopen=0 \
+                        --readpercent=50 \
+                        --db=' + dirpath + '\
+                        --max_key=100000000 ' + additional_opts],
+                                 stderr=subprocess.PIPE, shell=True)
+        print("Running db_stress with pid=%d and additional options=\n"
+              % child.pid + additional_opts + "\n")
+        time.sleep(interval)
+        while True:
+            if time.time() > killtime:
+                if child.poll() is not None:
+                    print("WARNING: db_stress ended before kill\n")
+                else:
+                    child.kill()
+                    print("KILLED %d\n" % child.pid)
+                    time.sleep(1)  # time to stabilize after a kill
+
+                while True:
+                    line = child.stderr.readline().strip()
+                    if line != '':
+                        run_had_errors = True
+                        print('***' + line + '^')
+                    else:
+                        break
+                if run_had_errors:
+                    sys.exit(2)
+                break
+
+            time.sleep(1)  # time to stabilize before the next run
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))

tools/db_crashtest2.py

@@ -0,0 +1,148 @@
+#! /usr/bin/env python
+import os
+import sys
+import time
+import random
+import getopt
+import logging
+import tempfile
+import subprocess
+
+# This python script runs db_stress multiple times with kill_random_test
+# that causes leveldb to crash at various points in code.
+# It also has test-batches-snapshot ON so that basic atomic/consistency
+# checks can be performed.
+#
+def main(argv):
+    try:
+        opts, args = getopt.getopt(argv, "hd:t:k:o:b:")
+    except getopt.GetoptError:
+        print str(getopt.GetoptError)
+        print "db_crashtest2.py -d <duration_test> -t <#threads> " \
+            "-k <kills with prob 1/k> -o <ops_per_thread> "\
+            "-b <write_buffer_size>\n"
+        sys.exit(2)
+
+    # default values, will be overridden by cmdline args
+    kill_random_test = 97  # kill with probability 1/97 by default
+    duration = 6000  # total time for this script to test db_stress
+    threads = 32
+    ops_per_thread = 200000
+    write_buf_size = 4 * 1024 * 1024
+
+    for opt, arg in opts:
+        if opt == '-h':
+            print "db_crashtest2.py -d <duration_test> -t <#threads> " \
+                "-k <kills with prob 1/k> -o <ops_per_thread> "\
+                "-b <write_buffer_size>\n"
+            sys.exit()
+        elif opt == ("-d"):
+            duration = int(arg)
+        elif opt == ("-t"):
+            threads = int(arg)
+        elif opt == ("-k"):
+            kill_random_test = int(arg)
+        elif opt == ("-i"):
+            interval = int(arg)
+        elif opt == ("-o"):
+            ops_per_thread = int(arg)
+        elif opt == ("-b"):
+            write_buf_size = int(arg)
+        else:
+            print "unrecognized option " + str(opt) + "\n"
+            print "db_crashtest2.py -d <duration_test> -t <#threads> " \
+                "-k <kills with prob 1/k> -o <ops_per_thread> " \
+                "-b <write_buffer_size>\n"
+            sys.exit(2)
+
+    exit_time = time.time() + duration
+
+    dirpath = tempfile.mkdtemp()
+
+    print("Running whitebox-crash-test with \ntotal-duration=" + str(duration)
+          + "\nthreads=" + str(threads) + "\nops_per_thread="
+          + str(ops_per_thread) + "\nwrite_buffer_size="
+          + str(write_buf_size) + "\n")
+
+    # kill in every alternate run. toggle tracks which run we are doing.
+    toggle = True
+
+    while time.time() < exit_time:
+        run_had_errors = False
+        additional_opts = ' --disable_seek_compaction=' + \
+                          str(random.randint(0, 1)) + \
+                          ' --mmap_read=' + str(random.randint(0, 1)) + \
+                          ' --block_size=16384 ' + \
+                          ' --cache_size=1048576 ' + \
+                          ' --open_files=500000 ' + \
+                          ' --verify_checksum=1 ' + \
+                          ' --sync=' + str(random.randint(0, 1)) + \
+                          ' --disable_wal=0 ' + \
+                          ' --disable_data_sync=' + \
+                          str(random.randint(0, 1)) + \
+                          ' --target_file_size_base=2097152 ' + \
+                          ' --target_file_size_multiplier=2 ' + \
+                          ' --max_write_buffer_number=3 ' + \
+                          ' --max_background_compactions=20 ' + \
+                          ' --max_bytes_for_level_base=10485760'
+        print ("Running db_stress with additional options=\n"
+               + additional_opts + "\n")
+
+        if toggle:
+            # since we are going to kill anyway, use more ops per thread
+            new_ops_per_thread = 100 * ops_per_thread
+            killoption = '--kill_random_test=' + str(kill_random_test)
+        else:
+            new_ops_per_thread = ops_per_thread
+            killoption = ''
+
+        toggle = not toggle
+
+        cmd = ['./db_stress \
+                --test_batches_snapshots=1 \
+                --ops_per_thread=0' + str(new_ops_per_thread) + ' \
+                --threads=0' + str(threads) + ' \
+                --write_buffer_size=' + str(write_buf_size) + ' \
+                --destroy_db_initially=0 ' + killoption + ' \
+                --reopen=0 \
+                --readpercent=50 \
+                --db=' + dirpath + ' \
+                --max_key=100000000 ' + additional_opts]
+
+        popen = subprocess.Popen(cmd, stdout=subprocess.PIPE,
+                                 stderr=subprocess.STDOUT,
+                                 shell=True)
+        stdoutdata, stderrdata = popen.communicate()
+        retncode = popen.returncode
+        msg = ("kill option = {0}, exitcode = {1}".format(
+               killoption, retncode))
+        print msg
+        print stdoutdata
+
+        expected = False
+        if (killoption == '') and (retncode == 0):
+            # we expect zero retncode if no kill option
+            expected = True
+        elif killoption != '' and retncode < 0:
+            # we expect negative retncode if kill option was given
+            expected = True
+
+        if not expected:
+            print "TEST FAILED. See kill option and exit code above!!!\n"
+            sys.exit(1)
+
+        stdoutdata = stdoutdata.lower()
+        errorcount = (stdoutdata.count('error') -
+                      stdoutdata.count('got errors 0 times'))
+        print "#times error occured in output is " + str(errorcount) + "\n"
+
+        if (errorcount > 0):
+            print "TEST FAILED. Output has 'error'!!!\n"
+            sys.exit(2)
+        if (stdoutdata.find('fail') >= 0):
+            print "TEST FAILED. Output has 'fail'!!!\n"
+            sys.exit(2)
+        time.sleep(1)  # time to stabilize after a kill
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))

tools/db_stress.cc

@@ -21,7 +21,7 @@
 #include "db/version_set.h"
 #include "db/db_statistics.h"
 #include "leveldb/cache.h"
-#include "leveldb/db.h"
+#include "utilities/utility_db.h"
 #include "leveldb/env.h"
 #include "leveldb/write_batch.h"
 #include "leveldb/statistics.h"
@@ -31,6 +31,8 @@
 #include "util/mutexlock.h"
 #include "util/random.h"
 #include "util/testutil.h"
+#include "util/logging.h"
+#include "utilities/ttl/db_ttl.h"
 #include "hdfs/env_hdfs.h"
 
 static const long KB = 1024;
@@ -54,6 +56,11 @@ static bool FLAGS_test_batches_snapshots = false;
 // Number of concurrent threads to run.
 static int FLAGS_threads = 32;
 
+// Opens the db with this ttl value if this is not -1
+// Carefully specify a large value such that verifications on deleted
+//  values don't fail
+static int FLAGS_ttl = -1;
+
 // Size of each value will be this number times rand_int(1,3) bytes
 static int FLAGS_value_size_mult = 8;
 
@@ -62,6 +69,9 @@ static bool FLAGS_verify_before_write = false;
 // Print histogram of operation timings
 static bool FLAGS_histogram = false;
 
+// Destroys the database dir before start if this is true
+static bool FLAGS_destroy_db_initially = true;
+
 static bool FLAGS_verbose = false;
 
 // Number of bytes to buffer in memtable before compacting
@@ -100,8 +110,11 @@ static const char* FLAGS_db = nullptr;
 // Verify checksum for every block read from storage
 static bool FLAGS_verify_checksum = false;
 
+// Allow reads to occur via mmap-ing files
+static bool FLAGS_use_mmap_reads = leveldb::EnvOptions().use_mmap_reads;
+
 // Database statistics
-static class leveldb::DBStatistics* dbstats;
+static std::shared_ptr<leveldb::Statistics> dbstats;
 
 // Sync all writes to disk
 static bool FLAGS_sync = false;
@@ -112,6 +125,10 @@ static bool FLAGS_disable_data_sync = false;
 // If true, issue fsync instead of fdatasync
 static bool FLAGS_use_fsync = false;
 
+// If non-zero, kill at various points in source code with probability 1/this
+static int FLAGS_kill_random_test = 0;
+extern int leveldb_kill_odds;
+
 // If true, do not write WAL for write.
 static bool FLAGS_disable_wal = false;
 
@@ -163,9 +180,8 @@ static uint32_t FLAGS_log2_keys_per_lock = 2; // implies 2^2 keys per lock
 // Percentage of times we want to purge redundant keys in memory before flushing
 static uint32_t FLAGS_purge_redundant_percent = 50;
 
-extern bool useOsBuffer;
-extern bool useFsReadAhead;
-extern bool useMmapRead;
+// Level0 compaction start trigger
+static int FLAGS_level0_file_num_compaction_trigger = 0;
 
 namespace leveldb {
 
@@ -232,8 +248,7 @@ class Stats {
       double micros = now - last_op_finish_;
       hist_.Add(micros);
       if (micros > 20000) {
-        fprintf(stderr, "long op: %.1f micros%30s\r", micros, "");
-        fflush(stderr);
+        fprintf(stdout, "long op: %.1f micros%30s\r", micros, "");
       }
       last_op_finish_ = now;
     }
@@ -247,8 +262,7 @@ class Stats {
       else if (next_report_ < 100000) next_report_ += 10000;
       else if (next_report_ < 500000) next_report_ += 50000;
       else                            next_report_ += 100000;
-      fprintf(stderr, "... finished %ld ops%30s\r", done_, "");
-      fflush(stderr);
+      fprintf(stdout, "... finished %ld ops%30s\r", done_, "");
     }
   }
 
@@ -474,14 +488,16 @@ class StressTest {
                        : nullptr),
         db_(nullptr),
         num_times_reopened_(0) {
-    std::vector<std::string> files;
-    FLAGS_env->GetChildren(FLAGS_db, &files);
-    for (unsigned int i = 0; i < files.size(); i++) {
-      if (Slice(files[i]).starts_with("heap-")) {
-        FLAGS_env->DeleteFile(std::string(FLAGS_db) + "/" + files[i]);
+    if (FLAGS_destroy_db_initially) {
+      std::vector<std::string> files;
+      FLAGS_env->GetChildren(FLAGS_db, &files);
+      for (unsigned int i = 0; i < files.size(); i++) {
+        if (Slice(files[i]).starts_with("heap-")) {
+          FLAGS_env->DeleteFile(std::string(FLAGS_db) + "/" + files[i]);
+        }
       }
+      DestroyDB(FLAGS_db, Options());
     }
-    DestroyDB(FLAGS_db, Options());
   }
 
   ~StressTest() {
@@ -688,7 +704,7 @@ class StressTest {
         char expected_prefix = (keys[i])[0];
         char actual_prefix = (values[i])[0];
         if (actual_prefix != expected_prefix) {
-          fprintf(stderr, "expected prefix = %c actual = %c\n",
+          fprintf(stderr, "error expected prefix = %c actual = %c\n",
                   expected_prefix, actual_prefix);
         }
         (values[i])[0] = ' '; // blank out the differing character
@@ -700,7 +716,7 @@ class StressTest {
     // Now that we retrieved all values, check that they all match
     for (int i = 1; i < 10; i++) {
       if (values[i] != values[0]) {
-        fprintf(stderr, "inconsistent values for key %s: %s, %s\n",
+        fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
                 key.ToString().c_str(), values[0].c_str(),
                 values[i].c_str());
       // we continue after error rather than exiting so that we can
@@ -859,6 +875,7 @@ class StressTest {
     for (size_t i=sizeof(uint32_t); i < value_sz; i++) {
       v[i] = (char)(rand ^ i);
     }
+    v[value_sz] = '\0';
     return value_sz; // the size of the value set.
   }
 
@@ -867,7 +884,13 @@ class StressTest {
             kMajorVersion, kMinorVersion);
     fprintf(stdout, "Number of threads   : %d\n", FLAGS_threads);
     fprintf(stdout, "Ops per thread      : %d\n", FLAGS_ops_per_thread);
+    std::string ttl_state("unused");
+    if (FLAGS_ttl > 0) {
+      ttl_state = NumberToString(FLAGS_ttl);
+    }
+    fprintf(stdout, "Time to live(sec)   : %s\n", ttl_state.c_str());
     fprintf(stdout, "Read percentage     : %d\n", FLAGS_readpercent);
+    fprintf(stdout, "Write-buffer-size   : %d\n", FLAGS_write_buffer_size);
     fprintf(stdout, "Delete percentage   : %d\n", FLAGS_delpercent);
     fprintf(stdout, "Max key             : %ld\n", FLAGS_max_key);
     fprintf(stdout, "Ratio #ops/#keys    : %ld\n",
@@ -914,6 +937,8 @@ class StressTest {
     options.env = FLAGS_env;
     options.disableDataSync = FLAGS_disable_data_sync;
     options.use_fsync = FLAGS_use_fsync;
+    options.allow_mmap_reads = FLAGS_use_mmap_reads;
+    leveldb_kill_odds = FLAGS_kill_random_test;
     options.target_file_size_base = FLAGS_target_file_size_base;
     options.target_file_size_multiplier = FLAGS_target_file_size_multiplier;
     options.max_bytes_for_level_base = FLAGS_max_bytes_for_level_base;
@@ -922,6 +947,8 @@ class StressTest {
     options.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger;
     options.level0_slowdown_writes_trigger =
       FLAGS_level0_slowdown_writes_trigger;
+    options.level0_file_num_compaction_trigger =
+      FLAGS_level0_file_num_compaction_trigger;
     options.compression = FLAGS_compression_type;
     options.create_if_missing = true;
     options.disable_seek_compaction = FLAGS_disable_seek_compaction;
@@ -932,7 +959,15 @@ class StressTest {
     if (purge_percent.Uniform(100) < FLAGS_purge_redundant_percent - 1) {
       options.purge_redundant_kvs_while_flush = false;
     }
-    Status s = DB::Open(options, FLAGS_db, &db_);
+
+    fprintf(stdout, "DB path: [%s]\n", FLAGS_db);
+
+    Status s;
+    if (FLAGS_ttl == -1) {
+      s = DB::Open(options, FLAGS_db, &db_);
+    } else {
+      s = UtilityDB::OpenTtlDB(options, FLAGS_db, &db_, FLAGS_ttl);
+    }
     if (!s.ok()) {
       fprintf(stderr, "open error: %s\n", s.ToString().c_str());
       exit(1);
@@ -942,7 +977,11 @@ class StressTest {
   void Reopen() {
     // do not close the db. Just delete the lock file. This
     // simulates a crash-recovery kind of situation.
-    ((DBImpl*) db_)->TEST_Destroy_DBImpl();
+    if (FLAGS_ttl != -1) {
+      ((DBWithTTL*) db_)->TEST_Destroy_DBWithTtl();
+    } else {
+      ((DBImpl*) db_)->TEST_Destroy_DBImpl();
+    }
     db_ = nullptr;
 
     num_times_reopened_++;
@@ -977,6 +1016,12 @@ int main(int argc, char** argv) {
   FLAGS_open_files = leveldb::Options().max_open_files;
   FLAGS_max_background_compactions =
     leveldb::Options().max_background_compactions;
+  FLAGS_level0_file_num_compaction_trigger =
+    leveldb::Options().level0_file_num_compaction_trigger;
+  FLAGS_level0_slowdown_writes_trigger =
+    leveldb::Options().level0_slowdown_writes_trigger;
+  FLAGS_level0_stop_writes_trigger =
+    leveldb::Options().level0_stop_writes_trigger;
   // Compression test code above refers to FLAGS_block_size
   FLAGS_block_size = leveldb::Options().block_size;
   std::string default_db_path;
@@ -1002,6 +1047,9 @@ int main(int argc, char** argv) {
     } else if (sscanf(argv[i], "--histogram=%d%c", &n, &junk) == 1 &&
                (n == 0 || n == 1)) {
       FLAGS_histogram = n;
+    } else if (sscanf(argv[i], "--destroy_db_initially=%d%c", &n, &junk) == 1 &&
+               (n == 0 || n == 1)) {
+      FLAGS_destroy_db_initially = n;
     } else if (sscanf(argv[i], "--verify_before_write=%d%c", &n, &junk) == 1 &&
                (n == 0 || n == 1)) {
       FLAGS_verify_before_write = n;
@@ -1010,6 +1058,8 @@ int main(int argc, char** argv) {
       FLAGS_test_batches_snapshots = n;
     } else if (sscanf(argv[i], "--threads=%d%c", &n, &junk) == 1) {
       FLAGS_threads = n;
+    } else if (sscanf(argv[i], "--ttl=%d%c", &n, &junk) == 1) {
+      FLAGS_ttl = n;
     } else if (sscanf(argv[i], "--value_size_mult=%d%c", &n, &junk) == 1) {
       FLAGS_value_size_mult = n;
     } else if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) {
@@ -1033,19 +1083,13 @@ int main(int argc, char** argv) {
     } else if (sscanf(argv[i], "--verify_checksum=%d%c", &n, &junk) == 1 &&
                (n == 0 || n == 1)) {
       FLAGS_verify_checksum = n;
-    } else if (sscanf(argv[i], "--bufferedio=%d%c", &n, &junk) == 1 &&
-               (n == 0 || n == 1)) {
-      useOsBuffer = n;
     } else if (sscanf(argv[i], "--mmap_read=%d%c", &n, &junk) == 1 &&
                (n == 0 || n == 1)) {
-      useMmapRead = n;
-    } else if (sscanf(argv[i], "--readhead=%d%c", &n, &junk) == 1 &&
-               (n == 0 || n == 1)) {
-      useFsReadAhead = n;
+      FLAGS_use_mmap_reads = n;
     } else if (sscanf(argv[i], "--statistics=%d%c", &n, &junk) == 1 &&
                (n == 0 || n == 1)) {
       if (n == 1) {
-        dbstats =  new leveldb::DBStatistics();
+        dbstats = leveldb::CreateDBStatistics();
       }
     } else if (sscanf(argv[i], "--sync=%d%c", &n, &junk) == 1 &&
                (n == 0 || n == 1)) {
@@ -1062,6 +1106,9 @@ int main(int argc, char** argv) {
     } else if (sscanf(argv[i], "--use_fsync=%d%c", &n, &junk) == 1 &&
         (n == 0 || n == 1)) {
       FLAGS_use_fsync = n;
+    } else if (sscanf(argv[i], "--kill_random_test=%d%c", &n, &junk) == 1 &&
+               (n >= 0)) {
+      FLAGS_kill_random_test = n;
     } else if (sscanf(argv[i], "--disable_wal=%d%c", &n, &junk) == 1 &&
         (n == 0 || n == 1)) {
       FLAGS_disable_wal = n;
@@ -1085,6 +1132,9 @@ int main(int argc, char** argv) {
     } else if (sscanf(argv[i],"--level0_slowdown_writes_trigger=%d%c",
         &n, &junk) == 1) {
       FLAGS_level0_slowdown_writes_trigger = n;
+    } else if (sscanf(argv[i],"--level0_file_num_compaction_trigger=%d%c",
+        &n, &junk) == 1) {
+      FLAGS_level0_file_num_compaction_trigger = n;
     } else if (strncmp(argv[i], "--compression_type=", 19) == 0) {
       const char* ctype = argv[i] + 19;
       if (!strcasecmp(ctype, "none"))
@@ -1121,6 +1171,16 @@ int main(int argc, char** argv) {
       fprintf(stderr, "Error: Read + Delete percents > 100!\n");
       exit(1);
   }
+  if (FLAGS_disable_wal == 1 && FLAGS_reopen > 0) {
+      fprintf(stderr, "Error: Db cannot reopen safely with disable_wal set!\n");
+      exit(1);
+  }
+  if ((unsigned)FLAGS_reopen >= FLAGS_ops_per_thread) {
+      fprintf(stderr, "Error: #DB-reopens should be < ops_per_thread\n"
+        "Provided reopens = %d and ops_per_thread = %u\n", FLAGS_reopen,
+        FLAGS_ops_per_thread);
+      exit(1);
+  }
 
