// Join joins an existing cluster.
func (s *RaftServer) Join(peers []string) error {
	command := &raft.DefaultJoinCommand{
		Name:             s.raftServer.Name(),
		ConnectionString: "http://" + s.httpAddr,
	}

	var err error
	var b bytes.Buffer
	json.NewEncoder(&b).Encode(command)
	for _, m := range peers {
		if m == s.httpAddr {
			continue
		}
		target := fmt.Sprintf("http://%s/cluster/join", strings.TrimSpace(m))
		glog.V(0).Infoln("Attempting to connect to:", target)

		err = postFollowingOneRedirect(target, "application/json", &b)

		if err != nil {
			glog.V(0).Infoln("Post returned error: ", err.Error())
			if _, ok := err.(*url.Error); ok {
				// If we receive a network error try the next member
				continue
			}
		} else {
			return nil
		}
	}

	return errors.New("Could not connect to any cluster peers")
}

func newFilePart(fullPathFilename string) (ret FilePart, err error) {
	fh, openErr := os.Open(fullPathFilename)
	if openErr != nil {
		glog.V(0).Info("Failed to open file: ", fullPathFilename)
		return ret, openErr
	}
	ret.Reader = fh

	if fi, fiErr := fh.Stat(); fiErr != nil {
		glog.V(0).Info("Failed to stat file:", fullPathFilename)
		return ret, fiErr
	} else {
		ret.ModTime = fi.ModTime().UTC().Unix()
		ret.FileSize = fi.Size()
	}
	ext := strings.ToLower(path.Ext(fullPathFilename))
	ret.IsGzipped = ext == ".gz"
	if ret.IsGzipped {
		ret.FileName = fullPathFilename[0 : len(fullPathFilename)-3]
	}
	ret.FileName = fullPathFilename
	if ext != "" {
		ret.MimeType = mime.TypeByExtension(ext)
	}

	return ret, nil
}

func (g *Guard) checkJwt(w http.ResponseWriter, r *http.Request) error {
	if g.checkWhiteList(w, r) == nil {
		return nil
	}

	if len(g.SecretKey) == 0 {
		return nil
	}

	tokenStr := GetJwt(r)

	if tokenStr == "" {
		return ErrUnauthorized
	}

	// Verify the token
	token, err := DecodeJwt(g.SecretKey, tokenStr)
	if err != nil {
		glog.V(1).Infof("Token verification error from %s: %v", r.RemoteAddr, err)
		return ErrUnauthorized
	}
	if !token.Valid {
		glog.V(1).Infof("Token invliad from %s: %v", r.RemoteAddr, tokenStr)
		return ErrUnauthorized
	}

	glog.V(1).Infof("No permission from %s", r.RemoteAddr)
	return fmt.Errorf("No write permisson from %s", r.RemoteAddr)
}

func LoadNeedleMap(file *os.File) (*NeedleMap, error) {
	nm := NewNeedleMap(file)
	e := WalkIndexFile(file, func(key uint64, offset, size uint32) error {
		if key > nm.MaximumFileKey {
			nm.MaximumFileKey = key
		}
		nm.FileCounter++
		nm.FileByteCounter = nm.FileByteCounter + uint64(size)
		if offset > 0 {
			oldSize := nm.m.Set(Key(key), offset, size)
			glog.V(3).Infoln("reading key", key, "offset", offset*NeedlePaddingSize, "size", size, "oldSize", oldSize)
			if oldSize > 0 {
				nm.DeletionCounter++
				nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
			}
		} else {
			oldSize := nm.m.Delete(Key(key))
			glog.V(3).Infoln("removing key", key, "offset", offset*NeedlePaddingSize, "size", size, "oldSize", oldSize)
			nm.DeletionCounter++
			nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
		}
		return nil
	})
	glog.V(1).Infoln("max file key:", nm.MaximumFileKey)
	return nm, e
}

