Merge branch 'filer2_development'

weed/command/filer.go

@@ -2,17 +2,18 @@ package command
 
 import (
 	"net/http"
-	"os"
 	"strconv"
 	"time"
 
+	"github.com/chrislusf/seaweedfs/weed/filer2"
 	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
 	"github.com/chrislusf/seaweedfs/weed/server"
 	"github.com/chrislusf/seaweedfs/weed/util"
 	"github.com/soheilhy/cmux"
-	"google.golang.org/grpc/reflection"
 	"google.golang.org/grpc"
-	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"google.golang.org/grpc/reflection"
+	"strings"
 )
 
 var (
@@ -20,50 +21,35 @@ var (
 )
 
 type FilerOptions struct {
-	master                  *string
+	masters                 *string
 	ip                      *string
 	port                    *int
 	publicPort              *int
 	collection              *string
 	defaultReplicaPlacement *string
-	dir                     *string
 	redirectOnRead          *bool
 	disableDirListing       *bool
-	confFile                *string
 	maxMB                   *int
 	secretKey               *string
-	cassandra_server        *string
-	cassandra_keyspace      *string
-	redis_server            *string
-	redis_password          *string
-	redis_database          *int
 }
 
 func init() {
 	cmdFiler.Run = runFiler // break init cycle
-	f.master = cmdFiler.Flag.String("master", "localhost:9333", "master server location")
+	f.masters = cmdFiler.Flag.String("master", "localhost:9333", "comma-separated master servers")
 	f.collection = cmdFiler.Flag.String("collection", "", "all data will be stored in this collection")
 	f.ip = cmdFiler.Flag.String("ip", "", "filer server http listen ip address")
 	f.port = cmdFiler.Flag.Int("port", 8888, "filer server http listen port")
 	f.publicPort = cmdFiler.Flag.Int("port.public", 0, "port opened to public")
-	f.dir = cmdFiler.Flag.String("dir", os.TempDir(), "directory to store meta data")
 	f.defaultReplicaPlacement = cmdFiler.Flag.String("defaultReplicaPlacement", "000", "default replication type if not specified")
 	f.redirectOnRead = cmdFiler.Flag.Bool("redirectOnRead", false, "whether proxy or redirect to volume server during file GET request")
 	f.disableDirListing = cmdFiler.Flag.Bool("disableDirListing", false, "turn off directory listing")
-	f.confFile = cmdFiler.Flag.String("confFile", "", "json encoded filer conf file")
 	f.maxMB = cmdFiler.Flag.Int("maxMB", 32, "split files larger than the limit")
-	f.cassandra_server = cmdFiler.Flag.String("cassandra.server", "", "host[:port] of the cassandra server")
-	f.cassandra_keyspace = cmdFiler.Flag.String("cassandra.keyspace", "seaweed", "keyspace of the cassandra server")
-	f.redis_server = cmdFiler.Flag.String("redis.server", "", "comma separated host:port[,host2:port2]* of the redis server, e.g., 127.0.0.1:6379")
-	f.redis_password = cmdFiler.Flag.String("redis.password", "", "password in clear text")
-	f.redis_database = cmdFiler.Flag.Int("redis.database", 0, "the database on the redis server")
 	f.secretKey = cmdFiler.Flag.String("secure.secret", "", "secret to encrypt Json Web Token(JWT)")
-
 }
 
 var cmdFiler = &Command{
-	UsageLine: "filer -port=8888 -dir=/tmp -master=<ip:port>",
-	Short:     "start a file server that points to a master server",
+	UsageLine: "filer -port=8888 -master=<ip:port>[,<ip:port>]*",
+	Short:     "start a file server that points to a master server, or a list of master servers",
 	Long: `start a file server which accepts REST operation for any files.
 
 	//create or overwrite the file, the directories /path/to will be automatically created
@@ -75,20 +61,15 @@ var cmdFiler = &Command{
 	//return a json format subdirectory and files listing
 	GET /path/to/
 
-  Current <fullpath~fileid> mapping metadata store is local embedded leveldb.
-  It should be highly scalable to hundreds of millions of files on a modest machine.
+	The configuration file "filer.toml" is read from ".", "$HOME/.seaweedfs/", or "/etc/seaweedfs/", in that order.
 