   // Choose a location for the test database if none given with --db=<path>
   if (FLAGS_db == nullptr) {
@@ -1131,8 +1191,5 @@ int main(int argc, char** argv) {
 
   leveldb::StressTest stress;
   stress.Run();
-  if (dbstats) {
-    delete dbstats;
-  }
   return 0;
 }

tools/ldb.cc

@@ -2,93 +2,10 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#include "util/ldb_cmd.h"
-
-namespace leveldb {
-
-class LDBCommandRunner {
-public:
-
-  static void PrintHelp(const char* exec_name) {
-    string ret;
-
-    ret.append("ldb - LevelDB Tool");
-    ret.append("\n\n");
-    ret.append("All commands MUST specify --" + LDBCommand::ARG_DB +
-        "=<full_path_to_db_directory>\n");
-    ret.append("\n");
-    ret.append("The following optional parameters control if keys/values are "
-        "input/output as hex or as plain strings:\n");
-    ret.append("  --" + LDBCommand::ARG_KEY_HEX +
-        " : Keys are input/output as hex\n");
-    ret.append("  --" + LDBCommand::ARG_VALUE_HEX +
-        " : Values are input/output as hex\n");
-    ret.append("  --" + LDBCommand::ARG_HEX +
-        " : Both keys and values are input/output as hex\n");
-    ret.append("\n");
-
-    ret.append("The following optional parameters control the database "
-        "internals:\n");
-    ret.append("  --" + LDBCommand::ARG_BLOOM_BITS + "=<int,e.g.:14>\n");
-    ret.append("  --" + LDBCommand::ARG_COMPRESSION_TYPE +
-        "=<no|snappy|zlib|bzip2>\n");
-    ret.append("  --" + LDBCommand::ARG_BLOCK_SIZE +
-        "=<block_size_in_bytes>\n");
-    ret.append("  --" + LDBCommand::ARG_AUTO_COMPACTION + "=<true|false>\n");
-    ret.append("  --" + LDBCommand::ARG_WRITE_BUFFER_SIZE +
-        "=<int,e.g.:4194304>\n");
-    ret.append("  --" + LDBCommand::ARG_FILE_SIZE + "=<int,e.g.:2097152>\n");
-
-    ret.append("\n\n");
-    ret.append("Data Access Commands:\n");
-    PutCommand::Help(ret);
-    GetCommand::Help(ret);
-    BatchPutCommand::Help(ret);
-    ScanCommand::Help(ret);
-    DeleteCommand::Help(ret);
-    DBQuerierCommand::Help(ret);
-    ApproxSizeCommand::Help(ret);
-
-    ret.append("\n\n");
-    ret.append("Admin Commands:\n");
-    WALDumperCommand::Help(ret);
-    CompactorCommand::Help(ret);
-    ReduceDBLevelsCommand::Help(ret);
-    DBDumperCommand::Help(ret);
-    DBLoaderCommand::Help(ret);
-
-    fprintf(stderr, "%s\n", ret.c_str());
-  }
-
-  static void RunCommand(int argc, char** argv) {
-    if (argc <= 2) {
-      PrintHelp(argv[0]);
-      exit(1);
-    }
-
-    LDBCommand* cmdObj = LDBCommand::InitFromCmdLineArgs(argc, argv);
-    if (cmdObj == NULL) {
-      fprintf(stderr, "Unknown command\n");
-      PrintHelp(argv[0]);
-      exit(1);
-    }
-
-    if (!cmdObj->ValidateCmdLineOptions()) {
-      exit(1);
-    }
-
-    cmdObj->Run();
-    LDBCommandExecuteResult ret = cmdObj->GetExecuteState();
-    fprintf(stderr, "%s\n", ret.ToString().c_str());
-    delete cmdObj;
-
-    exit(ret.IsFailed());
-  }
-
-};
-
-}
+#include "leveldb/ldb_tool.h"
 
 int main(int argc, char** argv) {
-  leveldb::LDBCommandRunner::RunCommand(argc, argv);
+  leveldb::LDBTool tool;
+  tool.Run(argc, argv);
+  return 0;
 }

tools/ldb_test.py

@@ -43,7 +43,7 @@ class LDBTestCase(unittest.TestCase):
     def dbParam(self, dbName):
         return "--db=%s" % os.path.join(self.TMP_DIR, dbName)
 
-    def assertRunOKFull(self, params, expectedOutput):
+    def assertRunOKFull(self, params, expectedOutput, unexpected=False):
         """
         All command-line params must be specified.
         Allows full flexibility in testing; for example: missing db param.
@@ -52,7 +52,10 @@ class LDBTestCase(unittest.TestCase):
 
         output = my_check_output("./ldb %s |grep -v \"Created bg thread\"" %
                             params, shell=True)
-        self.assertEquals(output.strip(), expectedOutput.strip());
+        if not unexpected:
+            self.assertEqual(output.strip(), expectedOutput.strip())
+        else:
+            self.assertNotEqual(output.strip(), expectedOutput.strip())
 
     def assertRunFAILFull(self, params):
         """
@@ -70,13 +73,13 @@ class LDBTestCase(unittest.TestCase):
             "Exception should have been raised for command with params: %s" %
             params)
 
-    def assertRunOK(self, params, expectedOutput):
+    def assertRunOK(self, params, expectedOutput, unexpected=False):
         """
         Uses the default test db.
 
         """
         self.assertRunOKFull("%s %s" % (self.dbParam(self.DB_NAME), params),
-                expectedOutput)
+                             expectedOutput, unexpected)
 
     def assertRunFAIL(self, params):
         """
@@ -123,7 +126,7 @@ class LDBTestCase(unittest.TestCase):
         self.assertRunOK("scan", "x2 : y2\nx3 : y3")
 
         self.assertRunOK("delete NonExistentKey", "OK")
-        # It is wierd that GET and SCAN raise exception for
+        # It is weird that GET and SCAN raise exception for
         # non-existent key, while delete does not
 
     def dumpDb(self, params, dumpFile):
@@ -142,6 +145,16 @@ class LDBTestCase(unittest.TestCase):
         self.assertRunFAIL("batchput k1")
         self.assertRunFAIL("batchput k1 v1 k2")
 
+    def testInvalidCmdLines(self):
+        print "Running testInvalidCmdLines..."
+        # db not specified
+        self.assertRunFAILFull("put 0x6133 0x6233 --hex --create_if_missing")
+        # No param called he
+        self.assertRunFAIL("put 0x6133 0x6233 --he --create_if_missing")
+        # max_keys is not applicable for put
+        self.assertRunFAIL("put 0x6133 0x6233 --max_keys=1 --create_if_missing")
+        # hex has invalid boolean value
+
     def testHexPutGet(self):
         print "Running testHexPutGet..."
         self.assertRunOK("put a1 b1 --create_if_missing", "OK")
@@ -171,6 +184,18 @@ class LDBTestCase(unittest.TestCase):
         self.assertRunOK("delete --hex 0x6133", "OK")
         self.assertRunOK("scan", "a1 : b1\na2 : b2\na4 : b4")
 
+    def testTtlPutGet(self):
+        print "Running testTtlPutGet..."
+        self.assertRunOK("put a1 b1 --ttl --create_if_missing", "OK")
+        self.assertRunOK("scan ", "a1 : b1", True)
+        self.assertRunOK("dump --ttl ", "a1 ==> b1", True)
+        self.assertRunOK("scan --hex --ttl", "0x6131 : 0x6231")
+        self.assertRunOK("get a1", "b1", True)
+        self.assertRunOK("get --ttl a1", "b1")
+        self.assertRunOK("put a3 b3 --create_if_missing", "OK")
+        # fails because timstamp's length is greater than value's
+        self.assertRunFAIL("get --ttl a3")
+
     def testInvalidCmdLines(self):
         print "Running testInvalidCmdLines..."
         # db not specified

tools/manifest_dump.cc

@@ -1,79 +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.
-
-#include "db/version_set.h"
-
-#include <algorithm>
-#include <stdio.h>
-#include "db/filename.h"
-#include "db/log_reader.h"
-#include "db/log_writer.h"
-#include "db/memtable.h"
-#include "db/table_cache.h"
-#include "leveldb/env.h"
-#include "leveldb/table_builder.h"
-#include "table/merger.h"
-#include "table/two_level_iterator.h"
-#include "util/coding.h"
-#include "util/logging.h"
-
-static int verbose = 0;
-static int output_hex = 0;
-
-using namespace leveldb;
-
-//
-// Takes a manifest file and dumps out all metedata
-//
-int main(int argc, char** argv) {
-
-  // parse command line options
-  size_t n;
-  char junk;
-  int foundfile = 0;
-  std::string manifestfile;
-  for (int i = 1; i < argc; i++) {
-    std::string param(argv[i]);
-    if ((n = param.find("--file=")) != std::string::npos) {
-      manifestfile = param.substr(strlen("--file="));
-      foundfile = 1;
-    } else if (sscanf(argv[i], "--verbose=%ld%c", &n, &junk) == 1 &&
-        (n == 0 || n == 1)) {
-      verbose = n;
-    } else if (param == "--output_hex") {
-      output_hex = n;
-    } else {
-      fprintf(stderr, "Unknown or badly-formatted option: %s\n",
-              argv[i]);
-      abort();
-    }
-  }
-  if (!foundfile) {
-    fprintf(stderr,
-      "%s [--verbose=0|1] [--output_hex] [--file=pathname of manifest file]\n",
-      argv[0]);
-    abort();
-  }
-
-  if (verbose) {
-    printf("Processing Manifest file %s\n", manifestfile.c_str());
-  }
-
-  Options options;
-  std::string file(manifestfile);
-  std::string dbname("dummy");
-  TableCache* tc = new TableCache(dbname, &options, 10);
-  const InternalKeyComparator* cmp = new InternalKeyComparator(options.comparator);
-
-  VersionSet* versions = new VersionSet(dbname, &options,
-                                   tc, cmp);
-  Status s = versions->DumpManifest(options, file, verbose, output_hex);
-  if (!s.ok()) {
-    printf("Error in processing file %s %s\n", manifestfile.c_str(),
-           s.ToString().c_str());
-  }
-  if (verbose) {
-    printf("Processing Manifest file %s done\n", manifestfile.c_str());
-  }
-}

tools/reduce_levels_test.cc

@@ -18,7 +18,7 @@ public:
   ReduceLevelTest() {
     dbname_ = test::TmpDir() + "/db_reduce_levels_test";
     DestroyDB(dbname_, Options());
-    db_ = NULL;
+    db_ = nullptr;
   }
 
   Status OpenDB(bool create_if_missing, int levels,
@@ -41,7 +41,7 @@ public:
   }
 
   Status CompactMemTable() {
-    if (db_ == NULL) {
+    if (db_ == nullptr) {
       return Status::InvalidArgument("DB not opened.");
     }
     DBImpl* db_impl = reinterpret_cast<DBImpl*>(db_);
@@ -49,9 +49,9 @@ public:
   }
 
   void CloseDB() {
-    if (db_ != NULL) {
+    if (db_ != nullptr) {
       delete db_;
-      db_ = NULL;
+      db_ = nullptr;
     }
   }
 

tools/sst_dump.cc

@@ -1,4 +1,4 @@
-#include "leveldb/table.h"
+#include "table/table.h"
 
 #include <map>
 #include <string>
@@ -22,7 +22,7 @@ namespace leveldb {
 
 class SstFileReader {
 public:
-  SstFileReader(std::string file_name,
+  explicit SstFileReader(std::string file_name,
                 bool verify_checksum = false,
                 bool output_hex = false);
   Status ReadSequential(bool print_kv, uint64_t read_num = -1);
@@ -34,6 +34,7 @@ private:
   uint64_t read_num_;
   bool verify_checksum_;
   bool output_hex_;
+  EnvOptions soptions_;
 };
 
 SstFileReader::SstFileReader(std::string file_path,
@@ -48,13 +49,14 @@ Status SstFileReader::ReadSequential(bool print_kv, uint64_t read_num)
   unique_ptr<Table> table;
   Options table_options;
   unique_ptr<RandomAccessFile> file;
-  Status s = table_options.env->NewRandomAccessFile(file_name_, &file);
+  Status s = table_options.env->NewRandomAccessFile(file_name_, &file,
+                                                    soptions_);
   if(!s.ok()) {
    return s;
   }
   uint64_t file_size;
   table_options.env->GetFileSize(file_name_, &file_size);
-  s = Table::Open(table_options, std::move(file), file_size, &table);
+  s = Table::Open(table_options, soptions_, std::move(file), file_size, &table);
   if(!s.ok()) {
    return s;
   }
@@ -93,7 +95,7 @@ static void print_help() {
 
 int main(int argc, char** argv) {
 
-  const char* dir_or_file = NULL;
+  const char* dir_or_file = nullptr;
   uint64_t read_num = -1;
   std::string command;
 
@@ -119,7 +121,7 @@ int main(int argc, char** argv) {
     }
   }
 
-  if(dir_or_file == NULL) {
+  if(dir_or_file == nullptr) {
     print_help();
     exit(1);
   }

util/auto_roll_logger.cc

@@ -1,4 +1,5 @@
 #include "util/auto_roll_logger.h"
+#include "util/mutexlock.h"
 
 using namespace std;
 
@@ -35,17 +36,26 @@ void AutoRollLogger::RollLogFile() {
 void AutoRollLogger::Logv(const char* format, va_list ap) {
   assert(GetStatus().ok());
 
-  if (kLogFileTimeToRoll > 0 && LogExpired()) {
-    RollLogFile();
-    ResetLogger();
+  std::shared_ptr<Logger> logger;
+  {
+    MutexLock l(&mutex_);
+    if ((kLogFileTimeToRoll > 0 && LogExpired()) ||
+        (kMaxLogFileSize > 0 && logger_->GetLogFileSize() >= kMaxLogFileSize)) {
+      RollLogFile();
+      ResetLogger();
+    }
+
+    // pin down the current logger_ instance before releasing the mutex.
+    logger = logger_;
   }
 
-  logger_->Logv(format, ap);
-
-  if (kMaxLogFileSize > 0 && logger_->GetLogFileSize() > kMaxLogFileSize) {
-    RollLogFile();
-    ResetLogger();
-  }
+  // Another thread could have put a new Logger instance into logger_ by now.
+  // However, since logger is still hanging on to the previous instance
+  // (reference count is not zero), we don't have to worry about it being
+  // deleted while we are accessing it.
+  // Note that logv itself is not mutex protected to allow maximum concurrency,
+  // as thread safety should have been handled by the underlying logger.
+  logger->Logv(format, ap);
 }
 
 bool AutoRollLogger::LogExpired() {

util/auto_roll_logger.h

@@ -8,8 +8,9 @@
 #ifndef STORAGE_LEVELDB_UTIL_AUTO_ROLL_LOGGER_H
 #define STORAGE_LEVELDB_UTIL_AUTO_ROLL_LOGGER_H
 
-#include "util/posix_logger.h"
 #include "db/filename.h"
+#include "port/port.h"
+#include "util/posix_logger.h"
 
 namespace leveldb {
 
@@ -29,7 +30,8 @@ class AutoRollLogger : public Logger {
      cached_now(static_cast<uint64_t>(env_->NowMicros() * 1e-6)),
      ctime_(cached_now),
      cached_now_access_count(0),
-     call_NowMicros_every_N_records_(100) {
+     call_NowMicros_every_N_records_(100),
+     mutex_() {
     env->GetAbsolutePath(dbname, &db_absolute_path_);
     log_fname_ = InfoLogFileName(dbname_, db_absolute_path_, db_log_dir_);
     RollLogFile();
@@ -75,6 +77,7 @@ class AutoRollLogger : public Logger {
   uint64_t ctime_;
   uint64_t cached_now_access_count;
   uint64_t call_NowMicros_every_N_records_;
+  port::Mutex mutex_;
 };
 
 // Facade to craete logger automatically

util/auto_roll_logger_test.cc

@@ -78,7 +78,11 @@ void AutoRollLoggerTest::RollLogFileBySizeTest(AutoRollLogger* logger,
 
