weed backup: efficient delta backup

fix https://github.com/chrislusf/seaweedfs/issues/399

weed/command/export.go

@@ -107,7 +107,7 @@ func (scanner *VolumeFileScanner4Export) VisitNeedle(n *storage.Needle, offset i
 	nv, ok := needleMap.Get(n.Id)
 	glog.V(3).Infof("key %d offset %d size %d disk_size %d gzip %v ok %v nv %+v",
 		n.Id, offset, n.Size, n.DiskSize(scanner.version), n.IsGzipped(), ok, nv)
-	if ok && nv.Size > 0 && int64(nv.Offset)*types.NeedlePaddingSize == offset {
+	if ok && nv.Size > 0 && nv.Size != types.TombstoneFileSize && int64(nv.Offset)*types.NeedlePaddingSize == offset {
 		if newerThanUnix >= 0 && n.HasLastModifiedDate() && n.LastModified < uint64(newerThanUnix) {
 			glog.V(3).Infof("Skipping this file, as it's old enough: LastModified %d vs %d",
 				n.LastModified, newerThanUnix)

weed/command/fix.go

@@ -44,7 +44,7 @@ func (scanner *VolumeFileScanner4Fix) ReadNeedleBody() bool {
 
 func (scanner *VolumeFileScanner4Fix) VisitNeedle(n *storage.Needle, offset int64) error {
 	glog.V(2).Infof("key %d offset %d size %d disk_size %d gzip %v", n.Id, offset, n.Size, n.DiskSize(scanner.version), n.IsGzipped())
-	if n.Size > 0 {
+	if n.Size > 0 && n.Size != types.TombstoneFileSize {
 		pe := scanner.nm.Put(n.Id, types.Offset(offset/types.NeedlePaddingSize), n.Size)
 		glog.V(2).Infof("saved %d with error %v", n.Size, pe)
 	} else {

weed/server/volume_grpc_follow.go

@@ -20,14 +20,14 @@ func (vs *VolumeServer) VolumeFollow(req *volume_server_pb.VolumeFollowRequest,
 	stopOffset := v.Size()
 	foundOffset, isLastOne, err := v.BinarySearchByAppendAtNs(req.Since)
 	if err != nil {
-		return fmt.Errorf("fail to locate by appendAtNs: %s", err)
+		return fmt.Errorf("fail to locate by appendAtNs %d: %s", req.Since, err)
 	}
 
 	if isLastOne {
 		return nil
 	}
 
-	startOffset := int64(foundOffset) * int64(types.NeedleEntrySize)
+	startOffset := int64(foundOffset) * int64(types.NeedlePaddingSize)
 
 	buf := make([]byte, 1024*1024*2)
 	return sendFileContent(v.DataFile(), buf, startOffset, stopOffset, stream)
@@ -40,7 +40,7 @@ func sendFileContent(datFile *os.File, buf []byte, startOffset, stopOffset int64
 		n, readErr := datFile.ReadAt(buf, startOffset+i)
 		if readErr == nil || readErr == io.EOF {
 			resp := &volume_server_pb.VolumeFollowResponse{}
-			resp.FileContent = buf[i : i+int64(n)]
+			resp.FileContent = buf[:int64(n)]
 			sendErr := stream.Send(resp)
 			if sendErr != nil {
 				return sendErr

weed/storage/volume_follow.go

@@ -87,9 +87,9 @@ func (v *Volume) Follow(volumeServer string, grpcDialOption grpc.DialOption) (er
 		return err
 	}
 
-	// TODO add to needle map
+	// add to needle map
+	return ScanVolumeFileFrom(v.version, v.dataFile, startFromOffset, &VolumeFileScanner4GenIdx{v:v})
 
-	return nil
 }
 
 func (v *Volume) findLastAppendAtNs() (uint64, error) {
@@ -138,9 +138,9 @@ func (v *Volume) readAppendAtNs(offset Offset) (uint64, error) {
 	if err != nil {
 		return 0, fmt.Errorf("ReadNeedleHeader: %v", err)
 	}
-	err = n.ReadNeedleBody(v.dataFile, v.SuperBlock.version, int64(offset)*NeedlePaddingSize, bodyLength)
+	err = n.ReadNeedleBody(v.dataFile, v.SuperBlock.version, int64(offset)*NeedlePaddingSize+int64(NeedleEntrySize), bodyLength)
 	if err != nil {
-		return 0, fmt.Errorf("ReadNeedleBody offset %d: %v", int64(offset)*NeedlePaddingSize, err)
+		return 0, fmt.Errorf("ReadNeedleBody offset %d, bodyLength %d: %v", int64(offset)*NeedlePaddingSize, bodyLength, err)
 	}
 	return n.AppendAtNs, nil
 
