use async write to persistent file to disk - part1

weed/storage/needle/async_request.go

@@ -0,0 +1,55 @@
+package needle
+
+type AsyncRequest struct {
+	N              *Needle
+	IsWriteRequest bool
+	ActualSize     int64
+	Fsync          bool
+	offset         uint64
+	size           uint64
+	doneChan       chan interface{}
+	isUnchanged    bool
+	err            error
+}
+
+func NewAsyncRequest(n *Needle, isWriteRequest bool, fsync bool) *AsyncRequest {
+	return &AsyncRequest{
+		offset:         0,
+		size:           0,
+		ActualSize:     0,
+		doneChan:       make(chan interface{}),
+		N:              n,
+		isUnchanged:    false,
+		IsWriteRequest: isWriteRequest,
+		Fsync:          fsync,
+		err:            nil,
+	}
+}
+
+func (r *AsyncRequest) WaitComplete() (uint64, uint64, bool, error) {
+	<-r.doneChan
+	return r.offset, r.size, r.isUnchanged, r.err
+}
+
+func (r *AsyncRequest) Complete(offset uint64, size uint64, isUnchanged bool, err error) {
+	r.offset = offset
+	r.size = size
+	r.isUnchanged = isUnchanged
+	r.err = err
+	close(r.doneChan)
+}
+
+func (r *AsyncRequest) UpdateResult(offset uint64, size uint64, isUnchanged bool, err error) {
+	r.offset = offset
+	r.size = size
+	r.isUnchanged = isUnchanged
+	r.err = err
+}
+
+func (r *AsyncRequest) Submit() {
+	close(r.doneChan)
+}
+
+func (r *AsyncRequest) IsSucceed() bool {
+	return r.err == nil
+}

weed/storage/store.go

@@ -264,12 +264,7 @@ func (s *Store) WriteVolumeNeedle(i needle.VolumeId, n *needle.Needle, fsync boo
 			err = fmt.Errorf("volume %d is read only", i)
 			return
 		}
-		// using len(n.Data) here instead of n.Size before n.Size is populated in v.writeNeedle(n)
+		_, _, isUnchanged, err = v.writeNeedle2(n, fsync)
-		if MaxPossibleVolumeSize >= v.ContentSize()+uint64(needle.GetActualSize(uint32(len(n.Data)), v.Version())) {
-			_, _, isUnchanged, err = v.writeNeedle(n, fsync)
-		} else {
-			err = fmt.Errorf("volume size limit %d exceeded! current size is %d", s.GetVolumeSizeLimit(), v.ContentSize())
-		}
 		return
 	}
 	glog.V(0).Infoln("volume", i, "not found!")
@@ -282,11 +277,7 @@ func (s *Store) DeleteVolumeNeedle(i needle.VolumeId, n *needle.Needle) (uint32,
 		if v.noWriteOrDelete {
 			return 0, fmt.Errorf("volume %d is read only", i)
 		}
-		if MaxPossibleVolumeSize >= v.ContentSize()+uint64(needle.GetActualSize(0, v.Version())) {
+		return v.deleteNeedle2(n)
-			return v.deleteNeedle(n)
-		} else {
-			return 0, fmt.Errorf("volume size limit %d exceeded! current size is %d", s.GetVolumeSizeLimit(), v.ContentSize())
-		}
 	}
 	return 0, fmt.Errorf("volume %d not found on %s:%d", i, s.Ip, s.Port)
 }

weed/storage/volume.go

@@ -33,6 +33,7 @@ type Volume struct {
 	super_block.SuperBlock
 
 	dataFileAccessLock    sync.RWMutex
+	asyncRequestsChan     chan *needle.AsyncRequest
 	lastModifiedTsSeconds uint64 //unix time in seconds
 	lastAppendAtNs        uint64 //unix time in nanoseconds
 
@@ -46,12 +47,15 @@ type Volume struct {
 
 func NewVolume(dirname string, collection string, id needle.VolumeId, needleMapKind NeedleMapType, replicaPlacement *super_block.ReplicaPlacement, ttl *needle.TTL, preallocate int64, memoryMapMaxSizeMb uint32) (v *Volume, e error) {
 	// if replicaPlacement is nil, the superblock will be loaded from disk
-	v = &Volume{dir: dirname, Collection: collection, Id: id, MemoryMapMaxSizeMb: memoryMapMaxSizeMb}
+	v = &Volume{dir: dirname, Collection: collection, Id: id, MemoryMapMaxSizeMb: memoryMapMaxSizeMb,
+		asyncRequestsChan: make(chan *needle.AsyncRequest, 128)}
 	v.SuperBlock = super_block.SuperBlock{ReplicaPlacement: replicaPlacement, Ttl: ttl}
 	v.needleMapKind = needleMapKind
 	e = v.load(true, true, needleMapKind, preallocate)
+	v.startWorker()
 	return
 }
+
 func (v *Volume) String() string {
 	return fmt.Sprintf("Id:%v, dir:%s, Collection:%s, dataFile:%v, nm:%v, noWrite:%v canDelete:%v", v.Id, v.dir, v.Collection, v.DataBackend, v.nm, v.noWriteOrDelete || v.noWriteCanDelete, v.noWriteCanDelete)
 }
@@ -65,6 +69,7 @@ func VolumeFileName(dir string, collection string, id int) (fileName string) {
 	}
 	return
 }
+
 func (v *Volume) FileName() (fileName string) {
 	return VolumeFileName(v.dir, v.Collection, int(v.Id))
 }

