switch to ReadAt() for thread-safe read

fix bugs during volume compaction

go/replication/volume_growth.go

@@ -137,7 +137,7 @@ func (vg *VolumeGrowth) grow(topo *topology.Topology, vid storage.VolumeId, coll
 		if err := AllocateVolume(server, vid, collection, rp); err == nil {
 			vi := storage.VolumeInfo{Id: vid, Size: 0, Collection: collection, ReplicaPlacement: rp, Version: storage.CurrentVersion}
 			server.AddOrUpdateVolume(vi)
-			topo.RegisterVolumeLayout(&vi, server)
+			topo.RegisterVolumeLayout(vi, server)
 			glog.V(0).Infoln("Created Volume", vid, "on", server)
 		} else {
 			glog.V(0).Infoln("Failed to assign", vid, "to", servers, "error", err)

go/storage/volume.go

@@ -106,12 +106,12 @@ func (v *Volume) load(alsoLoadIndex bool, createDatIfMissing bool) error {
 		if v.readOnly {
 			glog.V(1).Infoln("open to read file", fileName+".idx")
 			if indexFile, e = os.OpenFile(fileName+".idx", os.O_RDONLY, 0644); e != nil {
-				return fmt.Errorf("cannot read Volume Data %s.dat: %s", fileName, e.Error())
+				return fmt.Errorf("cannot read Volume Index %s.idx: %s", fileName, e.Error())
 			}
 		} else {
 			glog.V(1).Infoln("open to write file", fileName+".idx")
 			if indexFile, e = os.OpenFile(fileName+".idx", os.O_RDWR|os.O_CREATE, 0644); e != nil {
-				return fmt.Errorf("cannot write Volume Data %s.dat: %s", fileName, e.Error())
+				return fmt.Errorf("cannot write Volume Index %s.idx: %s", fileName, e.Error())
 			}
 		}
 		glog.V(0).Infoln("loading file", fileName+".idx", "readonly", v.readOnly)
@@ -287,7 +287,8 @@ func (v *Volume) Compact() error {
 	v.accessLock.Lock()
 	defer v.accessLock.Unlock()
 
-	filePath := path.Join(v.dir, v.Id.String())
+	filePath := v.FileName()
+	glog.V(3).Infof("creating copies for volume %d ...", v.Id)
 	return v.copyDataAndGenerateIndexFile(filePath+".cpd", filePath+".cpx")
 }
 func (v *Volume) commitCompact() error {
@@ -295,10 +296,10 @@ func (v *Volume) commitCompact() error {
 	defer v.accessLock.Unlock()
 	_ = v.dataFile.Close()
 	var e error
-	if e = os.Rename(path.Join(v.dir, v.Id.String()+".cpd"), path.Join(v.dir, v.Id.String()+".dat")); e != nil {
+	if e = os.Rename(v.FileName()+".cpd", v.FileName()+".dat"); e != nil {
 		return e
 	}
-	if e = os.Rename(path.Join(v.dir, v.Id.String()+".cpx"), path.Join(v.dir, v.Id.String()+".idx")); e != nil {
+	if e = os.Rename(v.FileName()+".cpx", v.FileName()+".idx"); e != nil {
 		return e
 	}
 	if e = v.load(true, false); e != nil {
@@ -337,10 +338,10 @@ func ScanVolumeFile(dirname string, collection string, id VolumeId,
 	visitNeedle func(n *Needle, offset int64) error) (err error) {
 	var v *Volume
 	if v, err = loadVolumeWithoutIndex(dirname, collection, id); err != nil {
-		return
+		return errors.New("Failed to load volume:" + err.Error())
 	}
 	if err = visitSuperBlock(v.SuperBlock); err != nil {
-		return
+		return errors.New("Failed to read super block:" + err.Error())
 	}
 
 	version := v.Version()
@@ -352,15 +353,14 @@ func ScanVolumeFile(dirname string, collection string, id VolumeId,
 		return
 	}
 	for n != nil {
-		offset += int64(NeedleHeaderSize)
-		if err = n.ReadNeedleBody(v.dataFile, version, offset, rest); err != nil {
+		if err = n.ReadNeedleBody(v.dataFile, version, offset+int64(NeedleHeaderSize), rest); err != nil {
 			err = fmt.Errorf("cannot read needle body: %s", err)
 			return
 		}
 		if err = visitNeedle(n, offset); err != nil {
 			return
 		}
-		offset += int64(rest)
+		offset += int64(NeedleHeaderSize) + int64(rest)
 		if n, rest, err = ReadNeedleHeader(v.dataFile, version, offset); err != nil {
 			if err == io.EOF {
 				return nil
@@ -386,6 +386,7 @@ func (v *Volume) copyDataAndGenerateIndexFile(dstName, idxName string) (err erro
 	}
 	defer idx.Close()
 