   // Now the log file will be rolled
   LogMessage(logger, log_message.c_str());
-  ASSERT_TRUE(0 == logger->GetLogFileSize());
+  // Since rotation is checked before actual logging, we need to
+  // trigger the rotation by logging another message.
+  LogMessage(logger, log_message.c_str());
+
+  ASSERT_TRUE(message_size == logger->GetLogFileSize());
 }
 
 uint64_t AutoRollLoggerTest::RollLogFileByTimeTest(

util/bit_set.h

@@ -0,0 +1,75 @@
+/**
+ * Copyright 2009 Facebook
+ * @author Tudor Bosman (tudorb@facebook.com)
+ */
+
+#ifndef STORAGE_LEVELDB_UTIL_BIT_SET_H_
+#define STORAGE_LEVELDB_UTIL_BIT_SET_H_
+
+#include <cassert>
+
+namespace leveldb {
+
+class BitSet {
+ public:
+  /**
+   * Create a bit set of numBits, with the bits set to either true or false.
+   */
+  explicit BitSet(size_t numBits, bool initial=false)
+    : numBits_(numBits),
+      data_(numWords(), initial ? ~0UL : 0UL) {
+  }
+
+  /**
+   * Set bit b to 1.
+   */
+  void set(size_t b) {
+    assert(b >= 0 && b < numBits_);
+    data_[word(b)] |= wordOffsetMask(b);
+  }
+
+  /**
+   * Set bit b to 0;
+   */
+  void reset(size_t b) {
+    assert(b >= 0 && b < numBits_);
+    data_[word(b)] &= ~wordOffsetMask(b);
+  }
+
+  /**
+   * Get a bit.
+   */
+  bool test(int b) const {
+    return data_[word(b)] & wordOffsetMask(b);
+  }
+
+ /**
+   * Return the size of the BitSet, in bits.
+   */
+  size_t size() const {
+    return numBits_;
+  }
+
+ private:
+
+ inline size_t numWords() const {
+    if (numBits_ == 0) return 0;
+    return 1 + (numBits_-1) / (8*sizeof(unsigned long));
+  }
+  inline static size_t word(int b) {
+    return b / (8*sizeof(unsigned long));
+  }
+  inline static int wordOffset(int b) {
+    return b % (8*sizeof(unsigned long));
+  }
+  inline static unsigned long wordOffsetMask(int b) {
+    return 1UL << wordOffset(b);
+  }
+
+  size_t numBits_;
+  std::vector<unsigned long> data_;
+};
+
+}  // namespace facebook
+
+#endif // STORAGE_LEVELDB_UTIL_BIT_SET_H_

util/cache.cc

@@ -189,7 +189,7 @@ void LRUCache::Unref(LRUHandle* e) {
   assert(e->refs > 0);
   e->refs--;
   if (e->refs <= 0) {
-    usage_ -= e->charge;
+
     (*e->deleter)(e->key(), e->value);
     free(e);
   }
@@ -198,6 +198,7 @@ void LRUCache::Unref(LRUHandle* e) {
 void LRUCache::LRU_Remove(LRUHandle* e) {
   e->next->prev = e->prev;
   e->prev->next = e->next;
+  usage_ -= e->charge;
 }
 
 void LRUCache::LRU_Append(LRUHandle* e) {
@@ -206,6 +207,7 @@ void LRUCache::LRU_Append(LRUHandle* e) {
   e->prev = lru_.prev;
   e->prev->next = e;
   e->next->prev = e;
+  usage_ += e->charge;
 }
 
 Cache::Handle* LRUCache::Lookup(const Slice& key, uint32_t hash) {
@@ -239,7 +241,6 @@ Cache::Handle* LRUCache::Insert(
   e->refs = 2;  // One from LRUCache, one for the returned handle
   memcpy(e->key_data, key.data(), key.size());
   LRU_Append(e);
-  usage_ += charge;
 
   LRUHandle* old = table_.Insert(e);
   if (old != nullptr) {
@@ -281,7 +282,8 @@ class ShardedLRUCache : public Cache {
   }
 
   uint32_t Shard(uint32_t hash) {
-    return hash >> (32 - numShardBits);
+    // Note, hash >> 32 yields hash in gcc, not the zero we expect!
+    return (numShardBits > 0) ? (hash >> (32 - numShardBits)) : 0;
   }
 
   void init(size_t capacity, int numbits) {

util/cache_test.cc

@@ -5,6 +5,8 @@
 #include "leveldb/cache.h"
 
 #include <vector>
+#include <string>
+#include <iostream>
 #include "util/coding.h"
 #include "util/testharness.h"
 
@@ -178,6 +180,51 @@ TEST(CacheTest, NewId) {
   ASSERT_NE(a, b);
 }
 
+
+class Value {
+ private:
+  int v_;
+ public:
+  Value(int v) : v_(v) { }
+
+  ~Value() { std::cout << v_ << " is destructed\n"; }
+};
+
+void deleter(const Slice& key, void* value) {
+  delete (Value *)value;
+}
+
+
+TEST(CacheTest, BadEviction) {
+  int n = 10;
+
+  // a LRUCache with n entries and one shard only
+  std::shared_ptr<Cache> cache = NewLRUCache(n, 0);
+
+  std::vector<Cache::Handle*> handles(n+1);
+
+  // Insert n+1 entries, but not releasing.
+  for (int i = 0; i < n+1; i++) {
+    std::string key = std::to_string(i+1);
+    handles[i] = cache->Insert(key, new Value(i+1), 1, &deleter);
+  }
+
+  // Guess what's in the cache now?
+  for (int i = 0; i < n+1; i++) {
+    std::string key = std::to_string(i+1);
+    auto h = cache->Lookup(key);
+    std::cout << key << (h?" found\n":" not found\n");
+    // Only the first entry should be missing
+    ASSERT_TRUE(h || i == 0);
+    if (h) cache->Release(h);
+  }
+
+  for (int i = 0; i < n+1; i++) {
+    cache->Release(handles[i]);
+  }
+  std::cout << "Poor entries\n";
+}
+
 }  // namespace leveldb
 
 int main(int argc, char** argv) {

util/env.cc

@@ -3,6 +3,7 @@
 // found in the LICENSE file. See the AUTHORS file for names of contributors.
 
 #include "leveldb/env.h"
+#include "leveldb/options.h"
 
 namespace leveldb {
 
@@ -46,7 +47,8 @@ static Status DoWriteStringToFile(Env* env, const Slice& data,
                                   const std::string& fname,
                                   bool should_sync) {
   unique_ptr<WritableFile> file;
-  Status s = env->NewWritableFile(fname, &file);
+  EnvOptions soptions;
+  Status s = env->NewWritableFile(fname, &file, soptions);
   if (!s.ok()) {
     return s;
   }
@@ -71,9 +73,10 @@ Status WriteStringToFileSync(Env* env, const Slice& data,
 }
 
 Status ReadFileToString(Env* env, const std::string& fname, std::string* data) {
+  EnvOptions soptions;
   data->clear();
   unique_ptr<SequentialFile> file;
-  Status s = env->NewSequentialFile(fname, &file);
+  Status s = env->NewSequentialFile(fname, &file, soptions);
   if (!s.ok()) {
     return s;
   }
@@ -97,4 +100,26 @@ Status ReadFileToString(Env* env, const std::string& fname, std::string* data) {
 EnvWrapper::~EnvWrapper() {
 }
 
+namespace {  // anonymous namespace
+
+void AssignEnvOptions(EnvOptions* env_options, const Options& options) {
+  env_options->use_os_buffer = options.allow_os_buffer;
+  env_options->use_mmap_reads = options.allow_mmap_reads;
+  env_options->use_mmap_writes = options.allow_mmap_writes;
+  env_options->set_fd_cloexec = options.is_fd_close_on_exec;
+  env_options->bytes_per_sync = options.bytes_per_sync;
+}
+
+}
+
+EnvOptions::EnvOptions(const Options& options) {
+  AssignEnvOptions(this, options);
+}
+
+EnvOptions::EnvOptions() {
+  Options options;
+  AssignEnvOptions(this, options);
+}
+
+
 }  // namespace leveldb

util/env_hdfs.cc

@@ -516,7 +516,8 @@ Status HdfsEnv::NewLogger(const std::string& fname,
 #include "hdfs/env_hdfs.h"
 namespace leveldb {
  Status HdfsEnv::NewSequentialFile(const std::string& fname,
-                                   unique_ptr<SequentialFile>* result) {
+                                   unique_ptr<SequentialFile>* result,
+                                   const EnvOptions& options) {
    return Status::NotSupported("Not compiled with hdfs support");
  }
 }

util/env_posix.cc

@@ -22,6 +22,7 @@
 #include <unistd.h>
 #if defined(OS_LINUX)
 #include <linux/fs.h>
+#include <fcntl.h>
 #endif
 #if defined(LEVELDB_PLATFORM_ANDROID)
 #include <sys/stat.h>
@@ -32,6 +33,8 @@
 #include "util/coding.h"
 #include "util/logging.h"
 #include "util/posix_logger.h"
+#include "util/random.h"
+#include <signal.h>
 
 #if !defined(TMPFS_MAGIC)
 #define TMPFS_MAGIC 0x01021994
@@ -43,13 +46,13 @@
 #define EXT4_SUPER_MAGIC 0xEF53
 #endif
 
-bool useOsBuffer = 1;     // cache data in OS buffers
-bool useFsReadAhead = 1;  // allow filesystem to do readaheads
-bool useMmapRead = 0;     // do not use mmaps for reading files
-bool useMmapWrite = 1;    // use mmaps for appending to files
+// This is only set from db_stress.cc and for testing only.
+// If non-zero, kill at various points in source code with probability 1/this
+int leveldb_kill_odds = 0;
 
 namespace leveldb {
 
+
 namespace {
 
 // list of pathnames that are locked
@@ -60,14 +63,52 @@ static Status IOError(const std::string& context, int err_number) {
   return Status::IOError(context, strerror(err_number));
 }
 
+#ifdef NDEBUG
+// empty in release build
+#define TEST_KILL_RANDOM(leveldb_kill_odds)
+#else
+
+// Kill the process with probablity 1/odds for testing.
+static void TestKillRandom(int odds, const std::string& srcfile,
+                           int srcline) {
+  time_t curtime = time(nullptr);
+  Random r((uint32_t)curtime);
+
+  assert(odds > 0);
+  bool crash = r.OneIn(odds);
+  if (crash) {
+    fprintf(stdout, "Crashing at %s:%d\n", srcfile.c_str(), srcline);
+    fflush(stdout);
+    kill(getpid(), SIGTERM);
+  }
+}
+
+// To avoid crashing always at some frequently executed codepaths (during
+// kill random test), use this factor to reduce odds
+#define REDUCE_ODDS 2
+#define REDUCE_ODDS2 4
+
+#define TEST_KILL_RANDOM(leveldb_kill_odds) {   \
+  if (leveldb_kill_odds > 0) { \
+    TestKillRandom(leveldb_kill_odds, __FILE__, __LINE__);     \
+  } \
+}
+
+#endif
+
 class PosixSequentialFile: public SequentialFile {
  private:
   std::string filename_;
   FILE* file_;
+  int fd_;
+  bool use_os_buffer_;
 
  public:
-  PosixSequentialFile(const std::string& fname, FILE* f)
-      : filename_(fname), file_(f) { }
+  PosixSequentialFile(const std::string& fname, FILE* f,
+      const EnvOptions& options)
+      : filename_(fname), file_(f), fd_(fileno(f)),
+        use_os_buffer_(options.use_os_buffer) {
+  }
   virtual ~PosixSequentialFile() { fclose(file_); }
 
   virtual Status Read(size_t n, Slice* result, char* scratch) {
@@ -82,6 +123,11 @@ class PosixSequentialFile: public SequentialFile {
         s = IOError(filename_, errno);
       }
     }
+    if (!use_os_buffer_) {
+      // we need to fadvise away the entire range of pages because
+      // we do not want readahead pages to be cached.
+      posix_fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED); // free OS pages
+    }
     return s;
   }
 
@@ -98,14 +144,13 @@ class PosixRandomAccessFile: public RandomAccessFile {
  private:
   std::string filename_;
   int fd_;
+  bool use_os_buffer_;
 
  public:
-  PosixRandomAccessFile(const std::string& fname, int fd)
-      : filename_(fname), fd_(fd) {
-    if (!useFsReadAhead) {
-      // disable read-aheads
-      posix_fadvise(fd, 0, 0, POSIX_FADV_RANDOM);
-    }
+  PosixRandomAccessFile(const std::string& fname, int fd,
+                        const EnvOptions& options)
+      : filename_(fname), fd_(fd), use_os_buffer_(options.use_os_buffer) {
+    assert(!options.use_mmap_reads);
   }
   virtual ~PosixRandomAccessFile() { close(fd_); }
 
@@ -118,7 +163,7 @@ class PosixRandomAccessFile: public RandomAccessFile {
       // An error: return a non-ok status
       s = IOError(filename_, errno);
     }
-    if (!useOsBuffer) {
+    if (!use_os_buffer_) {
       // we need to fadvise away the entire range of pages because
       // we do not want readahead pages to be cached.
       posix_fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED); // free OS pages
@@ -154,6 +199,30 @@ class PosixRandomAccessFile: public RandomAccessFile {
     return static_cast<size_t>(rid-id);
   }
 #endif
+
+  virtual void Hint(AccessPattern pattern) {
+    switch(pattern) {
+      case NORMAL:
+        posix_fadvise(fd_, 0, 0, POSIX_FADV_NORMAL);
+        break;
+      case RANDOM:
+        posix_fadvise(fd_, 0, 0, POSIX_FADV_RANDOM);
+        break;
+      case SEQUENTIAL:
+        posix_fadvise(fd_, 0, 0, POSIX_FADV_SEQUENTIAL);
+        break;
+      case WILLNEED:
+        posix_fadvise(fd_, 0, 0, POSIX_FADV_WILLNEED);
+        break;
+      case DONTNEED:
+        posix_fadvise(fd_, 0, 0, POSIX_FADV_DONTNEED);
+        break;
+      default:
+        assert(false);
+        break;
+    }
+  }
+
 };
 
 // mmap() based random-access
@@ -165,8 +234,12 @@ class PosixMmapReadableFile: public RandomAccessFile {
 
  public:
   // base[0,length-1] contains the mmapped contents of the file.
-  PosixMmapReadableFile(const std::string& fname, void* base, size_t length)
-      : filename_(fname), mmapped_region_(base), length_(length) { }
+  PosixMmapReadableFile(const std::string& fname, void* base, size_t length,
+                        const EnvOptions& options)
+      : filename_(fname), mmapped_region_(base), length_(length) {
+    assert(options.use_mmap_reads);
+    assert(options.use_os_buffer);
+  }
   virtual ~PosixMmapReadableFile() { munmap(mmapped_region_, length_); }
 
   virtual Status Read(uint64_t offset, size_t n, Slice* result,
@@ -214,6 +287,7 @@ class PosixMmapFile : public WritableFile {
 
   bool UnmapCurrentRegion() {
     bool result = true;
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     if (base_ != nullptr) {
       if (last_sync_ < limit_) {
         // Defer syncing this data until next Sync() call, if any
@@ -239,18 +313,22 @@ class PosixMmapFile : public WritableFile {
   Status MapNewRegion() {
     assert(base_ == nullptr);
 
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     int alloc_status = posix_fallocate(fd_, file_offset_, map_size_);
     if (alloc_status != 0) {
       return Status::IOError("Error allocating space to file : " + filename_ +
         "Error : " + strerror(alloc_status));
     }
 
-
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     void* ptr = mmap(nullptr, map_size_, PROT_READ | PROT_WRITE, MAP_SHARED,
                      fd_, file_offset_);
     if (ptr == MAP_FAILED) {
       return Status::IOError("MMap failed on " + filename_);
     }
+
+    TEST_KILL_RANDOM(leveldb_kill_odds);
+
     base_ = reinterpret_cast<char*>(ptr);
     limit_ = base_ + map_size_;
     dst_ = base_;
@@ -259,7 +337,8 @@ class PosixMmapFile : public WritableFile {
   }
 
  public:
-  PosixMmapFile(const std::string& fname, int fd, size_t page_size)
+  PosixMmapFile(const std::string& fname, int fd, size_t page_size,
+                const EnvOptions& options)
       : filename_(fname),
         fd_(fd),
         page_size_(page_size),
@@ -271,6 +350,7 @@ class PosixMmapFile : public WritableFile {
         file_offset_(0),
         pending_sync_(false) {
     assert((page_size & (page_size - 1)) == 0);
+    assert(options.use_mmap_writes);
   }
 
 
@@ -283,6 +363,7 @@ class PosixMmapFile : public WritableFile {
   virtual Status Append(const Slice& data) {
     const char* src = data.data();
     size_t left = data.size();
+    TEST_KILL_RANDOM(leveldb_kill_odds * REDUCE_ODDS);
     PrepareWrite(GetFileSize(), left);
     while (left > 0) {
       assert(base_ <= dst_);
@@ -294,6 +375,7 @@ class PosixMmapFile : public WritableFile {
           if (!s.ok()) {
             return s;
           }
+          TEST_KILL_RANDOM(leveldb_kill_odds);
         }
       }
 
@@ -303,12 +385,16 @@ class PosixMmapFile : public WritableFile {
       src += n;
       left -= n;
     }
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     return Status::OK();
   }
 
   virtual Status Close() {
     Status s;
     size_t unused = limit_ - dst_;
+
+    TEST_KILL_RANDOM(leveldb_kill_odds);
+
     if (!UnmapCurrentRegion()) {
       s = IOError(filename_, errno);
     } else if (unused > 0) {
@@ -318,6 +404,8 @@ class PosixMmapFile : public WritableFile {
       }
     }
 
+    TEST_KILL_RANDOM(leveldb_kill_odds);
+
     if (close(fd_) < 0) {
       if (s.ok()) {
         s = IOError(filename_, errno);
@@ -331,6 +419,7 @@ class PosixMmapFile : public WritableFile {
   }
 
   virtual Status Flush() {
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     return Status::OK();
   }
 
@@ -339,10 +428,12 @@ class PosixMmapFile : public WritableFile {
 
     if (pending_sync_) {
       // Some unmapped data was not synced
+      TEST_KILL_RANDOM(leveldb_kill_odds);
       pending_sync_ = false;
       if (fdatasync(fd_) < 0) {
         s = IOError(filename_, errno);
       }
+      TEST_KILL_RANDOM(leveldb_kill_odds * REDUCE_ODDS);
     }
 
     if (dst_ > last_sync_) {
@@ -351,9 +442,11 @@ class PosixMmapFile : public WritableFile {
       size_t p1 = TruncateToPageBoundary(last_sync_ - base_);
       size_t p2 = TruncateToPageBoundary(dst_ - base_ - 1);
       last_sync_ = dst_;
+      TEST_KILL_RANDOM(leveldb_kill_odds);
       if (msync(base_ + p1, p2 - p1 + page_size_, MS_SYNC) < 0) {
         s = IOError(filename_, errno);
       }
+      TEST_KILL_RANDOM(leveldb_kill_odds);
     }
 
     return s;
@@ -365,10 +458,12 @@ class PosixMmapFile : public WritableFile {
   virtual Status Fsync() {
     if (pending_sync_) {
       // Some unmapped data was not synced
+      TEST_KILL_RANDOM(leveldb_kill_odds);
       pending_sync_ = false;
       if (fsync(fd_) < 0) {
         return IOError(filename_, errno);
       }
+      TEST_KILL_RANDOM(leveldb_kill_odds);
     }
     // This invocation to Sync will not issue the call to
     // fdatasync because pending_sync_ has already been cleared.
@@ -387,6 +482,7 @@ class PosixMmapFile : public WritableFile {
 