-  Future we will ensure it can avoid of being SPOF.
+	The following are example filer.toml configuration file.
 
-  `,
+` + filer2.FILER_TOML_EXAMPLE + "\n",
 }
 
 func runFiler(cmd *Command, args []string) bool {
 
-	if err := util.TestFolderWritable(*f.dir); err != nil {
-		glog.Fatalf("Check Meta Folder (-dir) Writable %s : %s", *f.dir, err)
-	}
-
 	f.start()
 
 	return true
@@ -103,14 +84,13 @@ func (fo *FilerOptions) start() {
 		publicVolumeMux = http.NewServeMux()
 	}
 
+	masters := *f.masters
+
 	fs, nfs_err := weed_server.NewFilerServer(defaultMux, publicVolumeMux,
-		*fo.ip, *fo.port, *fo.master, *fo.dir, *fo.collection,
+		*fo.ip, *fo.port, strings.Split(masters, ","), *fo.collection,
 		*fo.defaultReplicaPlacement, *fo.redirectOnRead, *fo.disableDirListing,
-		*fo.confFile,
 		*fo.maxMB,
 		*fo.secretKey,
-		*fo.cassandra_server, *fo.cassandra_keyspace,
-		*fo.redis_server, *fo.redis_password, *fo.redis_database,
 	)
 	if nfs_err != nil {
 		glog.Fatalf("Filer startup error: %v", nfs_err)

weed/command/filer_copy.go

@@ -9,8 +9,15 @@ import (
 	"strings"
 
 	"github.com/chrislusf/seaweedfs/weed/operation"
-	filer_operation "github.com/chrislusf/seaweedfs/weed/operation/filer"
 	"github.com/chrislusf/seaweedfs/weed/security"
+	"path"
+	"net/http"
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"strconv"
+	"io"
+	"time"
+	"google.golang.org/grpc"
+	"context"
 )
 
 var (
@@ -68,20 +75,20 @@ func runCopy(cmd *Command, args []string) bool {
 		return false
 	}
 	filerDestination := args[len(args)-1]
-	fileOrDirs := args[0 : len(args)-1]
+	fileOrDirs := args[0: len(args)-1]
 
 	filerUrl, err := url.Parse(filerDestination)
 	if err != nil {
 		fmt.Printf("The last argument should be a URL on filer: %v\n", err)
 		return false
 	}
-	path := filerUrl.Path
-	if !strings.HasSuffix(path, "/") {
-		path = path + "/"
+	urlPath := filerUrl.Path
+	if !strings.HasSuffix(urlPath, "/") {
+		urlPath = urlPath + "/"
 	}
 
 	for _, fileOrDir := range fileOrDirs {
-		if !doEachCopy(fileOrDir, filerUrl.Host, path) {
+		if !doEachCopy(fileOrDir, filerUrl.Host, urlPath) {
 			return false
 		}
 	}
@@ -91,14 +98,14 @@ func runCopy(cmd *Command, args []string) bool {
 func doEachCopy(fileOrDir string, host string, path string) bool {
 	f, err := os.Open(fileOrDir)
 	if err != nil {
-		fmt.Printf("Failed to open file %s: %v", fileOrDir, err)
+		fmt.Printf("Failed to open file %s: %v\n", fileOrDir, err)
 		return false
 	}
 	defer f.Close()
 
 	fi, err := f.Stat()
 	if err != nil {
-		fmt.Printf("Failed to get stat for file %s: %v", fileOrDir, err)
+		fmt.Printf("Failed to get stat for file %s: %v\n", fileOrDir, err)
 		return false
 	}
 
@@ -120,28 +127,199 @@ func doEachCopy(fileOrDir string, host string, path string) bool {
 		}
 	}
 