func (t *Topology) StartRefreshWritableVolumes(garbageThreshold string) {
	go func() {
		for {
			if t.IsLeader() {
				freshThreshHold := time.Now().Unix() - 3*t.pulse //3 times of sleep interval
				t.CollectDeadNodeAndFullVolumes(freshThreshHold, t.volumeSizeLimit)
			}
			time.Sleep(time.Duration(float32(t.pulse*1e3)*(1+rand.Float32())) * time.Millisecond)
		}
	}()
	go func(garbageThreshold string) {
		c := time.Tick(15 * time.Minute)
		if t.IsLeader() {
			for _ = range c {
				t.Vacuum(garbageThreshold)
			}
		}
	}(garbageThreshold)
	go func() {
		for {
			select {
			case v := <-t.chanFullVolumes:
				t.SetVolumeCapacityFull(v)
			case dn := <-t.chanRecoveredDataNodes:
				t.RegisterRecoveredDataNode(dn)
				glog.V(0).Infoln("DataNode", dn, "is back alive!")
			case dn := <-t.chanDeadDataNodes:
				t.UnRegisterDataNode(dn)
				glog.V(0).Infoln("DataNode", dn, "is dead!")
			}
		}
	}()
}

func distributedOperation(masterNode string, store *storage.Store, volumeId storage.VolumeId, op func(location operation.Location) bool) bool {
	if lookupResult, lookupErr := operation.Lookup(masterNode, volumeId.String()); lookupErr == nil {
		length := 0
		selfUrl := (store.Ip + ":" + strconv.Itoa(store.Port))
		results := make(chan bool)
		for _, location := range lookupResult.Locations {
			if location.Url != selfUrl {
				length++
				go func(location operation.Location, results chan bool) {
					results <- op(location)
				}(location, results)
			}
		}
		ret := true
		for i := 0; i < length; i++ {
			ret = ret && <-results
		}
		if volume := store.GetVolume(volumeId); volume != nil {
			if length+1 < volume.ReplicaPlacement.GetCopyCount() {
				glog.V(0).Infof("replicating opetations [%d] is less than volume's replication copy count [%d]", length+1, volume.ReplicaPlacement.GetCopyCount())
				ret = false
			}
		}
		return ret
	} else {
		glog.V(0).Infoln("Failed to lookup for", volumeId, lookupErr.Error())
	}
	return false
}

func (l *DiskLocation) loadExistingVolumes(needleMapKind NeedleMapType) {
	if dirs, err := ioutil.ReadDir(l.Directory); err == nil {
		for _, dir := range dirs {
			name := dir.Name()
			if !dir.IsDir() && strings.HasSuffix(name, ".dat") {
				collection := ""
				base := name[:len(name)-len(".dat")]
				i := strings.LastIndex(base, "_")
				if i > 0 {
					collection, base = base[0:i], base[i+1:]
				}
				if vid, err := NewVolumeId(base); err == nil {
					if l.volumes[vid] == nil {
						if v, e := NewVolume(l.Directory, collection, vid, needleMapKind, nil, nil); e == nil {
							l.volumes[vid] = v
							glog.V(0).Infof("data file %s, replicaPlacement=%s v=%d size=%d ttl=%s", l.Directory+"/"+name, v.ReplicaPlacement, v.Version(), v.Size(), v.Ttl.String())
						} else {
							glog.V(0).Infof("new volume %s error %s", name, e)
						}
					}
				}
			}
		}
	}
	glog.V(0).Infoln("Store started on dir:", l.Directory, "with", len(l.volumes), "volumes", "max", l.MaxVolumeCount)
}

func (vs *VolumeServer) tryHandleChunkedFile(n *storage.Needle, fileName string, w http.ResponseWriter, r *http.Request) (processed bool) {
	if !n.IsChunkedManifest() {
		return false
	}