 #ifdef OS_LINUX
   virtual Status Allocate(off_t offset, off_t len) {
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     if (!fallocate(fd_, FALLOC_FL_KEEP_SIZE, offset, len)) {
       return Status::OK();
     } else {
@@ -403,13 +499,16 @@ class PosixWritableFile : public WritableFile {
   int fd_;
   size_t cursize_;      // current size of cached data in buf_
   size_t capacity_;     // max size of buf_
-  char* buf_;           // a buffer to cache writes
+  unique_ptr<char[]> buf_;           // a buffer to cache writes
   uint64_t filesize_;
   bool pending_sync_;
   bool pending_fsync_;
+  uint64_t last_sync_size_;
+  uint64_t bytes_per_sync_;
 
  public:
-  PosixWritableFile(const std::string& fname, int fd, size_t capacity) :
+  PosixWritableFile(const std::string& fname, int fd, size_t capacity,
+                    const EnvOptions& options) :
     filename_(fname),
     fd_(fd),
     cursize_(0),
@@ -417,15 +516,16 @@ class PosixWritableFile : public WritableFile {
     buf_(new char[capacity]),
     filesize_(0),
     pending_sync_(false),
-    pending_fsync_(false) {
+    pending_fsync_(false),
+    last_sync_size_(0),
+    bytes_per_sync_(options.bytes_per_sync) {
+    assert(!options.use_mmap_writes);
   }
 
   ~PosixWritableFile() {
     if (fd_ >= 0) {
       PosixWritableFile::Close();
     }
-    delete buf_;
-    buf_ = 0;
   }
 
   virtual Status Append(const Slice& data) {
@@ -435,6 +535,8 @@ class PosixWritableFile : public WritableFile {
     pending_sync_ = true;
     pending_fsync_ = true;
 
+    TEST_KILL_RANDOM(leveldb_kill_odds * REDUCE_ODDS2);
+
     PrepareWrite(GetFileSize(), left);
     // if there is no space in the cache, then flush
     if (cursize_ + left > capacity_) {
@@ -444,9 +546,8 @@ class PosixWritableFile : public WritableFile {
       }
       // Increase the buffer size, but capped at 1MB
       if (capacity_ < (1<<20)) {
-        delete buf_;
         capacity_ *= 2;
-        buf_ = new char[capacity_];
+        buf_.reset(new char[capacity_]);
       }
       assert(cursize_ == 0);
     }
@@ -454,7 +555,7 @@ class PosixWritableFile : public WritableFile {
     // if the write fits into the cache, then write to cache
     // otherwise do a write() syscall to write to OS buffers.
     if (cursize_ + left <= capacity_) {
-      memcpy(buf_+cursize_, src, left);
+      memcpy(buf_.get()+cursize_, src, left);
       cursize_ += left;
     } else {
       while (left != 0) {
@@ -462,6 +563,8 @@ class PosixWritableFile : public WritableFile {
         if (done < 0) {
           return IOError(filename_, errno);
         }
+        TEST_KILL_RANDOM(leveldb_kill_odds);
+
         left -= done;
         src += done;
       }
@@ -475,6 +578,9 @@ class PosixWritableFile : public WritableFile {
     s = Flush(); // flush cache to OS
     if (!s.ok()) {
     }
+
+    TEST_KILL_RANDOM(leveldb_kill_odds);
+
     if (close(fd_) < 0) {
       if (s.ok()) {
         s = IOError(filename_, errno);
@@ -486,32 +592,49 @@ class PosixWritableFile : public WritableFile {
 
   // write out the cached data to the OS cache
   virtual Status Flush() {
+    TEST_KILL_RANDOM(leveldb_kill_odds * REDUCE_ODDS2);
     size_t left = cursize_;
-    char* src = buf_;
+    char* src = buf_.get();
     while (left != 0) {
       ssize_t done = write(fd_, src, left);
       if (done < 0) {
         return IOError(filename_, errno);
       }
+      TEST_KILL_RANDOM(leveldb_kill_odds * REDUCE_ODDS2);
       left -= done;
       src += done;
     }
     cursize_ = 0;
+
+    // sync OS cache to disk for every bytes_per_sync_
+    // TODO: give log file and sst file different options (log
+    // files could be potentially cached in OS for their whole
+    // life time, thus we might not want to flush at all).
+    if (bytes_per_sync_ &&
+        filesize_ - last_sync_size_ >= bytes_per_sync_) {
+      RangeSync(last_sync_size_, filesize_ - last_sync_size_);
+      last_sync_size_ = filesize_;
+    }
+
     return Status::OK();
   }
 
   virtual Status Sync() {
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     if (pending_sync_ && fdatasync(fd_) < 0) {
       return IOError(filename_, errno);
     }
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     pending_sync_ = false;
     return Status::OK();
   }
 
   virtual Status Fsync() {
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     if (pending_fsync_ && fsync(fd_) < 0) {
       return IOError(filename_, errno);
     }
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     pending_fsync_ = false;
     pending_sync_ = false;
     return Status::OK();
@@ -523,12 +646,21 @@ class PosixWritableFile : public WritableFile {
 
 #ifdef OS_LINUX
   virtual Status Allocate(off_t offset, off_t len) {
+    TEST_KILL_RANDOM(leveldb_kill_odds);
     if (!fallocate(fd_, FALLOC_FL_KEEP_SIZE, offset, len)) {
       return Status::OK();
     } else {
       return IOError(filename_, errno);
     }
   }
+
+  virtual Status RangeSync(off64_t offset, off64_t nbytes) {
+    if (sync_file_range(fd_, offset, nbytes, SYNC_FILE_RANGE_WRITE) == 0) {
+      return Status::OK();
+    } else {
+      return IOError(filename_, errno);
+    }
+  }
 #endif
 };
 
@@ -579,32 +711,43 @@ class PosixFileLock : public FileLock {
 class PosixEnv : public Env {
  public:
   PosixEnv();
-  virtual ~PosixEnv() {
-    fprintf(stderr, "Destroying Env::Default()\n");
-    exit(1);
+
+  virtual ~PosixEnv(){
+    WaitForBGThreads();
+  }
+
+  void SetFD_CLOEXEC(int fd, const EnvOptions* options) {
+    if ((options == nullptr || options->set_fd_cloexec) && fd > 0) {
+      fcntl(fd, F_SETFD, fcntl(fd, F_GETFD) | FD_CLOEXEC);
+    }
   }
 
   virtual Status NewSequentialFile(const std::string& fname,
-                                   unique_ptr<SequentialFile>* result) {
+                                   unique_ptr<SequentialFile>* result,
+                                   const EnvOptions& options) {
     result->reset();
     FILE* f = fopen(fname.c_str(), "r");
     if (f == nullptr) {
       *result = nullptr;
       return IOError(fname, errno);
     } else {
-      result->reset(new PosixSequentialFile(fname, f));
+      int fd = fileno(f);
+      SetFD_CLOEXEC(fd, &options);
+      result->reset(new PosixSequentialFile(fname, f, options));
       return Status::OK();
     }
   }
 
   virtual Status NewRandomAccessFile(const std::string& fname,
-                                     unique_ptr<RandomAccessFile>* result) {
+                                     unique_ptr<RandomAccessFile>* result,
+                                     const EnvOptions& options) {
     result->reset();
     Status s;
     int fd = open(fname.c_str(), O_RDONLY);
+    SetFD_CLOEXEC(fd, &options);
     if (fd < 0) {
       s = IOError(fname, errno);
-    } else if (useMmapRead && sizeof(void*) >= 8) {
+    } else if (options.use_mmap_reads && sizeof(void*) >= 8) {
       // Use of mmap for random reads has been removed because it
       // kills performance when storage is fast.
       // Use mmap when virtual address-space is plentiful.
@@ -613,39 +756,42 @@ class PosixEnv : public Env {
       if (s.ok()) {
         void* base = mmap(nullptr, size, PROT_READ, MAP_SHARED, fd, 0);
         if (base != MAP_FAILED) {
-          result->reset(new PosixMmapReadableFile(fname, base, size));
+          result->reset(new PosixMmapReadableFile(fname, base, size, options));
         } else {
           s = IOError(fname, errno);
         }
       }
       close(fd);
     } else {
-      result->reset(new PosixRandomAccessFile(fname, fd));
+      result->reset(new PosixRandomAccessFile(fname, fd, options));
     }
     return s;
   }
 
   virtual Status NewWritableFile(const std::string& fname,
-                                 unique_ptr<WritableFile>* result) {
+                                 unique_ptr<WritableFile>* result,
+                                 const EnvOptions& options) {
     result->reset();
     Status s;
     const int fd = open(fname.c_str(), O_CREAT | O_RDWR | O_TRUNC, 0644);
     if (fd < 0) {
       s = IOError(fname, errno);
     } else {
-      if (!checkedDiskForMmap_) {
-        // this will be executed once in the program's lifetime.
-        if (useMmapWrite) {
+      SetFD_CLOEXEC(fd, &options);
+      if (options.use_mmap_writes) {
+        if (!checkedDiskForMmap_) {
+          // this will be executed once in the program's lifetime.
           // do not use mmapWrite on non ext-3/xfs/tmpfs systems.
-          useMmapWrite = SupportsFastAllocate(fname);
+          if (!SupportsFastAllocate(fname)) {
+            forceMmapOff = true;
+          }
+          checkedDiskForMmap_ = true;
         }
-        checkedDiskForMmap_ = true;
       }
-
-      if (useMmapWrite) {
-        result->reset(new PosixMmapFile(fname, fd, page_size_));
+      if (options.use_mmap_writes && !forceMmapOff) {
+        result->reset(new PosixMmapFile(fname, fd, page_size_, options));
       } else {
-        result->reset(new PosixWritableFile(fname, fd, 65536));
+        result->reset(new PosixWritableFile(fname, fd, 65536, options));
       }
     }
     return s;
@@ -747,6 +893,7 @@ class PosixEnv : public Env {
       result = IOError("lock " + fname, errno);
       close(fd);
     } else {
+      SetFD_CLOEXEC(fd, nullptr);
       PosixFileLock* my_lock = new PosixFileLock;
       my_lock->fd_ = fd;
       my_lock->filename = fname;
@@ -768,6 +915,8 @@ class PosixEnv : public Env {
 
   virtual void Schedule(void (*function)(void*), void* arg);
 
+  virtual void WaitForBGThreads();
+
   virtual void StartThread(void (*function)(void* arg), void* arg);
 
   virtual Status GetTestDirectory(std::string* result) {
@@ -798,6 +947,8 @@ class PosixEnv : public Env {
       result->reset();
       return IOError(fname, errno);
     } else {
+      int fd = fileno(f);
+      SetFD_CLOEXEC(fd, nullptr);
       result->reset(new PosixLogger(f, &PosixEnv::gettid));
       return Status::OK();
     }
@@ -852,10 +1003,12 @@ class PosixEnv : public Env {
 
   // Allow increasing the number of worker threads.
   virtual void SetBackgroundThreads(int num) {
+    PthreadCall("lock", pthread_mutex_lock(&mu_));
     if (num > num_threads_) {
       num_threads_ = num;
       bgthread_.resize(num_threads_);
     }
+    PthreadCall("unlock", pthread_mutex_unlock(&mu_));
   }
 
   virtual std::string TimeToString(uint64_t secondsSince1970) {
@@ -879,7 +1032,8 @@ class PosixEnv : public Env {
   }
 
  private:
-  bool checkedDiskForMmap_ = false;
+  bool checkedDiskForMmap_;
+  bool forceMmapOff; // do we override Env options?
 
   void PthreadCall(const char* label, int result) {
     if (result != 0) {
@@ -933,21 +1087,42 @@ class PosixEnv : public Env {
   struct BGItem { void* arg; void (*function)(void*); };
   typedef std::deque<BGItem> BGQueue;
   int queue_size_; // number of items in BGQueue
+  bool exit_all_threads_;
   BGQueue queue_;
+  std::vector<pthread_t> threads_to_join_;
 };
 
-PosixEnv::PosixEnv() : page_size_(getpagesize()),
+PosixEnv::PosixEnv() : checkedDiskForMmap_(false),
+                       forceMmapOff(false),
+                       page_size_(getpagesize()),
                        started_bgthread_(0),
                        num_threads_(1),
-                       queue_size_(0) {
+                       queue_size_(0),
+                       exit_all_threads_(false) {
   PthreadCall("mutex_init", pthread_mutex_init(&mu_, nullptr));
   PthreadCall("cvar_init", pthread_cond_init(&bgsignal_, nullptr));
   bgthread_.resize(num_threads_);
 }
 
+// Signal and Join all background threads started by calls to Schedule
+void PosixEnv::WaitForBGThreads() {
+  PthreadCall("lock", pthread_mutex_lock(&mu_));
+  assert(! exit_all_threads_);
+  exit_all_threads_ = true;
+  PthreadCall("signalall", pthread_cond_broadcast(&bgsignal_));
+  PthreadCall("unlock", pthread_mutex_unlock(&mu_));
+  for (unsigned int i = 0; i < threads_to_join_.size(); i++) {
+    pthread_join(threads_to_join_[i], nullptr);
+  }
+}
+
 void PosixEnv::Schedule(void (*function)(void*), void* arg) {
   PthreadCall("lock", pthread_mutex_lock(&mu_));
 
+  if (exit_all_threads_) {
+    PthreadCall("unlock", pthread_mutex_unlock(&mu_));
+    return;
+  }
   // Start background thread if necessary
   for (; started_bgthread_ < num_threads_; started_bgthread_++) {
     PthreadCall(
@@ -956,17 +1131,17 @@ void PosixEnv::Schedule(void (*function)(void*), void* arg) {
                        nullptr,
                        &PosixEnv::BGThreadWrapper,
                        this));
+    threads_to_join_.push_back(bgthread_[started_bgthread_]);
     fprintf(stdout, "Created bg thread 0x%lx\n", bgthread_[started_bgthread_]);
   }
 
-  // always wake up at least one waiting thread.
-  PthreadCall("signal", pthread_cond_signal(&bgsignal_));
-
   // Add to priority queue
   queue_.push_back(BGItem());
   queue_.back().function = function;
   queue_.back().arg = arg;
-  queue_size_++;
+
+  // always wake up at least one waiting thread.
+  PthreadCall("signal", pthread_cond_signal(&bgsignal_));
 
   PthreadCall("unlock", pthread_mutex_unlock(&mu_));
 }
@@ -975,14 +1150,16 @@ void PosixEnv::BGThread() {
   while (true) {
     // Wait until there is an item that is ready to run
     PthreadCall("lock", pthread_mutex_lock(&mu_));
-    while (queue_.empty()) {
+    while (queue_.empty() && !exit_all_threads_) {
       PthreadCall("wait", pthread_cond_wait(&bgsignal_, &mu_));
     }
-
+    if (exit_all_threads_) { // mechanism to let BG threads exit safely
+      PthreadCall("unlock", pthread_mutex_unlock(&mu_));
+      break;
+    }
     void (*function)(void*) = queue_.front().function;
     void* arg = queue_.front().arg;
     queue_.pop_front();
-    queue_size_--;
 
     PthreadCall("unlock", pthread_mutex_unlock(&mu_));
     (*function)(arg);
@@ -1009,17 +1186,14 @@ void PosixEnv::StartThread(void (*function)(void* arg), void* arg) {
   state->arg = arg;
   PthreadCall("start thread",
               pthread_create(&t, nullptr,  &StartThreadWrapper, state));
+  threads_to_join_.push_back(t);
 }
 
 }  // namespace
 
-static pthread_once_t once = PTHREAD_ONCE_INIT;
-static Env* default_env;
-static void InitDefaultEnv() { default_env = new PosixEnv; }
-
 Env* Env::Default() {
-  pthread_once(&once, InitDefaultEnv);
-  return default_env;
+  static PosixEnv default_env;
+  return &default_env;
 }
 
 }  // namespace leveldb

util/env_test.cc

@@ -119,21 +119,22 @@ char temp_id[MAX_ID_SIZE];
 
 TEST(EnvPosixTest, RandomAccessUniqueID) {
   // Create file.
+  const EnvOptions soptions;
   std::string fname = test::TmpDir() + "/" + "testfile";
   unique_ptr<WritableFile> wfile;
-  ASSERT_OK(env_->NewWritableFile(fname, &wfile));
+  ASSERT_OK(env_->NewWritableFile(fname, &wfile, soptions));
 
   unique_ptr<RandomAccessFile> file;
 
   // Get Unique ID
-  ASSERT_OK(env_->NewRandomAccessFile(fname, &file));
+  ASSERT_OK(env_->NewRandomAccessFile(fname, &file, soptions));
   size_t id_size = file->GetUniqueId(temp_id, MAX_ID_SIZE);
   ASSERT_TRUE(id_size > 0);
   std::string unique_id1(temp_id, id_size);
   ASSERT_TRUE(IsUniqueIDValid(unique_id1));
 
   // Get Unique ID again
-  ASSERT_OK(env_->NewRandomAccessFile(fname, &file));
+  ASSERT_OK(env_->NewRandomAccessFile(fname, &file, soptions));
   id_size = file->GetUniqueId(temp_id, MAX_ID_SIZE);
   ASSERT_TRUE(id_size > 0);
   std::string unique_id2(temp_id, id_size);
@@ -141,7 +142,7 @@ TEST(EnvPosixTest, RandomAccessUniqueID) {
 
   // Get Unique ID again after waiting some time.
   env_->SleepForMicroseconds(1000000);
-  ASSERT_OK(env_->NewRandomAccessFile(fname, &file));
+  ASSERT_OK(env_->NewRandomAccessFile(fname, &file, soptions));
   id_size = file->GetUniqueId(temp_id, MAX_ID_SIZE);
   ASSERT_TRUE(id_size > 0);
   std::string unique_id3(temp_id, id_size);
@@ -172,6 +173,7 @@ bool HasPrefix(const std::unordered_set<std::string>& ss) {
 
 TEST(EnvPosixTest, RandomAccessUniqueIDConcurrent) {
   // Check whether a bunch of concurrently existing files have unique IDs.
+  const EnvOptions soptions;
 
   // Create the files
   std::vector<std::string> fnames;
@@ -180,7 +182,7 @@ TEST(EnvPosixTest, RandomAccessUniqueIDConcurrent) {
 
     // Create file.
     unique_ptr<WritableFile> wfile;
-    ASSERT_OK(env_->NewWritableFile(fnames[i], &wfile));
+    ASSERT_OK(env_->NewWritableFile(fnames[i], &wfile, soptions));
   }
 
   // Collect and check whether the IDs are unique.
@@ -188,7 +190,7 @@ TEST(EnvPosixTest, RandomAccessUniqueIDConcurrent) {
   for (const std::string fname: fnames) {
     unique_ptr<RandomAccessFile> file;
     std::string unique_id;
-    ASSERT_OK(env_->NewRandomAccessFile(fname, &file));
+    ASSERT_OK(env_->NewRandomAccessFile(fname, &file, soptions));
     size_t id_size = file->GetUniqueId(temp_id, MAX_ID_SIZE);
     ASSERT_TRUE(id_size > 0);
     unique_id = std::string(temp_id, id_size);
@@ -207,6 +209,7 @@ TEST(EnvPosixTest, RandomAccessUniqueIDConcurrent) {
 }
 
