Use different option for max mem-table size and max table-file size. (#…

…167)

So we can set the table file to be much smaller to reduce compaction cost.

blob_test.go

@@ -17,6 +17,7 @@ func TestBlob(t *testing.T) {
 	opts := getTestOptions(dir)
 	opts.ValueThreshold = 20
 	opts.MaxTableSize = 4 * 1024
+	opts.MaxMemTableSize = 4 * 1024
 	opts.NumMemtables = 2
 	opts.NumLevelZeroTables = 1
 	opts.NumLevelZeroTablesStall = 2
@@ -70,6 +71,7 @@ func TestBlobGC(t *testing.T) {
 	opts := getTestOptions(dir)
 	opts.ValueThreshold = 20
 	opts.MaxTableSize = 6 * 1024
+	opts.MaxMemTableSize = 6 * 1024
 	opts.NumMemtables = 2
 	opts.NumLevelZeroTables = 1
 	opts.NumLevelZeroTablesStall = 2

db.go

@@ -125,7 +125,7 @@ func replayFunction(out *DB) func(Entry) error {
 	toLSM := func(nk []byte, vs y.ValueStruct) {
 		e := table.Entry{Key: nk, Value: vs}
 		mTbls := out.mtbls.Load().(*memTables)
-		if out.ensureRoomForWrite(mTbls.getMutable(), e.EstimateSize()) == out.opt.MaxTableSize {
+		if out.ensureRoomForWrite(mTbls.getMutable(), e.EstimateSize()) == out.opt.MaxMemTableSize {
 			mTbls = out.mtbls.Load().(*memTables)
 		}
 		mTbls.getMutable().PutToSkl(nk, vs)
@@ -190,7 +190,7 @@ func replayFunction(out *DB) func(Entry) error {
 
 // Open returns a new DB object.
 func Open(opt Options) (db *DB, err error) {
-	opt.maxBatchSize = (15 * opt.MaxTableSize) / 100
+	opt.maxBatchSize = (15 * opt.MaxMemTableSize) / 100
 	opt.maxBatchCount = opt.maxBatchSize / int64(skl.MaxNodeSize)
 
 	if opt.ValueThreshold > math.MaxUint16-16 {
@@ -773,12 +773,12 @@ func (db *DB) batchSetAsync(entries []*Entry, f func(error)) error {
 
 // ensureRoomForWrite is always called serially.
 func (db *DB) ensureRoomForWrite(mt *table.MemTable, minSize int64) int64 {
-	free := db.opt.MaxTableSize - mt.MemSize()
+	free := db.opt.MaxMemTableSize - mt.MemSize()
 	if free >= minSize {
 		return free
 	}
 	_ = db.flushMemTable()
-	return db.opt.MaxTableSize
+	return db.opt.MaxMemTableSize
 }
 
 func (db *DB) flushMemTable() *sync.WaitGroup {
@@ -795,7 +795,7 @@ func (db *DB) flushMemTable() *sync.WaitGroup {
 }
 
 func arenaSize(opt Options) int64 {
-	return opt.MaxTableSize + opt.maxBatchCount*int64(skl.MaxNodeSize)
+	return opt.MaxMemTableSize + opt.maxBatchCount*int64(skl.MaxNodeSize)
 }
 
 // WriteLevel0Table flushes memtable. It drops deleteValues.

db_test.go

@@ -50,8 +50,9 @@ func getTestCompression(tp options.CompressionType) []options.CompressionType {
 