	chunkManifest, e := operation.LoadChunkManifest(n.Data, n.IsGzipped())
	if e != nil {
		glog.V(0).Infof("load chunked manifest (%s) error: %v", r.URL.Path, e)
		return false
	}
	if fileName == "" && chunkManifest.Name != "" {
		fileName = chunkManifest.Name
	}
	mType := ""
	if chunkManifest.Mime != "" {
		mt := chunkManifest.Mime
		if !strings.HasPrefix(mt, "application/octet-stream") {
			mType = mt
		}
	}

	w.Header().Set("X-File-Store", "chunked")

	chunkedFileReader := &operation.ChunkedFileReader{
		Manifest: chunkManifest,
		Master:   vs.GetMasterNode(),
	}
	defer chunkedFileReader.Close()
	if e := writeResponseContent(fileName, mType, chunkedFileReader, w, r); e != nil {
		glog.V(2).Infoln("response write error:", e)
	}
	return true
}

func (vs *VolumeServer) DeleteHandler(w http.ResponseWriter, r *http.Request) {
	n := new(storage.Needle)
	vid, fid, _, _, _ := parseURLPath(r.URL.Path)
	volumeId, _ := storage.NewVolumeId(vid)
	n.ParsePath(fid)

	glog.V(2).Infoln("deleting", n)

	cookie := n.Cookie
	count, ok := vs.store.ReadVolumeNeedle(volumeId, n)

	if ok != nil {
		m := make(map[string]uint32)
		m["size"] = 0
		writeJsonQuiet(w, r, http.StatusNotFound, m)
		return
	}

	if n.Cookie != cookie {
		glog.V(0).Infoln("delete", r.URL.Path, "with unmaching cookie from ", r.RemoteAddr, "agent", r.UserAgent())
		return
	}

	ret := topology.ReplicatedDelete(vs.GetMasterNode(), vs.store, volumeId, n, r)

	if ret != 0 {
		m := make(map[string]uint32)
		m["size"] = uint32(count)
		writeJsonQuiet(w, r, http.StatusAccepted, m)
	} else {
		writeJsonError(w, r, http.StatusInternalServerError, errors.New("Deletion Failed."))
	}

}

//Experts only: takes multiple fid parameters. This function does not propagate deletes to replicas.
func (vs *VolumeServer) batchDeleteHandler(w http.ResponseWriter, r *http.Request) {
	r.ParseForm()
	var ret []operation.DeleteResult
	for _, fid := range r.Form["fid"] {
		vid, id_cookie, err := operation.ParseFileId(fid)
		if err != nil {
			ret = append(ret, operation.DeleteResult{Fid: fid, Error: err.Error()})
			continue
		}
		n := new(storage.Needle)
		volumeId, _ := storage.NewVolumeId(vid)
		n.ParsePath(id_cookie)
		glog.V(4).Infoln("batch deleting", n)
		cookie := n.Cookie
		if _, err := vs.store.ReadVolumeNeedle(volumeId, n); err != nil {
			ret = append(ret, operation.DeleteResult{Fid: fid, Error: err.Error()})
			continue
		}
		if n.Cookie != cookie {
			ret = append(ret, operation.DeleteResult{Fid: fid, Error: "File Random Cookie does not match."})
			glog.V(0).Infoln("deleting", fid, "with unmaching cookie from ", r.RemoteAddr, "agent", r.UserAgent())
			return
		}
		if size, err := vs.store.Delete(volumeId, n); err != nil {
			ret = append(ret, operation.DeleteResult{Fid: fid, Error: err.Error()})
		} else {
			ret = append(ret, operation.DeleteResult{Fid: fid, Size: int(size)})
		}
	}

	writeJsonQuiet(w, r, http.StatusAccepted, ret)
}

func (mn *MasterNodes) reset() {
	glog.V(4).Infof("Resetting master nodes: %v", mn)
	if len(mn.nodes) > 1 && mn.lastNode >= 0 {
		glog.V(0).Infof("Reset master %s from: %v", mn.nodes[mn.lastNode], mn.nodes)
		mn.lastNode = -mn.lastNode - 1
	}
}