 TEST(EnvPosixTest, RandomAccessUniqueIDDeletes) {
+  const EnvOptions soptions;
   std::string fname = test::TmpDir() + "/" + "testfile";
 
   // Check that after file is deleted we don't get same ID again in a new file.
@@ -215,14 +218,14 @@ TEST(EnvPosixTest, RandomAccessUniqueIDDeletes) {
     // Create file.
     {
       unique_ptr<WritableFile> wfile;
-      ASSERT_OK(env_->NewWritableFile(fname, &wfile));
+      ASSERT_OK(env_->NewWritableFile(fname, &wfile, soptions));
     }
 
     // Get Unique ID
     std::string unique_id;
     {
       unique_ptr<RandomAccessFile> file;
-      ASSERT_OK(env_->NewRandomAccessFile(fname, &file));
+      ASSERT_OK(env_->NewRandomAccessFile(fname, &file, soptions));
       size_t id_size = file->GetUniqueId(temp_id, MAX_ID_SIZE);
       ASSERT_TRUE(id_size > 0);
       unique_id = std::string(temp_id, id_size);

util/histogram.cc

@@ -85,17 +85,13 @@ void HistogramImpl::Add(uint64_t value) {
   sum_squares_ += (value * value);
 }
 
-void HistogramImpl::Add(double value) {
-  Add(static_cast<uint64_t>(value));
-}
-
 void HistogramImpl::Merge(const HistogramImpl& other) {
   if (other.min_ < min_) min_ = other.min_;
   if (other.max_ > max_) max_ = other.max_;
   num_ += other.num_;
   sum_ += other.sum_;
   sum_squares_ += other.sum_squares_;
-  for (int b = 0; b < bucketMapper.BucketCount(); b++) {
+  for (unsigned int b = 0; b < bucketMapper.BucketCount(); b++) {
     buckets_[b] += other.buckets_[b];
   }
 }
@@ -107,7 +103,7 @@ double HistogramImpl::Median() const {
 double HistogramImpl::Percentile(double p) const {
   double threshold = num_ * (p / 100.0);
   double sum = 0;
-  for (int b = 0; b < bucketMapper.BucketCount(); b++) {
+  for (unsigned int b = 0; b < bucketMapper.BucketCount(); b++) {
     sum += buckets_[b];
     if (sum >= threshold) {
       // Scale linearly within this bucket
@@ -152,13 +148,15 @@ std::string HistogramImpl::ToString() const {
            (num_ == 0.0 ? 0.0 : min_), Median(), max_);
   r.append(buf);
   snprintf(buf, sizeof(buf),
-           "Percentiles: P50: %.2f P75: %.2f P99: %.2f P99.9: %.2f\n",
-           Percentile(50), Percentile(75), Percentile(99), Percentile(99.9));
+           "Percentiles: "
+           "P50: %.2f P75: %.2f P99: %.2f P99.9: %.2f P99.99: %.2f\n",
+           Percentile(50), Percentile(75), Percentile(99), Percentile(99.9),
+           Percentile(99.99));
   r.append(buf);
   r.append("------------------------------------------------------\n");
   const double mult = 100.0 / num_;
   double sum = 0;
-  for (int b = 0; b < bucketMapper.BucketCount(); b++) {
+  for (unsigned int b = 0; b < bucketMapper.BucketCount(); b++) {
     if (buckets_[b] <= 0.0) continue;
     sum += buckets_[b];
     snprintf(buf, sizeof(buf),

util/histogram.h

@@ -53,7 +53,6 @@ class HistogramImpl {
   virtual ~HistogramImpl() {}
   virtual void Clear();
   virtual void Add(uint64_t value);
-  virtual void Add(double value);
   void Merge(const HistogramImpl& other);
 
   virtual std::string ToString() const;

util/ldb_cmd.cc

@@ -4,20 +4,31 @@
 
 #include "util/ldb_cmd.h"
 
-#include <boost/algorithm/string.hpp>
-#include <boost/lexical_cast.hpp>
-
-#include "leveldb/write_batch.h"
 #include "db/dbformat.h"
 #include "db/log_reader.h"
+#include "db/filename.h"
 #include "db/write_batch_internal.h"
+#include "leveldb/write_batch.h"
+#include "util/coding.h"
+
+#include <ctime>
+#include <dirent.h>
+#include <sstream>
+#include <string>
+#include <stdexcept>
 
 namespace leveldb {
 
+using namespace std;
+
 const string LDBCommand::ARG_DB = "db";
 const string LDBCommand::ARG_HEX = "hex";
 const string LDBCommand::ARG_KEY_HEX = "key_hex";
 const string LDBCommand::ARG_VALUE_HEX = "value_hex";
+const string LDBCommand::ARG_TTL = "ttl";
+const string LDBCommand::ARG_TTL_START = "start_time";
+const string LDBCommand::ARG_TTL_END = "end_time";
+const string LDBCommand::ARG_TIMESTAMP = "timestamp";
 const string LDBCommand::ARG_FROM = "from";
 const string LDBCommand::ARG_TO = "to";
 const string LDBCommand::ARG_MAX_KEYS = "max_keys";
@@ -31,12 +42,16 @@ const string LDBCommand::ARG_CREATE_IF_MISSING = "create_if_missing";
 
 const char* LDBCommand::DELIM = " ==> ";
 
-LDBCommand* LDBCommand::InitFromCmdLineArgs(int argc, char** argv) {
+LDBCommand* LDBCommand::InitFromCmdLineArgs(
+  int argc,
+  char** argv,
+  Options options
+) {
   vector<string> args;
   for (int i = 1; i < argc; i++) {
     args.push_back(argv[i]);
   }
-  return InitFromCmdLineArgs(args);
+  return InitFromCmdLineArgs(args, options);
 }
 
 /**
@@ -49,14 +64,17 @@ LDBCommand* LDBCommand::InitFromCmdLineArgs(int argc, char** argv) {
  * Command name is not included in args.
  * Returns nullptr if the command-line cannot be parsed.
  */
-LDBCommand* LDBCommand::InitFromCmdLineArgs(const vector<string>& args) {
+LDBCommand* LDBCommand::InitFromCmdLineArgs(
+  const vector<string>& args,
+  Options options
+) {
   // --x=y command line arguments are added as x->y map entries.
-  map<string, string> options;
+  map<string, string> option_map;
 
   // Command-line arguments of the form --hex end up in this array as hex
   vector<string> flags;
 
-  // Everything other than options and flags. Represents commands
+  // Everything other than option_map and flags. Represents commands
   // and their parameters.  For eg: put key1 value1 go into this vector.
   vector<string> cmdTokens;
 
@@ -65,12 +83,11 @@ LDBCommand* LDBCommand::InitFromCmdLineArgs(const vector<string>& args) {
   for (vector<string>::const_iterator itr = args.begin();
       itr != args.end(); itr++) {
     string arg = *itr;
-    if (boost::starts_with(arg, OPTION_PREFIX)){
-      vector<string> splits;
-      boost::split(splits, arg, boost::is_any_of("="));
+    if (arg[0] == '-' && arg[1] == '-'){
+      vector<string> splits = stringSplit(arg, '=');
       if (splits.size() == 2) {
         string optionKey = splits[0].substr(OPTION_PREFIX.size());
-        options[optionKey] = splits[1];
+        option_map[optionKey] = splits[1];
       } else {
         string optionKey = splits[0].substr(OPTION_PREFIX.size());
         flags.push_back(optionKey);
@@ -87,36 +104,57 @@ LDBCommand* LDBCommand::InitFromCmdLineArgs(const vector<string>& args) {
 
   string cmd = cmdTokens[0];
   vector<string> cmdParams(cmdTokens.begin()+1, cmdTokens.end());
+  LDBCommand* command = LDBCommand::SelectCommand(
+    cmd,
+    cmdParams,
+    option_map,
+    flags
+  );
+
+  if (command) {
+    command->SetOptions(options);
+  }
+  return command;
+}
+
+LDBCommand* LDBCommand::SelectCommand(
+    const std::string& cmd,
+    vector<string>& cmdParams,
+    map<string, string>& option_map,
+    vector<string>& flags
+  ) {
 
   if (cmd == GetCommand::Name()) {
-    return new GetCommand(cmdParams, options, flags);
+    return new GetCommand(cmdParams, option_map, flags);
   } else if (cmd == PutCommand::Name()) {
-    return new PutCommand(cmdParams, options, flags);
+    return new PutCommand(cmdParams, option_map, flags);
   } else if (cmd == BatchPutCommand::Name()) {
-    return new BatchPutCommand(cmdParams, options, flags);
+    return new BatchPutCommand(cmdParams, option_map, flags);
   } else if (cmd == ScanCommand::Name()) {
-    return new ScanCommand(cmdParams, options, flags);
+    return new ScanCommand(cmdParams, option_map, flags);
   } else if (cmd == DeleteCommand::Name()) {
-    return new DeleteCommand(cmdParams, options, flags);
+    return new DeleteCommand(cmdParams, option_map, flags);
   } else if (cmd == ApproxSizeCommand::Name()) {
-    return new ApproxSizeCommand(cmdParams, options, flags);
+    return new ApproxSizeCommand(cmdParams, option_map, flags);
   } else if (cmd == DBQuerierCommand::Name()) {
-    return new DBQuerierCommand(cmdParams, options, flags);
+    return new DBQuerierCommand(cmdParams, option_map, flags);
   } else if (cmd == CompactorCommand::Name()) {
-    return new CompactorCommand(cmdParams, options, flags);
+    return new CompactorCommand(cmdParams, option_map, flags);
   } else if (cmd == WALDumperCommand::Name()) {
-    return new WALDumperCommand(cmdParams, options, flags);
+    return new WALDumperCommand(cmdParams, option_map, flags);
   } else if (cmd == ReduceDBLevelsCommand::Name()) {
-    return new ReduceDBLevelsCommand(cmdParams, options, flags);
+    return new ReduceDBLevelsCommand(cmdParams, option_map, flags);
   } else if (cmd == DBDumperCommand::Name()) {
-    return new DBDumperCommand(cmdParams, options, flags);
+    return new DBDumperCommand(cmdParams, option_map, flags);
   } else if (cmd == DBLoaderCommand::Name()) {
-    return new DBLoaderCommand(cmdParams, options, flags);
+    return new DBLoaderCommand(cmdParams, option_map, flags);
+  } else if (cmd == ManifestDumpCommand::Name()) {
+    return new ManifestDumpCommand(cmdParams, option_map, flags);
   }
-
   return nullptr;
 }
 
+
 /**
  * Parses the specific integer option and fills in the value.
  * Returns true if the option is found.
@@ -127,30 +165,33 @@ LDBCommand* LDBCommand::InitFromCmdLineArgs(const vector<string>& args) {
 bool LDBCommand::ParseIntOption(const map<string, string>& options,
     string option, int& value, LDBCommandExecuteResult& exec_state) {
 
-  map<string, string>::const_iterator itr = options_.find(option);
-  if (itr != options_.end()) {
+  map<string, string>::const_iterator itr = option_map_.find(option);
+  if (itr != option_map_.end()) {
     try {
-      value = boost::lexical_cast<int>(itr->second);
+      value = stoi(itr->second);
       return true;
-    } catch( const boost::bad_lexical_cast & ) {
+    } catch(const invalid_argument&) {
       exec_state = LDBCommandExecuteResult::FAILED(option +
                       " has an invalid value.");
+    } catch(const out_of_range&) {
+      exec_state = LDBCommandExecuteResult::FAILED(option +
+                      " has a value out-of-range.");
     }
   }
   return false;
 }
 
-leveldb::Options LDBCommand::PrepareOptionsForOpenDB() {
+Options LDBCommand::PrepareOptionsForOpenDB() {
 
-  leveldb::Options opt;
+  Options opt = options_;
   opt.create_if_missing = false;
 
   map<string, string>::const_iterator itr;
 
   int bits;
-  if (ParseIntOption(options_, ARG_BLOOM_BITS, bits, exec_state_)) {
+  if (ParseIntOption(option_map_, ARG_BLOOM_BITS, bits, exec_state_)) {
     if (bits > 0) {
-      opt.filter_policy = leveldb::NewBloomFilterPolicy(bits);
+      opt.filter_policy = NewBloomFilterPolicy(bits);
     } else {
       exec_state_ = LDBCommandExecuteResult::FAILED(ARG_BLOOM_BITS +
                       " must be > 0.");
@@ -158,7 +199,7 @@ leveldb::Options LDBCommand::PrepareOptionsForOpenDB() {
   }
 
   int block_size;
-  if (ParseIntOption(options_, ARG_BLOCK_SIZE, block_size, exec_state_)) {
+  if (ParseIntOption(option_map_, ARG_BLOCK_SIZE, block_size, exec_state_)) {
     if (block_size > 0) {
       opt.block_size = block_size;
     } else {
@@ -167,22 +208,22 @@ leveldb::Options LDBCommand::PrepareOptionsForOpenDB() {
     }
   }
 
-  itr = options_.find(ARG_AUTO_COMPACTION);
-  if (itr != options_.end()) {
+  itr = option_map_.find(ARG_AUTO_COMPACTION);
+  if (itr != option_map_.end()) {
     opt.disable_auto_compactions = ! StringToBool(itr->second);
   }
 
-  itr = options_.find(ARG_COMPRESSION_TYPE);
-  if (itr != options_.end()) {
+  itr = option_map_.find(ARG_COMPRESSION_TYPE);
+  if (itr != option_map_.end()) {
     string comp = itr->second;
     if (comp == "no") {
-      opt.compression = leveldb::kNoCompression;
+      opt.compression = kNoCompression;
     } else if (comp == "snappy") {
-      opt.compression = leveldb::kSnappyCompression;
+      opt.compression = kSnappyCompression;
     } else if (comp == "zlib") {
-      opt.compression = leveldb::kZlibCompression;
+      opt.compression = kZlibCompression;
     } else if (comp == "bzip2") {
-      opt.compression = leveldb::kBZip2Compression;
+      opt.compression = kBZip2Compression;
     } else {
       // Unknown compression.
       exec_state_ = LDBCommandExecuteResult::FAILED(
@@ -191,7 +232,7 @@ leveldb::Options LDBCommand::PrepareOptionsForOpenDB() {
   }
 
   int write_buffer_size;
-  if (ParseIntOption(options_, ARG_WRITE_BUFFER_SIZE, write_buffer_size,
+  if (ParseIntOption(option_map_, ARG_WRITE_BUFFER_SIZE, write_buffer_size,
         exec_state_)) {
     if (write_buffer_size > 0) {
       opt.write_buffer_size = write_buffer_size;
@@ -202,7 +243,7 @@ leveldb::Options LDBCommand::PrepareOptionsForOpenDB() {
   }
 
   int file_size;
-  if (ParseIntOption(options_, ARG_FILE_SIZE, file_size, exec_state_)) {
+  if (ParseIntOption(option_map_, ARG_FILE_SIZE, file_size, exec_state_)) {
     if (file_size > 0) {
       opt.target_file_size_base = file_size;
     } else {
@@ -217,7 +258,7 @@ leveldb::Options LDBCommand::PrepareOptionsForOpenDB() {
 bool LDBCommand::ParseKeyValue(const string& line, string* key, string* value,
                               bool is_key_hex, bool is_value_hex) {
   size_t pos = line.find(DELIM);
-  if (pos != std::string::npos) {
+  if (pos != string::npos) {
     *key = line.substr(0, pos);
     *value = line.substr(pos + strlen(DELIM));
     if (is_key_hex) {
@@ -241,9 +282,9 @@ bool LDBCommand::ParseKeyValue(const string& line, string* key, string* value,
  */
 bool LDBCommand::ValidateCmdLineOptions() {
 
-  for (map<string, string>::const_iterator itr = options_.begin();
-        itr != options_.end(); itr++) {
-    if (std::find(valid_cmd_line_options_.begin(),
+  for (map<string, string>::const_iterator itr = option_map_.begin();
+        itr != option_map_.end(); itr++) {
+    if (find(valid_cmd_line_options_.begin(),
           valid_cmd_line_options_.end(), itr->first) ==
           valid_cmd_line_options_.end()) {
       fprintf(stderr, "Invalid command-line option %s\n", itr->first.c_str());
@@ -253,7 +294,7 @@ bool LDBCommand::ValidateCmdLineOptions() {
 
   for (vector<string>::const_iterator itr = flags_.begin();
         itr != flags_.end(); itr++) {
-    if (std::find(valid_cmd_line_options_.begin(),
+    if (find(valid_cmd_line_options_.begin(),
           valid_cmd_line_options_.end(), *itr) ==
           valid_cmd_line_options_.end()) {
       fprintf(stderr, "Invalid command-line flag %s\n", itr->c_str());
@@ -261,7 +302,7 @@ bool LDBCommand::ValidateCmdLineOptions() {
     }
   }
 
-  if (options_.find(ARG_DB) == options_.end()) {
+  if (!NoDBOpen() && option_map_.find(ARG_DB) == option_map_.end()) {
     fprintf(stderr, "%s must be specified\n", ARG_DB.c_str());
     return false;
   }
@@ -307,13 +348,13 @@ void CompactorCommand::Help(string& ret) {
 
 void CompactorCommand::DoCommand() {
 
-  leveldb::Slice* begin = nullptr;
-  leveldb::Slice* end = nullptr;
+  Slice* begin = nullptr;
+  Slice* end = nullptr;
   if (!null_from_) {
-    begin = new leveldb::Slice(from_);
+    begin = new Slice(from_);
   }
   if (!null_to_) {
-    end = new leveldb::Slice(to_);
+    end = new Slice(to_);
   }
 
   db_->CompactRange(begin, end);
@@ -353,8 +394,8 @@ void DBLoaderCommand::Help(string& ret) {
   ret.append("\n");
 }
 
-leveldb::Options DBLoaderCommand::PrepareOptionsForOpenDB() {
-  leveldb::Options opt = LDBCommand::PrepareOptionsForOpenDB();
+Options DBLoaderCommand::PrepareOptionsForOpenDB() {
+  Options opt = LDBCommand::PrepareOptionsForOpenDB();
   opt.create_if_missing = create_if_missing_;
   if (bulk_load_) {
     opt.PrepareForBulkLoad();
@@ -374,7 +415,7 @@ void DBLoaderCommand::DoCommand() {
 
   int bad_lines = 0;
   string line;
-  while (std::getline(std::cin, line, '\n')) {
+  while (getline(cin, line, '\n')) {
     string key;
     string value;
     if (ParseKeyValue(line, &key, &value, is_key_hex_, is_value_hex_)) {
@@ -389,22 +430,147 @@ void DBLoaderCommand::DoCommand() {
   }
 
   if (bad_lines > 0) {
-    std::cout << "Warning: " << bad_lines << " bad lines ignored." << std::endl;
+    cout << "Warning: " << bad_lines << " bad lines ignored." << endl;
   }
   if (compact_) {
     db_->CompactRange(nullptr, nullptr);
   }
 }
 