weed/storage/volume_read_write.go

@@ -46,6 +46,7 @@ func (v *Volume) Destroy() (err error) {
 		err = fmt.Errorf("volume %d is compacting", v.Id)
 		return
 	}
+	close(v.asyncRequestsChan)
 	storageName, storageKey := v.RemoteStorageNameKey()
 	if v.HasRemoteFile() && storageName != "" && storageKey != "" {
 		if backendStorage, found := backend.BackendStorages[storageName]; found {
@@ -63,7 +64,11 @@ func (v *Volume) Destroy() (err error) {
 	return
 }
 
-func (v *Volume) writeNeedle(n *needle.Needle, fsync bool) (offset uint64, size uint32, isUnchanged bool, err error) {
+func (v *Volume) asyncRequestAppend(request *needle.AsyncRequest) {
+	v.asyncRequestsChan <- request
+}
+
+func (v *Volume) writeNeedleDeprecated(n *needle.Needle, fsync bool) (offset uint64, size uint32, isUnchanged bool, err error) {
 	// glog.V(4).Infof("writing needle %s", needle.NewFileIdFromNeedle(v.Id, n).String())
 	v.dataFileAccessLock.Lock()
 	defer v.dataFileAccessLock.Unlock()
@@ -117,7 +122,66 @@ func (v *Volume) writeNeedle(n *needle.Needle, fsync bool) (offset uint64, size
 	return
 }
 
-func (v *Volume) deleteNeedle(n *needle.Needle) (uint32, error) {
+func (v *Volume) writeNeedle2(n *needle.Needle, fsync bool) (offset uint64, size uint32, isUnchanged bool, err error) {
+	// glog.V(4).Infof("writing needle %s", needle.NewFileIdFromNeedle(v.Id, n).String())
+	if n.Ttl == needle.EMPTY_TTL && v.Ttl != needle.EMPTY_TTL {
+		n.SetHasTtl()
+		n.Ttl = v.Ttl
+	}
+
+	asyncRequest := needle.NewAsyncRequest(n, true, fsync)
+	// using len(n.Data) here instead of n.Size before n.Size is populated in n.Append()
+	asyncRequest.ActualSize = needle.GetActualSize(uint32(len(n.Data)), v.Version())
+
+	v.asyncRequestAppend(asyncRequest)
+	offset, _, isUnchanged, err = asyncRequest.WaitComplete()
+
+	return
+}
+
+func (v *Volume) doWriteRequest(n *needle.Needle) (offset uint64, size uint32, isUnchanged bool, err error) {
+	// glog.V(4).Infof("writing needle %s", needle.NewFileIdFromNeedle(v.Id, n).String())
+	if v.isFileUnchanged(n) {
+		size = n.DataSize
+		isUnchanged = true
+		return
+	}
+
+	// check whether existing needle cookie matches
+	nv, ok := v.nm.Get(n.Id)
+	if ok {
+		existingNeedle, _, _, existingNeedleReadErr := needle.ReadNeedleHeader(v.DataBackend, v.Version(), nv.Offset.ToAcutalOffset())
+		if existingNeedleReadErr != nil {
+			err = fmt.Errorf("reading existing needle: %v", existingNeedleReadErr)
+			return
+		}
+		if existingNeedle.Cookie != n.Cookie {
+			glog.V(0).Infof("write cookie mismatch: existing %x, new %x", existingNeedle.Cookie, n.Cookie)
+			err = fmt.Errorf("mismatching cookie %x", n.Cookie)
+			return
+		}
+	}
+
+	// append to dat file
+	n.AppendAtNs = uint64(time.Now().UnixNano())
+	if offset, size, _, err = n.Append(v.DataBackend, v.Version()); err != nil {
+		return
+	}
+	v.lastAppendAtNs = n.AppendAtNs
+
+	// add to needle map
+	if !ok || uint64(nv.Offset.ToAcutalOffset()) < offset {
+		if err = v.nm.Put(n.Id, ToOffset(int64(offset)), n.Size); err != nil {
+			glog.V(4).Infof("failed to save in needle map %d: %v", n.Id, err)
+		}
+	}
+	if v.lastModifiedTsSeconds < n.LastModified {
+		v.lastModifiedTsSeconds = n.LastModified
+	}
+	return
+}
+
+func (v *Volume) deleteNeedleDeprecated(n *needle.Needle) (uint32, error) {
 	glog.V(4).Infof("delete needle %s", needle.NewFileIdFromNeedle(v.Id, n).String())
 	v.dataFileAccessLock.Lock()
 	defer v.dataFileAccessLock.Unlock()
@@ -140,6 +204,37 @@ func (v *Volume) deleteNeedle(n *needle.Needle) (uint32, error) {
 	return 0, nil
 }
 