-	parts, err := operation.NewFileParts([]string{fileOrDir})
+	// find the chunk count
+	chunkSize := int64(*copy.maxMB * 1024 * 1024)
+	chunkCount := 1
+	if chunkSize > 0 && fi.Size() > chunkSize {
+		chunkCount = int(fi.Size()/chunkSize) + 1
+	}
+
+	if chunkCount == 1 {
+		return uploadFileAsOne(host, path, f, fi)
+	}
+
+	return uploadFileInChunks(host, path, f, fi, chunkCount, chunkSize)
+}
+
+func uploadFileAsOne(filerUrl string, urlFolder string, f *os.File, fi os.FileInfo) bool {
+
+	// upload the file content
+	fileName := filepath.Base(f.Name())
+	mimeType := detectMimeType(f)
+	isGzipped := isGzipped(fileName)
+
+	var chunks []*filer_pb.FileChunk
+
+	if fi.Size() > 0 {
+
+		// assign a volume
+		assignResult, err := operation.Assign(*copy.master, &operation.VolumeAssignRequest{
+			Count:       1,
+			Replication: *copy.replication,
+			Collection:  *copy.collection,
+			Ttl:         *copy.ttl,
+		})
+		if err != nil {
+			fmt.Printf("Failed to assign from %s: %v\n", *copy.master, err)
+		}
+
+		targetUrl := "http://" + assignResult.Url + "/" + assignResult.Fid
+
+		uploadResult, err := operation.Upload(targetUrl, fileName, f, isGzipped, mimeType, nil, "")
+		if err != nil {
+			fmt.Printf("upload data %v to %s: %v\n", fileName, targetUrl, err)
+			return false
+		}
+		if uploadResult.Error != "" {
+			fmt.Printf("upload %v to %s result: %v\n", fileName, targetUrl, uploadResult.Error)
+			return false
+		}
+		fmt.Printf("uploaded %s to %s\n", fileName, targetUrl)
+
+		chunks = append(chunks, &filer_pb.FileChunk{
+			FileId: assignResult.Fid,
+			Offset: 0,
+			Size:   uint64(uploadResult.Size),
+			Mtime:  time.Now().UnixNano(),
+		})
+
+		fmt.Printf("copied %s => http://%s%s%s\n", fileName, filerUrl, urlFolder, fileName)
+	}
+
+	if err := withFilerClient(filerUrl, func(client filer_pb.SeaweedFilerClient) error {
+		request := &filer_pb.CreateEntryRequest{
+			Directory: urlFolder,
+			Entry: &filer_pb.Entry{
+				Name: fileName,
+				Attributes: &filer_pb.FuseAttributes{
+					Crtime:   time.Now().Unix(),
+					Mtime:    time.Now().Unix(),
+					Gid:      uint32(os.Getgid()),
+					Uid:      uint32(os.Getuid()),
+					FileSize: uint64(fi.Size()),
+					FileMode: uint32(fi.Mode()),
+					Mime:     mimeType,
+				},
+				Chunks: chunks,
+			},
+		}
+
+		if _, err := client.CreateEntry(context.Background(), request); err != nil {
+			return fmt.Errorf("update fh: %v", err)
+		}
+		return nil
+	}); err != nil {
+		fmt.Printf("upload data %v to http://%s%s%s: %v\n", fileName, filerUrl, urlFolder, fileName, err)
+		return false
+	}
+
+	return true
+}
+
+func uploadFileInChunks(filerUrl string, urlFolder string, f *os.File, fi os.FileInfo, chunkCount int, chunkSize int64) bool {
+
+	fileName := filepath.Base(f.Name())
+	mimeType := detectMimeType(f)
+
+	var chunks []*filer_pb.FileChunk
+
+	for i := int64(0); i < int64(chunkCount); i++ {
+
+		// assign a volume
+		assignResult, err := operation.Assign(*copy.master, &operation.VolumeAssignRequest{
+			Count:       1,
+			Replication: *copy.replication,
+			Collection:  *copy.collection,
+			Ttl:         *copy.ttl,
+		})
 		if err != nil {
-		fmt.Printf("Failed to read file %s: %v", fileOrDir, err)
+			fmt.Printf("Failed to assign from %s: %v\n", *copy.master, err)
 		}
 