+// ----------------------------------------------------------------------------
+
+const string ManifestDumpCommand::ARG_VERBOSE = "verbose";
+const string ManifestDumpCommand::ARG_PATH    = "path";
+
+void ManifestDumpCommand::Help(string& ret) {
+  ret.append("  ");
+  ret.append(ManifestDumpCommand::Name());
+  ret.append(" [--" + ARG_VERBOSE + "]");
+  ret.append(" [--" + ARG_PATH + "=<path_to_manifest_file>]");
+  ret.append("\n");
+}
+
+ManifestDumpCommand::ManifestDumpCommand(const vector<string>& params,
+      const map<string, string>& options, const vector<string>& flags) :
+    LDBCommand(options, flags, false,
+               BuildCmdLineOptions({ARG_VERBOSE,ARG_PATH})),
+    verbose_(false),
+    path_("")
+{
+  verbose_ = IsFlagPresent(flags, ARG_VERBOSE);
+
+  map<string, string>::const_iterator itr = options.find(ARG_PATH);
+  if (itr != options.end()) {
+    path_ = itr->second;
+    if (path_.empty()) {
+      exec_state_ = LDBCommandExecuteResult::FAILED("--path: missing pathname");
+    }
+  }
+}
+
+void ManifestDumpCommand::DoCommand() {
+
+  std::string manifestfile;
+
+  if (!path_.empty()) {
+    manifestfile = path_;
+  } else {
+    bool found = false;
+    // We need to find the manifest file by searching the directory
+    // containing the db for files of the form MANIFEST_[0-9]+
+    DIR* d = opendir(db_path_.c_str());
+    if (d == nullptr) {
+      exec_state_ = LDBCommandExecuteResult::FAILED(
+        db_path_ + " is not a directory");
+      return;
+    }
+    struct dirent* entry;
+    while ((entry = readdir(d)) != nullptr) {
+      unsigned int match;
+      unsigned long long num;
+      if (sscanf(entry->d_name, "MANIFEST-%llu%n", &num, &match)
+          && match == strlen(entry->d_name)) {
+        if (!found) {
+          manifestfile = db_path_ + "/" + std::string(entry->d_name);
+          found = true;
+        } else {
+          exec_state_ = LDBCommandExecuteResult::FAILED(
+            "Multiple MANIFEST files found; use --path to select one");
+          return;
+        }
+      }
+    }
+    closedir(d);
+  }
+
+  if (verbose_) {
+    printf("Processing Manifest file %s\n", manifestfile.c_str());
+  }
+
+  Options options;
+  EnvOptions sopt;
+  std::string file(manifestfile);
+  std::string dbname("dummy");
+  TableCache* tc = new TableCache(dbname, &options, sopt, 10);
+  const InternalKeyComparator* cmp =
+    new InternalKeyComparator(options.comparator);
+
+  VersionSet* versions = new VersionSet(dbname, &options, sopt, tc, cmp);
+  Status s = versions->DumpManifest(options, file, verbose_, is_key_hex_);
+  if (!s.ok()) {
+    printf("Error in processing file %s %s\n", manifestfile.c_str(),
+           s.ToString().c_str());
+  }
+  if (verbose_) {
+    printf("Processing Manifest file %s done\n", manifestfile.c_str());
+  }
+}
+
+// ----------------------------------------------------------------------------
+
+string ReadableTime(int unixtime) {
+  char time_buffer [80];
+  time_t rawtime = unixtime;
+  struct tm * timeinfo = localtime(&rawtime);
+  strftime(time_buffer, 80, "%c", timeinfo);
+  return string(time_buffer);
+}
+
+// This function only called when it's the sane case of >1 buckets in time-range
+// Also called only when timekv falls between ttl_start and ttl_end provided
+void IncBucketCounts(vector<uint64_t>& bucket_counts, int ttl_start,
+      int time_range, int bucket_size, int timekv, int num_buckets) {
+  assert(time_range > 0 && timekv >= ttl_start && bucket_size > 0 &&
+    timekv < (ttl_start + time_range) && num_buckets > 1);
+  int bucket = (timekv - ttl_start) / bucket_size;
+  bucket_counts[bucket]++;
+}
+
+void PrintBucketCounts(const vector<uint64_t>& bucket_counts, int ttl_start,
+      int ttl_end, int bucket_size, int num_buckets) {
+  int time_point = ttl_start;
+  for(int i = 0; i < num_buckets - 1; i++, time_point += bucket_size) {
+    fprintf(stdout, "Keys in range %s to %s : %lu\n",
+            ReadableTime(time_point).c_str(),
+            ReadableTime(time_point + bucket_size).c_str(), bucket_counts[i]);
+  }
+  fprintf(stdout, "Keys in range %s to %s : %lu\n",
+          ReadableTime(time_point).c_str(),
+          ReadableTime(ttl_end).c_str(), bucket_counts[num_buckets - 1]);
+}
+
 const string DBDumperCommand::ARG_COUNT_ONLY = "count_only";
 const string DBDumperCommand::ARG_STATS = "stats";
+const string DBDumperCommand::ARG_TTL_BUCKET = "bucket";
 
 DBDumperCommand::DBDumperCommand(const vector<string>& params,
       const map<string, string>& options, const vector<string>& flags) :
     LDBCommand(options, flags, true,
-               BuildCmdLineOptions({ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX,
-                                    ARG_FROM, ARG_TO, ARG_MAX_KEYS,
-                                    ARG_COUNT_ONLY, ARG_STATS})),
+               BuildCmdLineOptions({ARG_TTL, ARG_HEX, ARG_KEY_HEX,
+                                    ARG_VALUE_HEX, ARG_FROM, ARG_TO,
+                                    ARG_MAX_KEYS, ARG_COUNT_ONLY, ARG_STATS,
+                                    ARG_TTL_START, ARG_TTL_END,
+                                    ARG_TTL_BUCKET, ARG_TIMESTAMP})),
     null_from_(true),
     null_to_(true),
     max_keys_(-1),
@@ -426,10 +592,13 @@ DBDumperCommand::DBDumperCommand(const vector<string>& params,
   itr = options.find(ARG_MAX_KEYS);
   if (itr != options.end()) {
     try {
-      max_keys_ = boost::lexical_cast<int>(itr->second);
-    } catch( const boost::bad_lexical_cast & ) {
+      max_keys_ = stoi(itr->second);
+    } catch(const invalid_argument&) {
       exec_state_ = LDBCommandExecuteResult::FAILED(ARG_MAX_KEYS +
                         " has an invalid value");
+    } catch(const out_of_range&) {
+      exec_state_ = LDBCommandExecuteResult::FAILED(ARG_MAX_KEYS +
+                        " has a value out-of-range");
     }
   }
 
@@ -450,9 +619,14 @@ void DBDumperCommand::Help(string& ret) {
   ret.append("  ");
   ret.append(DBDumperCommand::Name());
   ret.append(HelpRangeCmdArgs());
+  ret.append(" [--" + ARG_TTL + "]");
   ret.append(" [--" + ARG_MAX_KEYS + "=<N>]");
+  ret.append(" [--" + ARG_TIMESTAMP + "]");
   ret.append(" [--" + ARG_COUNT_ONLY + "]");
   ret.append(" [--" + ARG_STATS + "]");
+  ret.append(" [--" + ARG_TTL_BUCKET + "=<N>]");
+  ret.append(" [--" + ARG_TTL_START + "=<N>:- is inclusive]");
+  ret.append(" [--" + ARG_TTL_END + "=<N>:- is exclusive]");
   ret.append("\n");
 }
 
@@ -470,8 +644,8 @@ void DBDumperCommand::DoCommand() {
   }
 
   // Setup key iterator
-  leveldb::Iterator* iter = db_->NewIterator(leveldb::ReadOptions());
-  leveldb::Status st = iter->status();
+  Iterator* iter = db_->NewIterator(ReadOptions());
+  Status st = iter->status();
   if (!st.ok()) {
     exec_state_ = LDBCommandExecuteResult::FAILED("Iterator error."
         + st.ToString());
@@ -484,25 +658,73 @@ void DBDumperCommand::DoCommand() {
   }
 
   int max_keys = max_keys_;
+  int ttl_start;
+  if (!ParseIntOption(option_map_, ARG_TTL_START, ttl_start, exec_state_)) {
+    ttl_start = DBWithTTL::kMinTimestamp; // TTL introduction time
+  }
+  int ttl_end;
+  if (!ParseIntOption(option_map_, ARG_TTL_END, ttl_end, exec_state_)) {
+    ttl_end = DBWithTTL::kMaxTimestamp; // Max time allowed by TTL feature
+  }
+  if (ttl_end < ttl_start) {
+    fprintf(stderr, "Error: End time can't be less than start time\n");
+    delete iter;
+    return;
+  }
+  int time_range = ttl_end - ttl_start;
+  int bucket_size;
+  if (!ParseIntOption(option_map_, ARG_TTL_BUCKET, bucket_size, exec_state_) ||
+      bucket_size <= 0) {
+    bucket_size = time_range; // Will have just 1 bucket by default
+  }
+  // At this point, bucket_size=0 => time_range=0
+  uint64_t num_buckets = (bucket_size >= time_range) ? 1 :
+    ((time_range + bucket_size - 1) / bucket_size);
+  vector<uint64_t> bucket_counts(num_buckets, 0);
+  if (is_db_ttl_ && !count_only_ && timestamp_) {
+    fprintf(stdout, "Dumping key-values from %s to %s\n",
+            ReadableTime(ttl_start).c_str(), ReadableTime(ttl_end).c_str());
+  }
+
   for (; iter->Valid(); iter->Next()) {
+    int rawtime = 0;
     // If end marker was specified, we stop before it
     if (!null_to_ && (iter->key().ToString() >= to_))
       break;
     // Terminate if maximum number of keys have been dumped
     if (max_keys == 0)
       break;
+    if (is_db_ttl_) {
+      TtlIterator* it_ttl = (TtlIterator*)iter;
+      rawtime = it_ttl->timestamp();
+      if (rawtime < ttl_start || rawtime >= ttl_end) {
+        continue;
+      }
+    }
     if (max_keys > 0) {
       --max_keys;
     }
+    if (is_db_ttl_ && num_buckets > 1) {
+      IncBucketCounts(bucket_counts, ttl_start, time_range, bucket_size,
+                      rawtime, num_buckets);
+    }
     ++count;
     if (!count_only_) {
+      if (is_db_ttl_ && timestamp_) {
+        fprintf(stdout, "%s ", ReadableTime(rawtime).c_str());
+      }
       string str = PrintKeyValue(iter->key().ToString(),
-                                      iter->value().ToString(),
-                                      is_key_hex_, is_value_hex_);
+                                 iter->value().ToString(), is_key_hex_,
+                                 is_value_hex_);
       fprintf(stdout, "%s\n", str.c_str());
     }
   }
-  fprintf(stdout, "Keys in range: %lld\n", (long long) count);
+  if (num_buckets > 1 && is_db_ttl_) {
+    PrintBucketCounts(bucket_counts, ttl_start, ttl_end, bucket_size,
+                      num_buckets);
+  } else {
+    fprintf(stdout, "Keys in range: %lld\n", (long long) count);
+  }
   // Clean up
   delete iter;
 }
@@ -519,7 +741,7 @@ ReduceDBLevelsCommand::ReduceDBLevelsCommand(const vector<string>& params,
     print_old_levels_(false) {
 
 
-  ParseIntOption(options_, ARG_NEW_LEVELS, new_levels_, exec_state_);
+  ParseIntOption(option_map_, ARG_NEW_LEVELS, new_levels_, exec_state_);
   print_old_levels_ = IsFlagPresent(flags, ARG_PRINT_OLD_LEVELS);
 
   if(new_levels_ <= 0) {
@@ -533,7 +755,7 @@ vector<string> ReduceDBLevelsCommand::PrepareArgs(const string& db_path,
   vector<string> ret;
   ret.push_back("reduce_levels");
   ret.push_back("--" + ARG_DB + "=" + db_path);
-  ret.push_back("--" + ARG_NEW_LEVELS + "=" + std::to_string(new_levels));
+  ret.push_back("--" + ARG_NEW_LEVELS + "=" + to_string(new_levels));
   if(print_old_level) {
     ret.push_back("--" + ARG_PRINT_OLD_LEVELS);
   }
@@ -548,9 +770,10 @@ void ReduceDBLevelsCommand::Help(string& ret) {
   ret.append("\n");
 }
 
-leveldb::Options ReduceDBLevelsCommand::PrepareOptionsForOpenDB() {
-  leveldb::Options opt = LDBCommand::PrepareOptionsForOpenDB();
+Options ReduceDBLevelsCommand::PrepareOptionsForOpenDB() {
+  Options opt = LDBCommand::PrepareOptionsForOpenDB();
   opt.num_levels = old_levels_;
+  opt.max_bytes_for_level_multiplier_additional.resize(opt.num_levels, 1);
   // Disable size compaction
   opt.max_bytes_for_level_base = 1UL << 50;
   opt.max_bytes_for_level_multiplier = 1;
@@ -558,11 +781,12 @@ leveldb::Options ReduceDBLevelsCommand::PrepareOptionsForOpenDB() {
   return opt;
 }
 
-Status ReduceDBLevelsCommand::GetOldNumOfLevels(leveldb::Options& opt,
+Status ReduceDBLevelsCommand::GetOldNumOfLevels(Options& opt,
     int* levels) {
-  TableCache tc(db_path_, &opt, 10);
+  EnvOptions soptions;
+  TableCache tc(db_path_, &opt, soptions, 10);
   const InternalKeyComparator cmp(opt.comparator);
-  VersionSet versions(db_path_, &opt, &tc, &cmp);
+  VersionSet versions(db_path_, &opt, soptions, &tc, &cmp);
   // We rely the VersionSet::Recover to tell us the internal data structures
   // in the db. And the Recover() should never do any change
   // (like LogAndApply) to the manifest file.
@@ -588,8 +812,8 @@ void ReduceDBLevelsCommand::DoCommand() {
     return;
   }
 
-  leveldb::Status st;
-  leveldb::Options opt = PrepareOptionsForOpenDB();
+  Status st;
+  Options opt = PrepareOptionsForOpenDB();
   int old_level_num = -1;
   st = GetOldNumOfLevels(opt, &old_level_num);
   if (!st.ok()) {
@@ -616,9 +840,10 @@ void ReduceDBLevelsCommand::DoCommand() {
   db_->CompactRange(nullptr, nullptr);
   CloseDB();
 
-  TableCache tc(db_path_, &opt, 10);
+  EnvOptions soptions;
+  TableCache tc(db_path_, &opt, soptions, 10);
   const InternalKeyComparator cmp(opt.comparator);
-  VersionSet versions(db_path_, &opt, &tc, &cmp);
+  VersionSet versions(db_path_, &opt, soptions, &tc, &cmp);
   // We rely the VersionSet::Recover to tell us the internal data structures
   // in the db. And the Recover() should never do any change (like LogAndApply)
   // to the manifest file.
@@ -658,8 +883,8 @@ class InMemoryHandler : public WriteBatch::Handler {
   }
 
  private:
-  std::map<string, string> putMap_;
-  std::vector<string> deleteList_;
+  map<string, string> putMap_;
+  vector<string> deleteList_;
 };
 
 const string WALDumperCommand::ARG_WAL_FILE = "walfile";
@@ -701,29 +926,30 @@ void WALDumperCommand::Help(string& ret) {
 void WALDumperCommand::DoCommand() {
   struct StdErrReporter : public log::Reader::Reporter {
     virtual void Corruption(size_t bytes, const Status& s) {
-      std::cerr<<"Corruption detected in log file "<<s.ToString()<<"\n";
+      cerr<<"Corruption detected in log file "<<s.ToString()<<"\n";
     }
   };
 
   unique_ptr<SequentialFile> file;
   Env* env_ = Env::Default();
-  Status status = env_->NewSequentialFile(wal_file_, &file);
+  EnvOptions soptions;
+  Status status = env_->NewSequentialFile(wal_file_, &file, soptions);
   if (!status.ok()) {
     exec_state_ = LDBCommandExecuteResult::FAILED("Failed to open WAL file " +
       status.ToString());
   } else {
     StdErrReporter reporter;
-    log::Reader reader(std::move(file), &reporter, true, 0);
+    log::Reader reader(move(file), &reporter, true, 0);
     string scratch;
     WriteBatch batch;
     Slice record;
-    std::stringstream row;
+    stringstream row;
     if (print_header_) {
-      std::cout<<"Sequence,Count,ByteSize,Physical Offset,Key(s)";
+      cout<<"Sequence,Count,ByteSize,Physical Offset,Key(s)";
       if (print_values_) {
-        std::cout << " : value ";
+        cout << " : value ";
       }
-      std::cout << "\n";
+      cout << "\n";
     }
     while(reader.ReadRecord(&record, &scratch)) {
       row.str("");
@@ -741,8 +967,8 @@ void WALDumperCommand::DoCommand() {
         row << "PUT : ";
         if (print_values_) {
           for (auto& kv : handler.PutMap()) {
-            std::string k = StringToHex(kv.first);
-            std::string v = StringToHex(kv.second);
+            string k = StringToHex(kv.first);
+            string v = StringToHex(kv.second);
             row << k << " : ";
             row << v << " ";
           }
@@ -758,7 +984,7 @@ void WALDumperCommand::DoCommand() {
         }
         row<<"\n";
       }
-      std::cout<<row.str();
+      cout<<row.str();
     }
   }
 }
@@ -766,8 +992,9 @@ void WALDumperCommand::DoCommand() {
 