+func (v *Volume) deleteNeedle2(n *needle.Needle) (uint32, error) {
+	asyncRequest := needle.NewAsyncRequest(n, false, false)
+	asyncRequest.ActualSize = needle.GetActualSize(0, v.Version())
+
+	v.asyncRequestAppend(asyncRequest)
+	_, size, _, err := asyncRequest.WaitComplete()
+
+	return uint32(size), err
+}
+
+func (v *Volume) doDeleteRequest(n *needle.Needle) (uint32, error) {
+	glog.V(4).Infof("delete needle %s", needle.NewFileIdFromNeedle(v.Id, n).String())
+	nv, ok := v.nm.Get(n.Id)
+	//fmt.Println("key", n.Id, "volume offset", nv.Offset, "data_size", n.Size, "cached size", nv.Size)
+	if ok && nv.Size != TombstoneFileSize {
+		size := nv.Size
+		n.Data = nil
+		n.AppendAtNs = uint64(time.Now().UnixNano())
+		offset, _, _, err := n.Append(v.DataBackend, v.Version())
+		if err != nil {
+			return size, err
+		}
+		v.lastAppendAtNs = n.AppendAtNs
+		if err = v.nm.Delete(n.Id, ToOffset(int64(offset))); err != nil {
+			return size, err
+		}
+		return size, err
+	}
+	return 0, nil
+}
+
 // read fills in Needle content by looking up n.Id from NeedleMapper
 func (v *Volume) readNeedle(n *needle.Needle) (int, error) {
 	v.dataFileAccessLock.RLock()
@@ -176,6 +271,87 @@ func (v *Volume) readNeedle(n *needle.Needle) (int, error) {
 	return -1, ErrorNotFound
 }
 
+func (v *Volume) startWorker() {
+	go func() {
+		chanClosed := false
+		for {
+			// chan closed. go thread will exit
+			if chanClosed {
+				break
+			}
+			fsync := false
+			currentRequests := make([]*needle.AsyncRequest, 0, 128)
+			currentBytesToWrite := int64(0)
+			for {
+				request, ok := <-v.asyncRequestsChan
+				//volume may be closed
+				if !ok {
+					chanClosed = true
+					break
+				}
+				if MaxPossibleVolumeSize < v.ContentSize()+uint64(currentBytesToWrite+request.ActualSize) {
+					request.Complete(0, 0, false,
+						fmt.Errorf("volume size limit %d exceeded! current size is %d", MaxPossibleVolumeSize, v.ContentSize()))
+					break
+				}
+				currentRequests = append(currentRequests, request)
+				currentBytesToWrite += request.ActualSize
+				if request.Fsync {
+					fsync = true
+				}
+				// submit at most 4M bytes or 128 requests at one time to decrease request delay.
+				// it also need to break if there is no data in channel to avoid io hang.
+				if currentBytesToWrite >= 4*1024*1024 || len(currentRequests) >= 128 || len(v.asyncRequestsChan) == 0 {
+					break
+				}
+			}
+			if len(currentRequests) == 0 {
+				continue
+			}
+			v.dataFileAccessLock.Lock()
+			end, _, e := v.DataBackend.GetStat()
+			if e != nil {
+				for i := 0; i < len(currentRequests); i++ {
+					currentRequests[i].Complete(0, 0, false,
+						fmt.Errorf("cannot read current volume position: %v", e))
+				}
+				v.dataFileAccessLock.Unlock()
+				continue
+			}
+
+			for i := 0; i < len(currentRequests); i++ {
+				if currentRequests[i].IsWriteRequest {
+					offset, size, isUnchanged, err := v.doWriteRequest(currentRequests[i].N)
+					currentRequests[i].UpdateResult(offset, uint64(size), isUnchanged, err)
+				} else {
+					size, err := v.doDeleteRequest(currentRequests[i].N)
+					currentRequests[i].UpdateResult(0, uint64(size), false, err)
+				}
+			}
+
+			if fsync {
+				// if sync error, data is not reliable, we should mark the completed request as fail and rollback
+				if err := v.DataBackend.Sync(); err != nil {
+					// todo: this may generate dirty data or cause data inconsistent, may be weed need to panic?
+					if te := v.DataBackend.Truncate(end); te != nil {
+						glog.V(0).Infof("Failed to truncate %s back to %d with error: %v", v.DataBackend.Name(), end, te)
+					}
+					for i := 0; i < len(currentRequests); i++ {
+						if currentRequests[i].IsSucceed() {
+							currentRequests[i].UpdateResult(0, 0, false, err)
+						}
+					}
+				}
+			}
+
+			for i := 0; i < len(currentRequests); i++ {
+				currentRequests[i].Submit()
+			}
+			v.dataFileAccessLock.Unlock()
+		}
+	}()
+}
+
 type VolumeFileScanner interface {
 	VisitSuperBlock(super_block.SuperBlock) error
 	ReadNeedleBody() bool