 func getTestOptions(dir string) Options {
 	opt := DefaultOptions
-	opt.MaxTableSize = 4 << 15 // Force more compaction.
-	opt.LevelOneSize = 4 << 15 // Force more compaction.
+	opt.MaxTableSize = 4 << 15    // Force more compaction.
+	opt.MaxMemTableSize = 4 << 15 // Force more compaction.
+	opt.LevelOneSize = 4 << 15    // Force more compaction.
 	opt.Dir = dir
 	opt.ValueDir = dir
 	opt.SyncWrites = false
@@ -1158,6 +1159,7 @@ func TestCompactionFilter(t *testing.T) {
 	opts := getTestOptions(dir)
 	opts.ValueThreshold = 8 * 1024
 	opts.MaxTableSize = 32 * 1024
+	opts.MaxMemTableSize = 32 * 1024
 	opts.NumMemtables = 2
 	opts.NumLevelZeroTables = 1
 	opts.NumLevelZeroTablesStall = 2
@@ -1293,6 +1295,7 @@ func TestIterateVLog(t *testing.T) {
 	defer os.RemoveAll(dir)
 	opts := getTestOptions(dir)
 	opts.MaxTableSize = 1 << 20
+	opts.MaxMemTableSize = 1 << 20
 	opts.ValueLogFileSize = 1 << 20
 	opts.ValueThreshold = 1000
 	opts.ValueLogMaxNumFiles = 1000

options.go

@@ -46,7 +46,8 @@ type Options struct {
 	// 3. Flags that user might want to review
 	// ----------------------------------------
 	// The following affect all levels of LSM tree.
-	MaxTableSize int64 // Each table (or file) is at most this size.
+	MaxMemTableSize int64 // Each mem table is at most this size.
+	MaxTableSize    int64 // Each table file is at most this size.
 	// If value size >= this threshold, only store value offsets in tree.
 	// If set to 0, all values are stored in SST.
 	ValueThreshold int
@@ -78,9 +79,6 @@ type Options struct {
 	// Number of compaction workers to run concurrently.
 	NumCompactors int
 
-	// Max number of sub compaction, set 1 or 0 to disable sub compaction.
-	MaxSubCompaction int
-
 	// Transaction start and commit timestamps are manaVgedTxns by end-user. This
 	// is a private option used by ManagedDB.
 	managedTxns bool
@@ -145,13 +143,11 @@ const (
 // DefaultOptions sets a list of recommended options for good performance.
 // Feel free to modify these to suit your needs.
 var DefaultOptions = Options{
-	DoNotCompact: false,
-	LevelOneSize: 256 << 20,
-	// table.MemoryMap to mmap() the tables.
-	// table.Nothing to not preload the tables.
-	MaxTableSize:            64 << 20,
+	DoNotCompact:            false,
+	LevelOneSize:            256 << 20,
+	MaxMemTableSize:         64 << 20,
+	MaxTableSize:            8 << 20,
 	NumCompactors:           3,
-	MaxSubCompaction:        3,
 	NumLevelZeroTables:      5,
 	NumLevelZeroTablesStall: 10,
 	NumMemtables:            5,

transaction.go

@@ -245,7 +245,7 @@ func (txn *Txn) checkSize(e *Entry) error {
 	}
 	// Extra bytes for version in key.
 	size := int64(e.estimateSize()) + 10
-	if size >= txn.db.opt.MaxTableSize {
+	if size >= txn.db.opt.MaxMemTableSize {
 		return ErrTxnTooBig
 	}
 	txn.count++

transaction_test.go

@@ -801,6 +801,7 @@ func TestArmV7Issue311Fix(t *testing.T) {
 	config.ValueLogFileSize = 16 << 20
 	config.LevelOneSize = 8 << 20
 	config.MaxTableSize = 2 << 20
+	config.MaxMemTableSize = 2 << 20
 	config.Dir = dir
 	config.ValueDir = dir
 	config.SyncWrites = false

writer.go

@@ -225,7 +225,7 @@ func (w *writeWorker) writeToLSM(entries []*Entry) error {
 	for len(entries) != 0 {
 		e := newEntry(entries[0])
 		free := w.ensureRoomForWrite(mTbls.getMutable(), e.EstimateSize())
-		if free == w.opt.MaxTableSize {
+		if free == w.opt.MaxMemTableSize {
 			mTbls = w.mtbls.Load().(*memTables)
 		}