fix ec related bugs

weed/server/volume_server_handlers_read.go

@@ -76,7 +76,7 @@ func (vs *VolumeServer) GetOrHeadHandler(w http.ResponseWriter, r *http.Request)
 	}
 	glog.V(4).Infoln("read bytes", count, "error", err)
 	if err != nil || count < 0 {
-		glog.V(0).Infof("read %s error: %v", r.URL.Path, err)
+		glog.V(0).Infof("read %s isNormalVolume %v error: %v", r.URL.Path, hasVolume, err)
 		w.WriteHeader(http.StatusNotFound)
 		return
 	}

weed/shell/command_ec_balance.go

@@ -199,6 +199,7 @@ func doDeduplicateEcShards(ctx context.Context, commandEnv *CommandEnv, collecti
 			if err := sourceServerDeleteEcShards(ctx, commandEnv.option.GrpcDialOption, collection, vid, ecNode.info.Id, duplicatedShardIds); err != nil {
 				return err
 			}
+			deleteEcVolumeShards(ecNode, vid, duplicatedShardIds)
 			ecNode.freeEcSlot++
 		}
 	}
@@ -273,3 +274,27 @@ func findEcVolumeShards(ecNode *EcNode, vid needle.VolumeId) erasure_coding.Shar
 
 	return 0
 }
+
+func addEcVolumeShards(ecNode *EcNode, vid needle.VolumeId, shardIds []uint32){
+
+	for _, shardInfo := range ecNode.info.EcShardInfos {
+		if needle.VolumeId(shardInfo.Id) == vid {
+			for _, shardId := range shardIds{
+				shardInfo.EcIndexBits = uint32(erasure_coding.ShardBits(shardInfo.EcIndexBits).AddShardId(erasure_coding.ShardId(shardId)))
+			}
+		}
+	}
+
+}
+
+func deleteEcVolumeShards(ecNode *EcNode, vid needle.VolumeId, shardIds []uint32){
+
+	for _, shardInfo := range ecNode.info.EcShardInfos {
+		if needle.VolumeId(shardInfo.Id) == vid {
+			for _, shardId := range shardIds{
+				shardInfo.EcIndexBits = uint32(erasure_coding.ShardBits(shardInfo.EcIndexBits).RemoveShardId(erasure_coding.ShardId(shardId)))
+			}
+		}
+	}
+
+}

weed/shell/command_ec_common.go

@@ -35,7 +35,14 @@ func moveMountedShardToEcNode(ctx context.Context, commandEnv *CommandEnv, exist
 	}
 
 	// ask source node to delete the shard, and maybe the ecx file
-	return sourceServerDeleteEcShards(ctx, commandEnv.option.GrpcDialOption, collection, vid, existingLocation.info.Id, copiedShardIds)
+	err = sourceServerDeleteEcShards(ctx, commandEnv.option.GrpcDialOption, collection, vid, existingLocation.info.Id, copiedShardIds)
+	if err != nil {
+		return err
+	}
+
+	deleteEcVolumeShards(existingLocation, vid, copiedShardIds)
+
+	return nil
 
 }
 

weed/shell/command_ec_rebuild.go

@@ -163,6 +163,8 @@ func rebuildOneEcVolume(ctx context.Context, commandEnv *CommandEnv, rebuilder *
 		return err
 	}
 
+	addEcVolumeShards(rebuilder, volumeId, generatedShardIds)
+
 	return nil
 }
 

weed/storage/disk_location.go

@@ -223,8 +223,8 @@ func (l *DiskLocation) Close() {
 	l.Unlock()
 
 	l.ecVolumesLock.Lock()
-	for _, shards := range l.ecVolumes {
+	for _, ecVolume := range l.ecVolumes {
-		shards.Close()
+		ecVolume.Close()
 	}
 	l.ecVolumesLock.Unlock()
 

weed/storage/disk_location_ec.go

@@ -87,8 +87,8 @@ func (l *DiskLocation) UnloadEcShard(vid needle.VolumeId, shardId erasure_coding
 	if _, deleted := ecVolume.DeleteEcVolumeShard(shardId); deleted {
 		if len(ecVolume.Shards) == 0 {
 			delete(l.ecVolumes, vid)
-		}
 			ecVolume.Close()
+		}
 		return true
 	}
 

weed/storage/erasure_coding/ec_volume.go

@@ -155,7 +155,7 @@ func (ev *EcVolume) LocateEcShardNeedle(n *needle.Needle, version needle.Version
 	// find the needle from ecx file
 	offset, size, err = ev.findNeedleFromEcx(n.Id)
 	if err != nil {
-		return types.Offset{}, 0, nil, err
+		return types.Offset{}, 0, nil, fmt.Errorf("findNeedleFromEcx: %v", err)
 	}
 
 	shard := ev.Shards[0]
@@ -173,7 +173,7 @@ func (ev *EcVolume) findNeedleFromEcx(needleId types.NeedleId) (offset types.Off
 	for l < h {
 		m := (l + h) / 2
 		if _, err := ev.ecxFile.ReadAt(buf, m*types.NeedleMapEntrySize); err != nil {
-			return types.Offset{}, 0, err
+			return types.Offset{}, 0, fmt.Errorf("ecx file %d read at %d: %v", ev.ecxFileSize, m*types.NeedleMapEntrySize, err)
 		}
 		key, offset, size = idx.IdxFileEntry(buf)
 		if key == needleId {

weed/storage/store_ec.go

@@ -118,7 +118,7 @@ func (s *Store) ReadEcShardNeedle(ctx context.Context, vid needle.VolumeId, n *n
 
 			offset, size, intervals, err := localEcVolume.LocateEcShardNeedle(n, version)
 			if err != nil {
-				return 0, err
+				return 0, fmt.Errorf("locate in local ec volume: %v", err)
 			}
 
 			glog.V(4).Infof("read ec volume %d offset %d size %d intervals:%+v", vid, offset.ToAcutalOffset(), size, intervals)

weed/topology/topology.go

@@ -78,7 +78,7 @@ func (t *Topology) Leader() (string, error) {
 	return l, nil
 }
 
-func (t *Topology) Lookup(collection string, vid needle.VolumeId) []*DataNode {
+func (t *Topology) Lookup(collection string, vid needle.VolumeId) (dataNodes []*DataNode) {
 	//maybe an issue if lots of collections?
 	if collection == "" {
 		for _, c := range t.collectionMap.Items() {
@@ -91,6 +91,14 @@ func (t *Topology) Lookup(collection string, vid needle.VolumeId) []*DataNode {
 			return c.(*Collection).Lookup(vid)
 		}
 	}
+
+	if locations, found := t.LookupEcShards(vid); found {
+		for _, loc := range locations.Locations {
+			dataNodes = append(dataNodes, loc...)
+		}
+		return dataNodes
+	}
+
 	return nil
 }