func (vs *VolumeServer) PostHandler(w http.ResponseWriter, r *http.Request) {
	if e := r.ParseForm(); e != nil {
		glog.V(0).Infoln("form parse error:", e)
		writeJsonError(w, r, http.StatusBadRequest, e)
		return
	}
	vid, _, _, _, _ := parseURLPath(r.URL.Path)
	volumeId, ve := storage.NewVolumeId(vid)
	if ve != nil {
		glog.V(0).Infoln("NewVolumeId error:", ve)
		writeJsonError(w, r, http.StatusBadRequest, ve)
		return
	}
	needle, ne := storage.NewNeedle(r, vs.FixJpgOrientation)
	if ne != nil {
		writeJsonError(w, r, http.StatusBadRequest, ne)
		return
	}

	ret := operation.UploadResult{}
	size, errorStatus := topology.ReplicatedWrite(vs.GetMasterNode(),
		vs.store, volumeId, needle, r)
	httpStatus := http.StatusCreated
	if errorStatus != "" {
		httpStatus = http.StatusInternalServerError
		ret.Error = errorStatus
	}
	if needle.HasName() {
		ret.Name = string(needle.Name)
	}
	ret.Size = size
	writeJsonQuiet(w, r, httpStatus, ret)
}

func (mn *MasterNodes) findMaster() (string, error) {
	if len(mn.nodes) == 0 {
		return "", errors.New("No master node found!")
	}
	if mn.lastNode < 0 {
		for _, m := range mn.nodes {
			glog.V(4).Infof("Listing masters on %s", m)
			if masters, e := operation.ListMasters(m); e == nil {
				if len(masters) == 0 {
					continue
				}
				mn.nodes = append(masters, m)
				mn.lastNode = rand.Intn(len(mn.nodes))
				glog.V(2).Infof("current master nodes is %v", mn)
				break
			} else {
				glog.V(4).Infof("Failed listing masters on %s: %v", m, e)
			}
		}
	}
	if mn.lastNode < 0 {
		return "", errors.New("No master node available!")
	}
	return mn.nodes[mn.lastNode], nil
}

func (ms *MasterServer) dirJoinHandler(w http.ResponseWriter, r *http.Request) {
	body, err := ioutil.ReadAll(r.Body)
	if err != nil {
		writeJsonError(w, r, http.StatusBadRequest, err)
		return
	}
	joinMessage := &operation.JoinMessage{}
	if err = proto.Unmarshal(body, joinMessage); err != nil {
		writeJsonError(w, r, http.StatusBadRequest, err)
		return
	}
	if *joinMessage.Ip == "" {
		*joinMessage.Ip = r.RemoteAddr[0:strings.Index(r.RemoteAddr, ":")]
	}
	if glog.V(4) {
		if jsonData, jsonError := json.Marshal(joinMessage); jsonError != nil {
			glog.V(0).Infoln("json marshaling error: ", jsonError)
			writeJsonError(w, r, http.StatusBadRequest, jsonError)
			return
		} else {
			glog.V(4).Infoln("Proto size", len(body), "json size", len(jsonData), string(jsonData))
		}
	}

	ms.Topo.ProcessJoinMessage(joinMessage)
	writeJsonQuiet(w, r, http.StatusOK, operation.JoinResult{
		VolumeSizeLimit: uint64(ms.volumeSizeLimitMB) * 1024 * 1024,
		SecretKey:       string(ms.guard.SecretKey),
	})
}