+	nm := NewNeedleMap(idx)
 	new_offset := int64(SuperBlockSize)
 
 	err = ScanVolumeFile(v.dir, v.Collection, v.Id, func(superBlock SuperBlock) error {
@@ -393,13 +394,16 @@ func (v *Volume) copyDataAndGenerateIndexFile(dstName, idxName string) (err erro
 		return err
 	}, func(n *Needle, offset int64) error {
 		nv, ok := v.nm.Get(n.Id)
-		//glog.V(0).Infoln("file size is", n.Size, "rest", rest)
+		glog.V(3).Infoln("needle expected offset ", offset, "ok", ok, "nv", nv)
 		if ok && int64(nv.Offset)*NeedlePaddingSize == offset && nv.Size > 0 {
+			if _, err = nm.Put(n.Id, uint32(new_offset/NeedlePaddingSize), n.Size); err != nil {
+				return fmt.Errorf("cannot put needle: %s", err)
+			}
 			if _, err = n.Append(dst, v.Version()); err != nil {
 				return fmt.Errorf("cannot append needle: %s", err)
 			}
 			new_offset += n.DiskSize()
-			//glog.V(0).Infoln("saving key", n.Id, "volume offset", old_offset, "=>", new_offset, "data_size", n.Size, "rest", rest)
+			glog.V(3).Infoln("saving key", n.Id, "volume offset", offset, "=>", new_offset, "data_size", n.Size)
 		}
 		return nil
 	})

go/topology/topology.go

@@ -129,8 +129,8 @@ func (t *Topology) DeleteCollection(collectionName string) {
 	delete(t.collectionMap, collectionName)
 }
 
-func (t *Topology) RegisterVolumeLayout(v *storage.VolumeInfo, dn *DataNode) {
-	t.GetVolumeLayout(v.Collection, v.ReplicaPlacement).RegisterVolume(v, dn)
+func (t *Topology) RegisterVolumeLayout(v storage.VolumeInfo, dn *DataNode) {
+	t.GetVolumeLayout(v.Collection, v.ReplicaPlacement).RegisterVolume(&v, dn)
 }
 
 func (t *Topology) RegisterVolumes(init bool, volumeInfos []storage.VolumeInfo, ip string, port int, publicUrl string, maxVolumeCount int, dcName string, rackName string) {
@@ -144,7 +144,7 @@ func (t *Topology) RegisterVolumes(init bool, volumeInfos []storage.VolumeInfo,
 	dn = rack.GetOrCreateDataNode(ip, port, publicUrl, maxVolumeCount)
 	dn.UpdateVolumes(volumeInfos)
 	for _, v := range volumeInfos {
-		t.RegisterVolumeLayout(&v, dn)
+		t.RegisterVolumeLayout(v, dn)
 	}
 }
 

go/topology/volume_layout.go

@@ -33,15 +33,22 @@ func (vl *VolumeLayout) RegisterVolume(v *storage.VolumeInfo, dn *DataNode) {
 	if _, ok := vl.vid2location[v.Id]; !ok {
 		vl.vid2location[v.Id] = NewVolumeLocationList()
 	}
-	if vl.vid2location[v.Id].Add(dn) {
-		if len(vl.vid2location[v.Id].list) == v.ReplicaPlacement.GetCopyCount() {
-			if vl.isWritable(v) {
-				vl.writables = append(vl.writables, v.Id)
-			} else {
-				vl.removeFromWritable(v.Id)
-			}
+	vl.vid2location[v.Id].Set(dn)
+	glog.V(3).Infoln("volume", v.Id, "added to dn", dn, "len", vl.vid2location[v.Id].Length(), "copy", v.ReplicaPlacement.GetCopyCount())
+	if vl.vid2location[v.Id].Length() == vl.rp.GetCopyCount() && vl.isWritable(v) {
+		vl.AddToWritable(v.Id)
+	} else {
+		vl.removeFromWritable(v.Id)
+	}
+}
+
+func (vl *VolumeLayout) AddToWritable(vid storage.VolumeId) {
+	for _, id := range vl.writables {
+		if vid == id {
+			return
 		}
 	}
+	vl.writables = append(vl.writables, vid)
 }
 
 func (vl *VolumeLayout) isWritable(v *storage.VolumeInfo) bool {
@@ -156,10 +163,9 @@ func (vl *VolumeLayout) SetVolumeAvailable(dn *DataNode, vid storage.VolumeId) b
 	vl.accessLock.Lock()
 	defer vl.accessLock.Unlock()
 
-	if vl.vid2location[vid].Add(dn) {
-		if vl.vid2location[vid].Length() >= vl.rp.GetCopyCount() {
-			return vl.setVolumeWritable(vid)
-		}
+	vl.vid2location[vid].Set(dn)
+	if vl.vid2location[vid].Length() >= vl.rp.GetCopyCount() {
+		return vl.setVolumeWritable(vid)
 	}
 	return false
 }

go/topology/volume_location_list.go

@@ -18,14 +18,14 @@ func (dnll *VolumeLocationList) Length() int {
 	return len(dnll.list)
 }
 
-func (dnll *VolumeLocationList) Add(loc *DataNode) bool {
-	for _, dnl := range dnll.list {
-		if loc.Ip == dnl.Ip && loc.Port == dnl.Port {
-			return false
+func (dnll *VolumeLocationList) Set(loc *DataNode) {
+	for i := 0; i < len(dnll.list); i++ {
+		if loc.Ip == dnll.list[i].Ip && loc.Port == dnll.list[i].Port {
+			dnll.list[i] = loc
+			return
 		}
 	}
 	dnll.list = append(dnll.list, loc)
-	return true
 }
 
 func (dnll *VolumeLocationList) Remove(loc *DataNode) bool {