@@ -218,3 +218,23 @@ func (v *Volume) readAppendAtNsForIndexEntry(indexFile *os.File, bytes []byte, m
 	_, offset, _ := IdxFileEntry(bytes)
 	return offset, nil
 }
+
+// generate the volume idx
+type VolumeFileScanner4GenIdx struct {
+	v *Volume
+}
+
+func (scanner *VolumeFileScanner4GenIdx) VisitSuperBlock(superBlock SuperBlock) error {
+	return nil
+
+}
+func (scanner *VolumeFileScanner4GenIdx) ReadNeedleBody() bool {
+	return false
+}
+
+func (scanner *VolumeFileScanner4GenIdx) VisitNeedle(n *Needle, offset int64) error {
+	if n.Size > 0 && n.Size != TombstoneFileSize {
+		return scanner.v.nm.Put(n.Id, Offset(offset/NeedlePaddingSize), n.Size)
+	}
+	return scanner.v.nm.Delete(n.Id, Offset(offset/NeedlePaddingSize))
+}

weed/storage/volume_read_write.go

@@ -180,30 +180,37 @@ func ScanVolumeFile(dirname string, collection string, id VolumeId,
 	volumeFileScanner VolumeFileScanner) (err error) {
 	var v *Volume
 	if v, err = loadVolumeWithoutIndex(dirname, collection, id, needleMapKind); err != nil {
-		return fmt.Errorf("Failed to load volume %d: %v", id, err)
+		return fmt.Errorf("failed to load volume %d: %v", id, err)
 	}
 	if err = volumeFileScanner.VisitSuperBlock(v.SuperBlock); err != nil {
-		return fmt.Errorf("Failed to process volume %d super block: %v", id, err)
+		return fmt.Errorf("failed to process volume %d super block: %v", id, err)
 	}
 	defer v.Close()
 
 	version := v.Version()
 
 	offset := int64(v.SuperBlock.BlockSize())
-	n, rest, e := ReadNeedleHeader(v.dataFile, version, offset)
+
+	return ScanVolumeFileFrom(version, v.dataFile, offset, volumeFileScanner)
+}
+
+func ScanVolumeFileFrom(version Version, dataFile *os.File, offset int64, volumeFileScanner VolumeFileScanner) (err error) {
+	n, rest, e := ReadNeedleHeader(dataFile, version, offset)
 	if e != nil {
-		err = fmt.Errorf("cannot read needle header: %v", e)
-		return
+		if e == io.EOF {
+			return nil
+		}
+		return fmt.Errorf("cannot read %s at offset %d: %v", dataFile.Name(), offset, e)
 	}
 	for n != nil {
 		if volumeFileScanner.ReadNeedleBody() {
-			if err = n.ReadNeedleBody(v.dataFile, version, offset+NeedleEntrySize, rest); err != nil {
+			if err = n.ReadNeedleBody(dataFile, version, offset+NeedleEntrySize, rest); err != nil {
 				glog.V(0).Infof("cannot read needle body: %v", err)
 				//err = fmt.Errorf("cannot read needle body: %v", err)
 				//return
 			}
 		}
-		err = volumeFileScanner.VisitNeedle(n, offset)
+		err := volumeFileScanner.VisitNeedle(n, offset)
 		if err == io.EOF {
 			return nil
 		}
@@ -212,14 +219,13 @@ func ScanVolumeFile(dirname string, collection string, id VolumeId,
 		}
 		offset += NeedleEntrySize + rest
 		glog.V(4).Infof("==> new entry offset %d", offset)
-		if n, rest, err = ReadNeedleHeader(v.dataFile, version, offset); err != nil {
+		if n, rest, err = ReadNeedleHeader(dataFile, version, offset); err != nil {
 			if err == io.EOF {
 				return nil
 			}
-			return fmt.Errorf("cannot read needle header: %v", err)
+			return fmt.Errorf("cannot read needle header at offset %d: %v", offset, err)
 		}
 		glog.V(4).Infof("new entry needle size:%d rest:%d", n.Size, rest)
 	}
-
-	return
+	return nil
 }

weed/storage/volume_vacuum.go

@@ -261,7 +261,7 @@ func (scanner *VolumeFileScanner4Vacuum) VisitNeedle(n *Needle, offset int64) er
 	}
 	nv, ok := scanner.v.nm.Get(n.Id)
 	glog.V(4).Infoln("needle expected offset ", offset, "ok", ok, "nv", nv)
-	if ok && int64(nv.Offset)*NeedlePaddingSize == offset && nv.Size > 0 {
+	if ok && int64(nv.Offset)*NeedlePaddingSize == offset && nv.Size > 0 && nv.Size != TombstoneFileSize {
 		if err := scanner.nm.Put(n.Id, Offset(scanner.newOffset/NeedlePaddingSize), n.Size); err != nil {
 			return fmt.Errorf("cannot put needle: %s", err)
 		}