// walks through the index file, calls fn function with each key, offset, size
// stops with the error returned by the fn function
func WalkIndexFile(r *os.File, fn func(key uint64, offset, size uint32) error) error {
	var readerOffset int64
	bytes := make([]byte, 16*RowsToRead)
	count, e := r.ReadAt(bytes, readerOffset)
	glog.V(3).Infoln("file", r.Name(), "readerOffset", readerOffset, "count", count, "e", e)
	readerOffset += int64(count)
	var (
		key          uint64
		offset, size uint32
		i            int
	)

	for count > 0 && e == nil || e == io.EOF {
		for i = 0; i+16 <= count; i += 16 {
			key, offset, size = idxFileEntry(bytes[i : i+16])
			if e = fn(key, offset, size); e != nil {
				return e
			}
		}
		if e == io.EOF {
			return nil
		}
		count, e = r.ReadAt(bytes, readerOffset)
		glog.V(3).Infoln("file", r.Name(), "readerOffset", readerOffset, "count", count, "e", e)
		readerOffset += int64(count)
	}
	return e
}

func batchVacuumVolumeCompact(vl *VolumeLayout, vid storage.VolumeId, locationlist *VolumeLocationList) bool {
	vl.removeFromWritable(vid)
	ch := make(chan bool, locationlist.Length())
	for index, dn := range locationlist.list {
		go func(index int, url string, vid storage.VolumeId) {
			glog.V(0).Infoln(index, "Start vacuuming", vid, "on", url)
			if e := vacuumVolume_Compact(url, vid); e != nil {
				glog.V(0).Infoln(index, "Error when vacuuming", vid, "on", url, e)
				ch <- false
			} else {
				glog.V(0).Infoln(index, "Complete vacuuming", vid, "on", url)
				ch <- true
			}
		}(index, dn.Url(), vid)
	}
	isVacuumSuccess := true
	for _ = range locationlist.list {
		select {
		case _ = <-ch:
		case <-time.After(30 * time.Minute):
			isVacuumSuccess = false
			break
		}
	}
	return isVacuumSuccess
}

// a workaround because http POST following redirection misses request body
func postFollowingOneRedirect(target string, contentType string, b *bytes.Buffer) error {
	backupReader := bytes.NewReader(b.Bytes())
	resp, err := http.Post(target, contentType, b)
	if err != nil {
		return err
	}
	defer resp.Body.Close()
	reply, _ := ioutil.ReadAll(resp.Body)
	statusCode := resp.StatusCode

	if statusCode == http.StatusMovedPermanently {
		var urlStr string
		if urlStr = resp.Header.Get("Location"); urlStr == "" {
			return fmt.Errorf("%d response missing Location header", resp.StatusCode)
		}

		glog.V(0).Infoln("Post redirected to ", urlStr)
		resp2, err2 := http.Post(urlStr, contentType, backupReader)
		if err2 != nil {
			return err2
		}
		defer resp2.Body.Close()
		reply, _ = ioutil.ReadAll(resp2.Body)
		statusCode = resp2.StatusCode
	}

	glog.V(0).Infoln("Post returned status: ", statusCode, string(reply))
	if statusCode != http.StatusOK {
		return errors.New(string(reply))
	}

	return nil
}

// trySynchronizing sync with remote volume server incrementally by
// make up the local and remote delta.
func (v *Volume) trySynchronizing(volumeServer string, masterMap CompactMap, compactRevision uint16) error {
	slaveIdxFile, err := os.Open(v.nm.IndexFileName())
	if err != nil {
		return fmt.Errorf("Open volume %d index file: %v", v.Id, err)
	}
	defer slaveIdxFile.Close()
	slaveMap, err := LoadNeedleMap(slaveIdxFile)
	if err != nil {
		return fmt.Errorf("Load volume %d index file: %v", v.Id, err)
	}
	var delta []NeedleValue
	if err := masterMap.Visit(func(needleValue NeedleValue) error {
		if needleValue.Key == 0 {
			return nil
		}
		if _, ok := slaveMap.Get(uint64(needleValue.Key)); ok {
			return nil // skip intersection
		}
		delta = append(delta, needleValue)
		return nil
	}); err != nil {
		return fmt.Errorf("Add master entry: %v", err)
	}
	if err := slaveMap.m.Visit(func(needleValue NeedleValue) error {
		if needleValue.Key == 0 {
			return nil
		}
		if _, ok := masterMap.Get(needleValue.Key); ok {
			return nil // skip intersection
		}
		needleValue.Size = 0
		delta = append(delta, needleValue)
		return nil
	}); err != nil {
		return fmt.Errorf("Remove local entry: %v", err)
	}