 GetCommand::GetCommand(const vector<string>& params,
       const map<string, string>& options, const vector<string>& flags) :
-  LDBCommand(options, flags, true,
-             BuildCmdLineOptions({ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX})) {
+  LDBCommand(options, flags, true, BuildCmdLineOptions({ARG_TTL, ARG_HEX,
+                                                        ARG_KEY_HEX,
+                                                        ARG_VALUE_HEX})) {
 
   if (params.size() != 1) {
     exec_state_ = LDBCommandExecuteResult::FAILED(
@@ -785,12 +1012,13 @@ void GetCommand::Help(string& ret) {
   ret.append("  ");
   ret.append(GetCommand::Name());
   ret.append(" <key>");
+  ret.append(" [--" + ARG_TTL + "]");
   ret.append("\n");
 }
 
 void GetCommand::DoCommand() {
   string value;
-  leveldb::Status st = db_->Get(leveldb::ReadOptions(), key_, &value);
+  Status st = db_->Get(ReadOptions(), key_, &value);
   if (st.ok()) {
     fprintf(stdout, "%s\n",
               (is_value_hex_ ? StringToHex(value) : value).c_str());
@@ -837,8 +1065,8 @@ void ApproxSizeCommand::Help(string& ret) {
 
 void ApproxSizeCommand::DoCommand() {
 
-  leveldb::Range ranges[1];
-  ranges[0] = leveldb::Range(start_key_, end_key_);
+  Range ranges[1];
+  ranges[0] = Range(start_key_, end_key_);
   uint64_t sizes[1];
   db_->GetApproximateSizes(ranges, 1, sizes);
   fprintf(stdout, "%ld\n", sizes[0]);
@@ -854,7 +1082,7 @@ void ApproxSizeCommand::DoCommand() {
 BatchPutCommand::BatchPutCommand(const vector<string>& params,
       const map<string, string>& options, const vector<string>& flags) :
   LDBCommand(options, flags, false,
-             BuildCmdLineOptions({ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX,
+             BuildCmdLineOptions({ARG_TTL, ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX,
                                   ARG_CREATE_IF_MISSING})) {
 
   if (params.size() < 2) {
@@ -867,7 +1095,7 @@ BatchPutCommand::BatchPutCommand(const vector<string>& params,
     for (size_t i = 0; i < params.size(); i += 2) {
       string key = params.at(i);
       string value = params.at(i+1);
-      key_values_.push_back(std::pair<string, string>(
+      key_values_.push_back(pair<string, string>(
                     is_key_hex_ ? HexToString(key) : key,
                     is_value_hex_ ? HexToString(value) : value));
     }
@@ -878,17 +1106,18 @@ void BatchPutCommand::Help(string& ret) {
   ret.append("  ");
   ret.append(BatchPutCommand::Name());
   ret.append(" <key> <value> [<key> <value>] [..]");
+  ret.append(" [--" + ARG_TTL + "]");
   ret.append("\n");
 }
 
 void BatchPutCommand::DoCommand() {
-  leveldb::WriteBatch batch;
+  WriteBatch batch;
 
-  for (vector<std::pair<string, string>>::const_iterator itr
+  for (vector<pair<string, string>>::const_iterator itr
         = key_values_.begin(); itr != key_values_.end(); itr++) {
       batch.Put(itr->first, itr->second);
   }
-  leveldb::Status st = db_->Write(leveldb::WriteOptions(), &batch);
+  Status st = db_->Write(WriteOptions(), &batch);
   if (st.ok()) {
     fprintf(stdout, "OK\n");
   } else {
@@ -896,8 +1125,8 @@ void BatchPutCommand::DoCommand() {
   }
 }
 
-leveldb::Options BatchPutCommand::PrepareOptionsForOpenDB() {
-  leveldb::Options opt = LDBCommand::PrepareOptionsForOpenDB();
+Options BatchPutCommand::PrepareOptionsForOpenDB() {
+  Options opt = LDBCommand::PrepareOptionsForOpenDB();
   opt.create_if_missing = IsFlagPresent(flags_, ARG_CREATE_IF_MISSING);
   return opt;
 }
@@ -906,8 +1135,9 @@ leveldb::Options BatchPutCommand::PrepareOptionsForOpenDB() {
 ScanCommand::ScanCommand(const vector<string>& params,
       const map<string, string>& options, const vector<string>& flags) :
     LDBCommand(options, flags, true,
-               BuildCmdLineOptions({ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX,
-                                    ARG_FROM, ARG_TO, ARG_MAX_KEYS})),
+               BuildCmdLineOptions({ARG_TTL, ARG_HEX, ARG_KEY_HEX, ARG_TO,
+                                    ARG_VALUE_HEX, ARG_FROM, ARG_TIMESTAMP,
+                                    ARG_MAX_KEYS, ARG_TTL_START, ARG_TTL_END})),
     start_key_specified_(false),
     end_key_specified_(false),
     max_keys_scanned_(-1) {
@@ -932,10 +1162,13 @@ ScanCommand::ScanCommand(const vector<string>& params,
   itr = options.find(ARG_MAX_KEYS);
   if (itr != options.end()) {
     try {
-      max_keys_scanned_ = boost::lexical_cast< int >(itr->second);
-    } catch( const boost::bad_lexical_cast & ) {
+      max_keys_scanned_ = stoi(itr->second);
+    } catch(const invalid_argument&) {
       exec_state_ = LDBCommandExecuteResult::FAILED(ARG_MAX_KEYS +
                         " has an invalid value");
+    } catch(const out_of_range&) {
+      exec_state_ = LDBCommandExecuteResult::FAILED(ARG_MAX_KEYS +
+                        " has a value out-of-range");
     }
   }
 }
@@ -944,23 +1177,54 @@ void ScanCommand::Help(string& ret) {
   ret.append("  ");
   ret.append(ScanCommand::Name());
   ret.append(HelpRangeCmdArgs());
-  ret.append("--" + ARG_MAX_KEYS + "=N] ");
+  ret.append(" [--" + ARG_TTL + "]");
+  ret.append(" [--" + ARG_TIMESTAMP + "]");
+  ret.append(" [--" + ARG_MAX_KEYS + "=<N>q] ");
+  ret.append(" [--" + ARG_TTL_START + "=<N>:- is inclusive]");
+  ret.append(" [--" + ARG_TTL_END + "=<N>:- is exclusive]");
   ret.append("\n");
 }
 
 void ScanCommand::DoCommand() {
 
   int num_keys_scanned = 0;
-  Iterator* it = db_->NewIterator(leveldb::ReadOptions());
+  Iterator* it = db_->NewIterator(ReadOptions());
   if (start_key_specified_) {
     it->Seek(start_key_);
   } else {
     it->SeekToFirst();
   }
+  int ttl_start;
+  if (!ParseIntOption(option_map_, ARG_TTL_START, ttl_start, exec_state_)) {
+    ttl_start = DBWithTTL::kMinTimestamp; // TTL introduction time
+  }
+  int ttl_end;
+  if (!ParseIntOption(option_map_, ARG_TTL_END, ttl_end, exec_state_)) {
+    ttl_end = DBWithTTL::kMaxTimestamp; // Max time allowed by TTL feature
+  }
+  if (ttl_end < ttl_start) {
+    fprintf(stderr, "Error: End time can't be less than start time\n");
+    delete it;
+    return;
+  }
+  if (is_db_ttl_ && timestamp_) {
+    fprintf(stdout, "Scanning key-values from %s to %s\n",
+            ReadableTime(ttl_start).c_str(), ReadableTime(ttl_end).c_str());
+  }
   for ( ;
         it->Valid() && (!end_key_specified_ || it->key().ToString() < end_key_);
         it->Next()) {
     string key = it->key().ToString();
+    if (is_db_ttl_) {
+      TtlIterator* it_ttl = (TtlIterator*)it;
+      int rawtime = it_ttl->timestamp();
+      if (rawtime < ttl_start || rawtime >= ttl_end) {
+        continue;
+      }
+      if (timestamp_) {
+        fprintf(stdout, "%s ", ReadableTime(rawtime).c_str());
+      }
+    }
     string value = it->value().ToString();
     fprintf(stdout, "%s : %s\n",
           (is_key_hex_ ? StringToHex(key) : key).c_str(),
@@ -1001,7 +1265,7 @@ void DeleteCommand::Help(string& ret) {
 }
 
 void DeleteCommand::DoCommand() {
-  leveldb::Status st = db_->Delete(leveldb::WriteOptions(), key_);
+  Status st = db_->Delete(WriteOptions(), key_);
   if (st.ok()) {
     fprintf(stdout, "OK\n");
   } else {
@@ -1013,7 +1277,7 @@ void DeleteCommand::DoCommand() {
 PutCommand::PutCommand(const vector<string>& params,
       const map<string, string>& options, const vector<string>& flags) :
   LDBCommand(options, flags, false,
-             BuildCmdLineOptions({ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX,
+             BuildCmdLineOptions({ARG_TTL, ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX,
                                   ARG_CREATE_IF_MISSING})) {
 
   if (params.size() != 2) {
@@ -1037,11 +1301,12 @@ void PutCommand::Help(string& ret) {
   ret.append("  ");
   ret.append(PutCommand::Name());
   ret.append(" <key> <value> ");
+  ret.append(" [--" + ARG_TTL + "]");
   ret.append("\n");
 }
 
 void PutCommand::DoCommand() {
-  leveldb::Status st = db_->Put(leveldb::WriteOptions(), key_, value_);
+  Status st = db_->Put(WriteOptions(), key_, value_);
   if (st.ok()) {
     fprintf(stdout, "OK\n");
   } else {
@@ -1049,8 +1314,8 @@ void PutCommand::DoCommand() {
   }
 }
 
-leveldb::Options PutCommand::PrepareOptionsForOpenDB() {
-  leveldb::Options opt = LDBCommand::PrepareOptionsForOpenDB();
+Options PutCommand::PrepareOptionsForOpenDB() {
+  Options opt = LDBCommand::PrepareOptionsForOpenDB();
   opt.create_if_missing = IsFlagPresent(flags_, ARG_CREATE_IF_MISSING);
   return opt;
 }
@@ -1064,13 +1329,15 @@ const char* DBQuerierCommand::DELETE_CMD = "delete";
 DBQuerierCommand::DBQuerierCommand(const vector<string>& params,
     const map<string, string>& options, const vector<string>& flags) :
   LDBCommand(options, flags, false,
-             BuildCmdLineOptions({ARG_HEX, ARG_KEY_HEX, ARG_VALUE_HEX})) {
+             BuildCmdLineOptions({ARG_TTL, ARG_HEX, ARG_KEY_HEX,
+                                  ARG_VALUE_HEX})) {
 
 }
 
 void DBQuerierCommand::Help(string& ret) {
   ret.append("  ");
   ret.append(DBQuerierCommand::Name());
+  ret.append(" [--" + ARG_TTL + "]");
   ret.append("\n");
   ret.append("    Starts a REPL shell.  Type help for list of available "
              "commands.");
@@ -1082,13 +1349,13 @@ void DBQuerierCommand::DoCommand() {
     return;
   }
 
-  leveldb::ReadOptions read_options;
-  leveldb::WriteOptions write_options;
+  ReadOptions read_options;
+  WriteOptions write_options;
 
   string line;
   string key;
   string value;
-  while (getline(std::cin, line, '\n')) {
+  while (getline(cin, line, '\n')) {
 
     // Parse line into vector<string>
     vector<string> tokens;

util/ldb_cmd.h

@@ -2,8 +2,8 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
-#ifndef LEVELDB_UTIL_LDB_CMD_H_
-#define LEVELDB_UTIL_LDB_CMD_H_
+#ifndef STORAGE_LEVELDB_UTIL_LDB_CMD_H_
+#define STORAGE_LEVELDB_UTIL_LDB_CMD_H_
 
 #include <string>
 #include <iostream>
@@ -12,17 +12,16 @@
 #include <algorithm>
 #include <stdio.h>
 
-#include <boost/algorithm/string/case_conv.hpp>
-#include <boost/algorithm/string/predicate.hpp>
-
-#include "leveldb/db.h"
+#include "db/version_set.h"
 #include "leveldb/env.h"
 #include "leveldb/options.h"
 #include "leveldb/iterator.h"
 #include "leveldb/slice.h"
-#include "db/version_set.h"
 #include "util/logging.h"
 #include "util/ldb_cmd_execute_result.h"
+#include "util/string_util.h"
+#include "utilities/utility_db.h"
+#include "utilities/ttl/db_ttl.h"
 
 using std::string;
 using std::map;
@@ -39,6 +38,10 @@ public:
   static const string ARG_HEX;
   static const string ARG_KEY_HEX;
   static const string ARG_VALUE_HEX;
+  static const string ARG_TTL;
+  static const string ARG_TTL_START;
+  static const string ARG_TTL_END;
+  static const string ARG_TIMESTAMP;
   static const string ARG_FROM;
   static const string ARG_TO;
   static const string ARG_MAX_KEYS;
@@ -50,11 +53,24 @@ public:
   static const string ARG_FILE_SIZE;
   static const string ARG_CREATE_IF_MISSING;
 
-  static LDBCommand* InitFromCmdLineArgs(const vector<string>& args);
-  static LDBCommand* InitFromCmdLineArgs(int argc, char** argv);
+  static LDBCommand* InitFromCmdLineArgs(
+    const vector<string>& args,
+    Options options = Options()
+  );
+
+  static LDBCommand* InitFromCmdLineArgs(
+    int argc,
+    char** argv,
+    Options options = Options()
+  );
+
   bool ValidateCmdLineOptions();
 
-  virtual leveldb::Options PrepareOptionsForOpenDB();
+  virtual Options PrepareOptionsForOpenDB();
+
+  virtual void SetOptions(Options options) {
+    options_ = options;
+  }
 
   virtual bool NoDBOpen() {
     return false;
@@ -102,7 +118,7 @@ public:
 
   static string HexToString(const string& str) {
     string parsed;
-    if (!boost::starts_with(str, "0x")) {
+    if (str[0] != '0' || str[1] != 'x') {
       fprintf(stderr, "Invalid hex input %s.  Must start with 0x\n",
               str.c_str());
       throw "Invalid hex input";
@@ -132,8 +148,8 @@ public:
 protected:
 
   LDBCommandExecuteResult exec_state_;
-  std::string db_path_;
-  leveldb::DB* db_;
+  string db_path_;
+  DB* db_;
 
   /**
    * true implies that this command can work if the db is opened in read-only
@@ -147,10 +163,16 @@ protected:
   /** If true, the value is input/output as hex in get/put/scan/delete etc. */
   bool is_value_hex_;
 
+  /** If true, the value is treated as timestamp suffixed */
+  bool is_db_ttl_;
+
+  // If true, the kvs are output with their insert/modify timestamp in a ttl db
+  bool timestamp_;
+
   /**
    * Map of options passed on the command-line.
    */
-  const map<string, string> options_;
+  const map<string, string> option_map_;
 
   /**
    * Flags passed on the command-line.
@@ -169,7 +191,9 @@ protected:
       is_read_only_(is_read_only),
       is_key_hex_(false),
       is_value_hex_(false),
-      options_(options),
+      is_db_ttl_(false),
+      timestamp_(false),
+      option_map_(options),
       flags_(flags),
       valid_cmd_line_options_(valid_cmd_line_options) {
 
@@ -180,22 +204,30 @@ protected:
 
     is_key_hex_ = IsKeyHex(options, flags);
     is_value_hex_ = IsValueHex(options, flags);
+    is_db_ttl_ = IsFlagPresent(flags, ARG_TTL);
+    timestamp_ = IsFlagPresent(flags, ARG_TIMESTAMP);
   }
 
   void OpenDB() {
-    leveldb::Options opt = PrepareOptionsForOpenDB();
+    Options opt = PrepareOptionsForOpenDB();
     if (!exec_state_.IsNotStarted()) {
       return;
     }
     // Open the DB.
-    leveldb::Status st;
-    if (is_read_only_) {
-      st = leveldb::DB::OpenForReadOnly(opt, db_path_, &db_);
+    Status st;
+    if (is_db_ttl_) {
+      if (is_read_only_) {
+        st = UtilityDB::OpenTtlDB(opt, db_path_, &db_, 0, true);
+      } else {
+        st = UtilityDB::OpenTtlDB(opt, db_path_, &db_);
+      }
+    } else if (is_read_only_) {
+      st = DB::OpenForReadOnly(opt, db_path_, &db_);
     } else {
-      st = leveldb::DB::Open(opt, db_path_, &db_);
+      st = DB::Open(opt, db_path_, &db_);
     }
     if (!st.ok()) {
-      std::string msg = st.ToString();
+      string msg = st.ToString();
       exec_state_ = LDBCommandExecuteResult::FAILED(msg);
     }
   }
@@ -254,6 +286,8 @@ protected:
 
 private:
 
+  Options options_;
+
   /**
    * Interpret command line options and flags to determine if the key
    * should be input/output in hex.
@@ -301,7 +335,7 @@ private:
    * Otherwise an exception is thrown.
    */
   bool StringToBool(string val) {
-    boost::algorithm::to_lower(val);
+    std::transform(val.begin(), val.end(), val.begin(), ::tolower);
     if (val == "true") {
       return true;
     } else if (val == "false") {
@@ -311,6 +345,13 @@ private:
     }
   }
 
+  static LDBCommand* SelectCommand(
+    const string& cmd,
+    vector<string>& cmdParams,
+    map<string, string>& option_map,
+    vector<string>& flags
+  );
+
 };
 
 class CompactorCommand: public LDBCommand {
@@ -353,6 +394,7 @@ private:
 
   static const string ARG_COUNT_ONLY;
   static const string ARG_STATS;
+  static const string ARG_TTL_BUCKET;
 };
 
 class DBLoaderCommand: public LDBCommand {
@@ -367,7 +409,7 @@ public:
   static void Help(string& ret);
   virtual void DoCommand();
 
-  virtual leveldb::Options PrepareOptionsForOpenDB();
+  virtual Options PrepareOptionsForOpenDB();
 
 private:
   bool create_if_missing_;
@@ -380,6 +422,28 @@ private:
   static const string ARG_COMPACT;
 };
 
+class ManifestDumpCommand: public LDBCommand {
+public:
+  static string Name() { return "manifest_dump"; }
+
+  ManifestDumpCommand(const vector<string>& params,
+      const map<string, string>& options, const vector<string>& flags);
+
+  static void Help(string& ret);
+  virtual void DoCommand();
+
+  virtual bool NoDBOpen() {
+    return true;
+  }
+
+private:
+  bool verbose_;
+  string path_;
+
+  static const string ARG_VERBOSE;
+  static const string ARG_PATH;
+};
+
 class ReduceDBLevelsCommand : public LDBCommand {
 public:
   static string Name() { return "reduce_levels"; }
@@ -387,7 +451,7 @@ public:
   ReduceDBLevelsCommand(const vector<string>& params,
       const map<string, string>& options, const vector<string>& flags);
 
-  virtual leveldb::Options PrepareOptionsForOpenDB();
+  virtual Options PrepareOptionsForOpenDB();
 
   virtual void DoCommand();
 
@@ -408,7 +472,7 @@ private:
   static const string ARG_NEW_LEVELS;
   static const string ARG_PRINT_OLD_LEVELS;
 
-  Status GetOldNumOfLevels(leveldb::Options& opt, int* levels);
+  Status GetOldNumOfLevels(Options& opt, int* levels);
 };
 
 class WALDumperCommand : public LDBCommand {
@@ -478,7 +542,7 @@ public:
 
   static void Help(string& ret);
 
-  virtual leveldb::Options PrepareOptionsForOpenDB();
+  virtual Options PrepareOptionsForOpenDB();
 
 private:
   /**
@@ -532,7 +596,7 @@ public:
 
   static void Help(string& ret);
 
-  virtual leveldb::Options PrepareOptionsForOpenDB();
+  virtual Options PrepareOptionsForOpenDB();
 
 private:
   string key_;
@@ -561,5 +625,5 @@ private:
   static const char* DELETE_CMD;
 };
 
