use simple delta encoding for key offsets

memiavl/README.md

@@ -97,17 +97,18 @@ IAVL snapshot is composed by four files:
 
   The implementation will read the mmap-ed content in a zero-copy way, won't use extra node cache, it will only rely on the OS page cache.
 
-- `keys`, sequence of leaf node keys, ordered and no duplication, the offsets are appended to the end of the file, user can look up the key offset by leaf node index.
+- `keys`, sequence of leaf node keys, ordered and no duplication, the offsets are encoded with custom format and appended to the end of the file, support query by leaf node index.
 
   ```
   payload
   *repeat*
-  key offset: uint32
+  offset restart: uint64
+  delta offsets: [65535]uint32
   *repeat*
-  offset: uint64    // begin offset of the offsets table
+  offset: uint64    // beginning offset of the above table
   ```
 
-- `values`, sequence of leaf node values, the offsets are encoded with elias-fano coding and appended to the end of the file, user can look up the key offset by leaf node index.
+- `values`, sequence of leaf node values, the offsets are encoded with elias-fano coding and appended to the end of the file, support query by leaf node index.
 
   ```
   payload

memiavl/layout_little_endian.go

@@ -53,11 +53,26 @@ type PlainOffsetTable struct {
 	data []byte
 }
 
-func (t PlainOffsetTable) Get2(i uint64) (uint32, uint32) {
-	offset := i * 4
-	start := binary.LittleEndian.Uint32(t.data[offset:])
-	end := binary.LittleEndian.Uint32(t.data[offset+4:])
-	return start, end
+func (t PlainOffsetTable) Get2(i uint64) (uint64, uint64) {
+	ichunk := i / OffsetRestartInteval
+	ii := i % OffsetRestartInteval
+	irestart := ichunk * (OffsetRestartInteval + 1) * 4
+	data := t.data[irestart:]
+
+	_ = data[3*4-1]
+	restart := binary.LittleEndian.Uint64(data[:8])
+
+	if ii == 0 {
+		return restart, restart + uint64(binary.LittleEndian.Uint32(data[8:12]))
+	}
+	if ii == OffsetRestartInteval-1 {
+		// the next one is at the beginning of the next chunk
+		return restart + uint64(binary.LittleEndian.Uint32(data[OffsetRestartInteval*4:])),
+			binary.LittleEndian.Uint64(data[(OffsetRestartInteval+1)*4:])
+	}
+	// the next one is in the same chunk
+	return restart + uint64(binary.LittleEndian.Uint32(data[(ii+1)*4:])),
+		restart + uint64(binary.LittleEndian.Uint32(data[(ii+2)*4:]))
 }
 
 func NewPlainOffsetTable(data []byte) (PlainOffsetTable, error) {

memiavl/layout_native.go

@@ -65,8 +65,24 @@ type PlainOffsetTable struct {
 	data []uint32
 }
 
-func (t PlainOffsetTable) Get2(i uint64) (uint32, uint32) {
-	return t.data[i], t.data[i+1]
+func (t PlainOffsetTable) Get2(i uint64) (uint64, uint64) {
+	ichunk := i / OffsetRestartInteval
+	ii := i % OffsetRestartInteval
+	irestart := ichunk * (OffsetRestartInteval + 1)
+	data := t.data[irestart:]
+
+	_ = data[2]
+	restart := uint64(data[0]) | uint64(data[1])<<32
+
+	if ii == 0 {
+		return restart, restart + uint64(data[2])
+	}
+	if ii == OffsetRestartInteval-1 {
+		data2 := data[OffsetRestartInteval+1:]
+		_ = data2[1]
+		return restart + uint64(data[OffsetRestartInteval]), uint64(data2[0]) | uint64(data2[1])<<32
+	}
+	return restart + uint64(data[ii+1]), restart + uint64(data[ii+2])
 }
 
 func NewPlainOffsetTable(buf []byte) (PlainOffsetTable, error) {

memiavl/snapshot.go

@@ -29,6 +29,8 @@ const (
 	EmptyRootNodeIndex = math.MaxUint32
 
 	Alignment = 8
+
+	OffsetRestartInteval = 65536
 )
 
 // Snapshot manage the lifecycle of mmap-ed files for the snapshot,
@@ -498,14 +500,25 @@ func writePadding(w io.Writer, offset uint64) (uint64, error) {
 
 // writePlainOffsets writes the offset table in plain little-endian format
 func writePlainOffsets(w io.Writer, bitmap *roaring64.Bitmap) error {
-	var numBuf [4]byte
+	var numBuf [8]byte
 	it := bitmap.Iterator()
+	var counter, restart uint64
 	for it.HasNext() {
 		v := it.Next()
-		binary.LittleEndian.PutUint32(numBuf[:], uint32(v))
-		if _, err := w.Write(numBuf[:]); err != nil {
-			return err
+		if counter%OffsetRestartInteval == 0 {
+			binary.LittleEndian.PutUint64(numBuf[:], v)
+			restart = v
+
+			if _, err := w.Write(numBuf[:]); err != nil {
+				return err
+			}
+		} else {
+			binary.LittleEndian.PutUint32(numBuf[:], uint32(v-restart))
+			if _, err := w.Write(numBuf[:4]); err != nil {
+				return err
+			}
 		}
+		counter++
 	}
 	return nil
 }