	// simulate to same ordering of remote .dat file needle entries
	sort.Sort(ByOffset(delta))

	// make up the delta
	fetchCount := 0
	volumeDataContentHandlerUrl := "http://" + volumeServer + "/admin/sync/data"
	for _, needleValue := range delta {
		if needleValue.Size == 0 {
			// remove file entry from local
			v.removeNeedle(needleValue.Key)
			continue
		}
		// add master file entry to local data file
		if err := v.fetchNeedle(volumeDataContentHandlerUrl, needleValue, compactRevision); err != nil {
			glog.V(0).Infof("Fetch needle %v from %s: %v", needleValue, volumeServer, err)
			return err
		}
		fetchCount++
	}
	glog.V(1).Infof("Fetched %d needles from %s", fetchCount, volumeServer)
	return nil
}

func ScanVolumeFile(dirname string, collection string, id VolumeId,
	needleMapKind NeedleMapType,
	visitSuperBlock func(SuperBlock) error,
	readNeedleBody bool,
	visitNeedle func(n *Needle, offset int64) error) (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)
	}
	if err = visitSuperBlock(v.SuperBlock); err != nil {
		return fmt.Errorf("Failed to process volume %d super block: %v", id, err)
	}

	version := v.Version()

	offset := int64(SuperBlockSize)
	n, rest, e := ReadNeedleHeader(v.dataFile, version, offset)
	if e != nil {
		err = fmt.Errorf("cannot read needle header: %v", e)
		return
	}
	for n != nil {
		if readNeedleBody {
			if err = n.ReadNeedleBody(v.dataFile, version, offset+int64(NeedleHeaderSize), rest); err != nil {
				glog.V(0).Infof("cannot read needle body: %v", err)
				//err = fmt.Errorf("cannot read needle body: %v", err)
				//return
			}
			if n.DataSize >= n.Size {
				// this should come from a bug reported on #87 and #93
				// fixed in v0.69
				// remove this whole "if" clause later, long after 0.69
				oldRest, oldSize := rest, n.Size
				padding := NeedlePaddingSize - ((n.Size + NeedleHeaderSize + NeedleChecksumSize) % NeedlePaddingSize)
				n.Size = 0
				rest = n.Size + NeedleChecksumSize + padding
				if rest%NeedlePaddingSize != 0 {
					rest += (NeedlePaddingSize - rest%NeedlePaddingSize)
				}
				glog.V(4).Infof("Adjusting n.Size %d=>0 rest:%d=>%d %+v", oldSize, oldRest, rest, n)
			}
		}
		if err = visitNeedle(n, offset); err != nil {
			glog.V(0).Infof("visit needle error: %v", err)
		}
		offset += int64(NeedleHeaderSize) + int64(rest)
		glog.V(4).Infof("==> new entry offset %d", offset)
		if n, rest, err = ReadNeedleHeader(v.dataFile, version, offset); err != nil {
			if err == io.EOF {
				return nil
			}
			return fmt.Errorf("cannot read needle header: %v", err)
		}
		glog.V(4).Infof("new entry needle size:%d rest:%d", n.Size, rest)
	}

	return
}

func (s *RaftServer) redirectToLeader(w http.ResponseWriter, req *http.Request) {
	if leader, e := s.topo.Leader(); e == nil {
		//http.StatusMovedPermanently does not cause http POST following redirection
		glog.V(0).Infoln("Redirecting to", http.StatusMovedPermanently, "http://"+leader+req.URL.Path)
		http.Redirect(w, req, "http://"+leader+req.URL.Path, http.StatusMovedPermanently)
	} else {
		glog.V(0).Infoln("Error: Leader Unknown")
		http.Error(w, "Leader unknown", http.StatusInternalServerError)
	}
}