-}
-#endif
+} // namespace leveldb
+#endif // STORAGE_LEVELDB_UTIL_LDB_CMD_H_

util/ldb_tool.cc

@@ -0,0 +1,98 @@
+// Copyright 2008-present Facebook. All Rights Reserved.
+
+#include "leveldb/ldb_tool.h"
+#include "util/ldb_cmd.h"
+
+namespace leveldb {
+
+class LDBCommandRunner {
+public:
+
+  static void PrintHelp(const char* exec_name) {
+    string ret;
+
+    ret.append("ldb - LevelDB Tool");
+    ret.append("\n\n");
+    ret.append("commands MUST specify --" + LDBCommand::ARG_DB +
+        "=<full_path_to_db_directory> when necessary\n");
+    ret.append("\n");
+    ret.append("The following optional parameters control if keys/values are "
+        "input/output as hex or as plain strings:\n");
+    ret.append("  --" + LDBCommand::ARG_KEY_HEX +
+        " : Keys are input/output as hex\n");
+    ret.append("  --" + LDBCommand::ARG_VALUE_HEX +
+        " : Values are input/output as hex\n");
+    ret.append("  --" + LDBCommand::ARG_HEX +
+        " : Both keys and values are input/output as hex\n");
+    ret.append("\n");
+
+    ret.append("The following optional parameters control the database "
+        "internals:\n");
+    ret.append("  --" + LDBCommand::ARG_TTL +
+        " with 'put','get','scan','dump','query','batchput'"
+        " : DB supports ttl and value is internally timestamp-suffixed\n");
+    ret.append("  --" + LDBCommand::ARG_BLOOM_BITS + "=<int,e.g.:14>\n");
+    ret.append("  --" + LDBCommand::ARG_COMPRESSION_TYPE +
+        "=<no|snappy|zlib|bzip2>\n");
+    ret.append("  --" + LDBCommand::ARG_BLOCK_SIZE +
+        "=<block_size_in_bytes>\n");
+    ret.append("  --" + LDBCommand::ARG_AUTO_COMPACTION + "=<true|false>\n");
+    ret.append("  --" + LDBCommand::ARG_WRITE_BUFFER_SIZE +
+        "=<int,e.g.:4194304>\n");
+    ret.append("  --" + LDBCommand::ARG_FILE_SIZE + "=<int,e.g.:2097152>\n");
+
+    ret.append("\n\n");
+    ret.append("Data Access Commands:\n");
+    PutCommand::Help(ret);
+    GetCommand::Help(ret);
+    BatchPutCommand::Help(ret);
+    ScanCommand::Help(ret);
+    DeleteCommand::Help(ret);
+    DBQuerierCommand::Help(ret);
+    ApproxSizeCommand::Help(ret);
+
+    ret.append("\n\n");
+    ret.append("Admin Commands:\n");
+    WALDumperCommand::Help(ret);
+    CompactorCommand::Help(ret);
+    ReduceDBLevelsCommand::Help(ret);
+    DBDumperCommand::Help(ret);
+    DBLoaderCommand::Help(ret);
+    ManifestDumpCommand::Help(ret);
+
+    fprintf(stderr, "%s\n", ret.c_str());
+  }
+
+  static void RunCommand(int argc, char** argv, Options options) {
+    if (argc <= 2) {
+      PrintHelp(argv[0]);
+      exit(1);
+    }
+
+    LDBCommand* cmdObj = LDBCommand::InitFromCmdLineArgs(argc, argv, options);
+    if (cmdObj == nullptr) {
+      fprintf(stderr, "Unknown command\n");
+      PrintHelp(argv[0]);
+      exit(1);
+    }
+
+    if (!cmdObj->ValidateCmdLineOptions()) {
+      exit(1);
+    }
+
+    cmdObj->Run();
+    LDBCommandExecuteResult ret = cmdObj->GetExecuteState();
+    fprintf(stderr, "%s\n", ret.ToString().c_str());
+    delete cmdObj;
+
+    exit(ret.IsFailed());
+  }
+
+};
+
+
+void LDBTool::Run(int argc, char** argv, Options options) {
+  LDBCommandRunner::RunCommand(argc, argv, options);
+}
+} // namespace leveldb
+

util/manual_compaction_test.cc

@@ -0,0 +1,88 @@
+// Test for issue 178: a manual compaction causes deleted data to reappear.
+#include <iostream>
+#include <sstream>
+#include <cstdlib>
+
+#include "leveldb/db.h"
+#include "leveldb/write_batch.h"
+#include "util/testharness.h"
+
+namespace {
+
+const int kNumKeys = 1100000;
+
+std::string Key1(int i) {
+  char buf[100];
+  snprintf(buf, sizeof(buf), "my_key_%d", i);
+  return buf;
+}
+
+std::string Key2(int i) {
+  return Key1(i) + "_xxx";
+}
+
+class ManualCompactionTest { };
+
+TEST(ManualCompactionTest, Test) {
+  // Get rid of any state from an old run.
+  std::string dbpath = leveldb::test::TmpDir() + "/leveldb_cbug_test";
+  DestroyDB(dbpath, leveldb::Options());
+
+  // Open database.  Disable compression since it affects the creation
+  // of layers and the code below is trying to test against a very
+  // specific scenario.
+  leveldb::DB* db;
+  leveldb::Options db_options;
+  db_options.create_if_missing = true;
+  db_options.compression = leveldb::kNoCompression;
+  ASSERT_OK(leveldb::DB::Open(db_options, dbpath, &db));
+
+  // create first key range
+  leveldb::WriteBatch batch;
+  for (int i = 0; i < kNumKeys; i++) {
+    batch.Put(Key1(i), "value for range 1 key");
+  }
+  ASSERT_OK(db->Write(leveldb::WriteOptions(), &batch));
+
+  // create second key range
+  batch.Clear();
+  for (int i = 0; i < kNumKeys; i++) {
+    batch.Put(Key2(i), "value for range 2 key");
+  }
+  ASSERT_OK(db->Write(leveldb::WriteOptions(), &batch));
+
+  // delete second key range
+  batch.Clear();
+  for (int i = 0; i < kNumKeys; i++) {
+    batch.Delete(Key2(i));
+  }
+  ASSERT_OK(db->Write(leveldb::WriteOptions(), &batch));
+
+  // compact database
+  std::string start_key = Key1(0);
+  std::string end_key = Key1(kNumKeys - 1);
+  leveldb::Slice least(start_key.data(), start_key.size());
+  leveldb::Slice greatest(end_key.data(), end_key.size());
+
+  // commenting out the line below causes the example to work correctly
+  db->CompactRange(&least, &greatest);
+
+  // count the keys
+  leveldb::Iterator* iter = db->NewIterator(leveldb::ReadOptions());
+  int num_keys = 0;
+  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
+    num_keys++;
+  }
+  delete iter;
+  ASSERT_EQ(kNumKeys, num_keys) << "Bad number of keys";
+
+  // close database
+  delete db;
+  DestroyDB(dbpath, leveldb::Options());
+}
+
+}  // anonymous namespace
+
+int main(int argc, char** argv) {
+  return leveldb::test::RunAllTests();
+}

util/options.cc

@@ -6,15 +6,19 @@
 
 #include <limits>
 
+#include "leveldb/cache.h"
+#include "leveldb/compaction_filter.h"
 #include "leveldb/comparator.h"
 #include "leveldb/env.h"
 #include "leveldb/filter_policy.h"
-#include "leveldb/cache.h"
+#include "leveldb/merge_operator.h"
 
 namespace leveldb {
 
 Options::Options()
     : comparator(BytewiseComparator()),
+      merge_operator(nullptr),
+      compaction_filter(nullptr),
       create_if_missing(false),
       error_if_exists(false),
       paranoid_checks(false),
@@ -22,6 +26,7 @@ Options::Options()
       info_log(nullptr),
       write_buffer_size(4<<20),
       max_write_buffer_number(2),
+      min_write_buffer_number_to_merge(1),
       max_open_files(1000),
       block_size(4096),
       block_restart_interval(16),
@@ -36,10 +41,10 @@ Options::Options()
       target_file_size_multiplier(1),
       max_bytes_for_level_base(10 * 1048576),
       max_bytes_for_level_multiplier(10),
+      max_bytes_for_level_multiplier_additional(num_levels, 1),
       expanded_compaction_factor(25),
       source_compaction_factor(1),
       max_grandparent_overlap_factor(10),
-      statistics(nullptr),
       disableDataSync(false),
       use_fsync(false),
       db_stats_log_interval(1800),
@@ -55,20 +60,37 @@ Options::Options()
       max_manifest_file_size(std::numeric_limits<uint64_t>::max()),
       no_block_cache(false),
       table_cache_numshardbits(4),
-      compaction_filter_args(nullptr),
-      CompactionFilter(nullptr),
       disable_auto_compactions(false),
       WAL_ttl_seconds(0),
       manifest_preallocation_size(4 * 1024 * 1024),
-      purge_redundant_kvs_while_flush(true) {
-
+      purge_redundant_kvs_while_flush(true),
+      allow_os_buffer(true),
+      allow_readahead(true),
+      allow_readahead_compactions(true),
+      allow_mmap_reads(false),
+      allow_mmap_writes(true),
+      is_fd_close_on_exec(true),
+      skip_log_error_on_recovery(false),
+      stats_dump_period_sec(3600),
+      block_size_deviation (10),
+      advise_random_on_open(true),
+      access_hint_on_compaction_start(NORMAL),
+      use_adaptive_mutex(false),
+      bytes_per_sync(0) {
 }
 
+static const char* const access_hints[] = {
+  "NONE", "NORMAL", "SEQUENTIAL", "WILLNEED"
+};
+
 void
 Options::Dump(Logger* log) const
 {
     Log(log,"              Options.comparator: %s", comparator->Name());
-    Log(log,"       Options.create_if_missing: %d", create_if_missing);
+    Log(log,"          Options.merge_operator: %s",
+        merge_operator? merge_operator->Name() : "None");
+    Log(log,"       Options.compaction_filter: %s",
+        compaction_filter? compaction_filter->Name() : "None");
     Log(log,"         Options.error_if_exists: %d", error_if_exists);
     Log(log,"         Options.paranoid_checks: %d", paranoid_checks);
     Log(log,"                     Options.env: %p", env);
@@ -84,7 +106,7 @@ Options::Dump(Logger* log) const
     Log(log,"              Options.block_size: %zd", block_size);
     Log(log,"  Options.block_restart_interval: %d", block_restart_interval);
     if (!compression_per_level.empty()) {
-      for (int i = 0; i < compression_per_level.size(); i++) {
+      for (unsigned int i = 0; i < compression_per_level.size(); i++) {
           Log(log,"       Options.compression[%d]: %d",
               i, compression_per_level[i]);
        }
@@ -103,6 +125,14 @@ Options::Dump(Logger* log) const
     Log(log,"     Options.keep_log_file_num: %ld", keep_log_file_num);
     Log(log," Options.db_stats_log_interval: %d",
         db_stats_log_interval);
+    Log(log,"       Options.allow_os_buffer: %d", allow_os_buffer);
+    Log(log,"       Options.allow_readahead: %d", allow_readahead);
+    Log(log,"      Options.allow_mmap_reads: %d", allow_mmap_reads);
+    Log(log,"     Options.allow_mmap_writes: %d", allow_mmap_writes);
+    Log(log,"       Options.min_write_buffer_number_to_merge: %d",
+        min_write_buffer_number_to_merge);
+    Log(log,"            Options.allow_readahead_compactions: %d",
+        allow_readahead_compactions);
     Log(log,"        Options.purge_redundant_kvs_while_flush: %d",
          purge_redundant_kvs_while_flush);
     Log(log,"           Options.compression_opts.window_bits: %d",
@@ -127,6 +157,10 @@ Options::Dump(Logger* log) const
         max_bytes_for_level_base);
     Log(log,"         Options.max_bytes_for_level_multiplier: %d",
         max_bytes_for_level_multiplier);
+    for (int i = 0; i < num_levels; i++) {
+      Log(log,"Options.max_bytes_for_level_multiplier_addtl[%d]: %d",
+          i, max_bytes_for_level_multiplier_additional[i]);
+    }
     Log(log,"             Options.expanded_compaction_factor: %d",
         expanded_compaction_factor);
     Log(log,"               Options.source_compaction_factor: %d",
@@ -140,7 +174,7 @@ Options::Dump(Logger* log) const
     Log(log,"                         Options.no_block_cache: %d",
         no_block_cache);
     Log(log,"               Options.table_cache_numshardbits: %d",
-  table_cache_numshardbits);
+        table_cache_numshardbits);
     Log(log,"    Options.delete_obsolete_files_period_micros: %ld",
         delete_obsolete_files_period_micros);
     Log(log,"             Options.max_background_compactions: %d",
@@ -149,16 +183,40 @@ Options::Dump(Logger* log) const
         rate_limit);
     Log(log,"          Options.rate_limit_delay_milliseconds: %d",
         rate_limit_delay_milliseconds);
-    Log(log,"                 Options.compaction_filter_args: %p",
-        compaction_filter_args);
-    Log(log,"                       Options.CompactionFilter: %p",
-        CompactionFilter);
     Log(log,"               Options.disable_auto_compactions: %d",
         disable_auto_compactions);
     Log(log,"                        Options.WAL_ttl_seconds: %ld",
         WAL_ttl_seconds);
     Log(log,"            Options.manifest_preallocation_size: %ld",
         manifest_preallocation_size);
+    Log(log,"         Options.purge_redundant_kvs_while_flush: %d",
+        purge_redundant_kvs_while_flush);
+    Log(log,"                         Options.allow_os_buffer: %d",
+        allow_os_buffer);
+    Log(log,"                         Options.allow_readahead: %d",
+        allow_readahead);
+    Log(log,"             Options.allow_readahead_compactions: %d",
+        allow_readahead_compactions);
+    Log(log,"                        Options.allow_mmap_reads: %d",
+        allow_mmap_reads);
+    Log(log,"                       Options.allow_mmap_writes: %d",
+        allow_mmap_writes);
+    Log(log,"                     Options.is_fd_close_on_exec: %d",
+        is_fd_close_on_exec);
+    Log(log,"              Options.skip_log_error_on_recovery: %d",
+        skip_log_error_on_recovery);
+    Log(log,"                   Options.stats_dump_period_sec: %d",
+        stats_dump_period_sec);
+    Log(log,"                    Options.block_size_deviation: %d",
+        block_size_deviation);
+    Log(log,"                   Options.advise_random_on_open: %d",
+        advise_random_on_open);
+    Log(log,"         Options.access_hint_on_compaction_start: %s",
+        access_hints[access_hint_on_compaction_start]);
+    Log(log,"                      Options.use_adaptive_mutex: %d",
+        use_adaptive_mutex);
+    Log(log,"                          Options.bytes_per_sync: %ld",
+        bytes_per_sync);
 }   // Options::Dump
 
 //

util/posix_logger.h

@@ -18,6 +18,7 @@
 #include <linux/falloc.h>
 #endif
 #include "leveldb/env.h"
+#include <atomic>
 
 namespace leveldb {
 
@@ -27,8 +28,7 @@ class PosixLogger : public Logger {
  private:
   FILE* file_;
   uint64_t (*gettid_)();  // Return the thread id for the current thread
-
-  size_t log_size_;
+  std::atomic_size_t log_size_;
   int fd_;
  public:
   PosixLogger(FILE* f, uint64_t (*gettid)()) :
@@ -94,16 +94,17 @@ class PosixLogger : public Logger {
       }
 
       assert(p <= limit);
+      const size_t write_size = p - base;
 
 #ifdef OS_LINUX
       // If this write would cross a boundary of kDebugLogChunkSize
       // space, pre-allocate more space to avoid overly large
       // allocations from filesystem allocsize options.
-      const size_t write_size = p - base;
+      const size_t log_size = log_size_;
       const int last_allocation_chunk =
-        ((kDebugLogChunkSize - 1 + log_size_) / kDebugLogChunkSize);
+        ((kDebugLogChunkSize - 1 + log_size) / kDebugLogChunkSize);
       const int desired_allocation_chunk =
-        ((kDebugLogChunkSize - 1 + log_size_ + write_size) /
+        ((kDebugLogChunkSize - 1 + log_size + write_size) /
            kDebugLogChunkSize);
       if (last_allocation_chunk != desired_allocation_chunk) {
         fallocate(fd_, FALLOC_FL_KEEP_SIZE, 0,
@@ -111,11 +112,12 @@ class PosixLogger : public Logger {
       }
 #endif
 
-      fwrite(base, 1, write_size, file_);
-      fflush(file_);
-
-      log_size_ += write_size;
-
+      size_t sz = fwrite(base, 1, write_size, file_);
+      assert(sz == write_size);
+      if (sz > 0) {
+        fflush(file_);
+        log_size_ += write_size;
+      }
       if (base != buffer) {
         delete[] base;
       }

util/random.h

@@ -5,6 +5,7 @@
 #ifndef STORAGE_LEVELDB_UTIL_RANDOM_H_
 #define STORAGE_LEVELDB_UTIL_RANDOM_H_
 
+#include <random>
 #include <stdint.h>
 
 namespace leveldb {
@@ -54,6 +55,35 @@ class Random {
   }
 };
 
+// A simple 64bit random number generator based on std::mt19937_64
+class Random64 {
+ private:
+  std::mt19937_64 generator_;
+
+ public:
+  explicit Random64(uint64_t s) : generator_(s) { }
+
+  // Generates the next random number
+  uint64_t Next() { return generator_(); }
+
+  // Returns a uniformly distributed value in the range [0..n-1]
+  // REQUIRES: n > 0
+  uint64_t Uniform(uint64_t n) {
+    return std::uniform_int_distribution<uint64_t>(0, n - 1)(generator_);
+  }
+
+  // Randomly returns true ~"1/n" of the time, and false otherwise.
+  // REQUIRES: n > 0
+  bool OneIn(uint64_t n) { return Uniform(n) == 0; }
+
+  // Skewed: pick "base" uniformly from range [0,max_log] and then
+  // return "base" random bits.  The effect is to pick a number in the
+  // range [0,2^max_log-1] with exponential bias towards smaller numbers.
+  uint64_t Skewed(int max_log) {
+    return Uniform(1 << Uniform(max_log + 1));
+  }
+};
+
 }  // namespace leveldb
 
 #endif  // STORAGE_LEVELDB_UTIL_RANDOM_H_

util/signal_test.cc

@@ -0,0 +1,27 @@
+#include "util/stack_trace.h"
+#include <assert.h>
+
+void f0() {
+  char *p = nullptr;
+  *p = 10;  /* SIGSEGV here!! */
+}
+
+void f1() {
+  f0();
+}
+
+void f2() {
+  f1();
+}
+
+void f3() {
+  f2();
+}
+
+int main() {
+  leveldb::InstallStackTraceHandler();
+
+  f3();
+
+  return 0;
+}

util/stack_trace.h

@@ -0,0 +1,13 @@
+#ifndef STACK_TRACE_H
+#define STACK_TRACE_H
+
+namespace leveldb {
+
+// Install a signal handler to print callstack on the following signals:
+// SIGILL SIGSEGV SIGBUS SIGABRT
+// Currently supports linux only. No-op otherwise.
+void InstallStackTraceHandler();
+
+}   // namespace leveldb
+
+#endif