-	results, err := operation.SubmitFiles(*copy.master, parts,
-		*copy.replication, *copy.collection, "",
-		*copy.ttl, *copy.maxMB, copy.secret)
+		targetUrl := "http://" + assignResult.Url + "/" + assignResult.Fid
+
+		uploadResult, err := operation.Upload(targetUrl,
+			fileName+"-"+strconv.FormatInt(i+1, 10),
+			io.LimitReader(f, chunkSize),
+			false, "application/octet-stream", nil, "")
 		if err != nil {
-		fmt.Printf("Failed to submit file %s: %v", fileOrDir, err)
+			fmt.Printf("upload data %v to %s: %v\n", fileName, targetUrl, err)
+			return false
+		}
+		if uploadResult.Error != "" {
+			fmt.Printf("upload %v to %s result: %v\n", fileName, targetUrl, uploadResult.Error)
+			return false
+		}
+		chunks = append(chunks, &filer_pb.FileChunk{
+			FileId: assignResult.Fid,
+			Offset: i * chunkSize,
+			Size:   uint64(uploadResult.Size),
+			Mtime:  time.Now().UnixNano(),
+		})
+		fmt.Printf("uploaded %s-%d to %s [%d,%d)\n", fileName, i+1, targetUrl, i*chunkSize, i*chunkSize+int64(uploadResult.Size))
 	}
 
-	if strings.HasSuffix(path, "/") {
-		path = path + fi.Name()
+	if err := withFilerClient(filerUrl, func(client filer_pb.SeaweedFilerClient) error {
+		request := &filer_pb.CreateEntryRequest{
+			Directory: urlFolder,
+			Entry: &filer_pb.Entry{
+				Name: fileName,
+				Attributes: &filer_pb.FuseAttributes{
+					Crtime:   time.Now().Unix(),
+					Mtime:    time.Now().Unix(),
+					Gid:      uint32(os.Getgid()),
+					Uid:      uint32(os.Getuid()),
+					FileSize: uint64(fi.Size()),
+					FileMode: uint32(fi.Mode()),
+					Mime:     mimeType,
+				},
+				Chunks: chunks,
+			},
 		}
 
-	if err = filer_operation.RegisterFile(host, path, results[0].Fid, copy.secret); err != nil {
-		fmt.Printf("Failed to register file %s on %s: %v", fileOrDir, host, err)
+		if _, err := client.CreateEntry(context.Background(), request); err != nil {
+			return fmt.Errorf("update fh: %v", err)
+		}
+		return nil
+	}); err != nil {
+		fmt.Printf("upload data %v to http://%s%s%s: %v\n", fileName, filerUrl, urlFolder, fileName, err)
 		return false
 	}
 
-	fmt.Printf("Copy %s => http://%s%s\n", fileOrDir, host, path)
+	fmt.Printf("copied %s => http://%s%s%s\n", fileName, filerUrl, urlFolder, fileName)
 
 	return true
 }
+
+func isGzipped(filename string) bool {
+	return strings.ToLower(path.Ext(filename)) == ".gz"
+}
+
+func detectMimeType(f *os.File) string {
+	head := make([]byte, 512)
+	f.Seek(0, 0)
+	n, err := f.Read(head)
+	if err == io.EOF {
+		return ""
+	}
+	if err != nil {
+		fmt.Printf("read head of %v: %v\n", f.Name(), err)
+		return "application/octet-stream"
+	}
+	f.Seek(0, 0)
+	mimeType := http.DetectContentType(head[:n])
+	return mimeType
+}
+
+func withFilerClient(filerAddress string, fn func(filer_pb.SeaweedFilerClient) error) error {
+
+	grpcConnection, err := grpc.Dial(filerAddress, grpc.WithInsecure())
+	if err != nil {
+		return fmt.Errorf("fail to dial %s: %v", filerAddress, err)
+	}
+	defer grpcConnection.Close()
+
+	client := filer_pb.NewSeaweedFilerClient(grpcConnection)
+
+	return fn(client)
+}

weed/command/mount.go

@@ -3,6 +3,9 @@ package command
 type MountOptions struct {
 	filer            *string
 	dir              *string
+	collection       *string
+	replication      *string
+	chunkSizeLimitMB *int
 }
 
 var (
@@ -11,9 +14,11 @@ var (
 
 func init() {
 	cmdMount.Run = runMount // break init cycle
-	cmdMount.IsDebug = cmdMount.Flag.Bool("debug", false, "verbose debug information")
 	mountOptions.filer = cmdMount.Flag.String("filer", "localhost:8888", "weed filer location")
 	mountOptions.dir = cmdMount.Flag.String("dir", ".", "mount weed filer to this directory")
+	mountOptions.collection = cmdMount.Flag.String("collection", "", "collection to create the files")
+	mountOptions.replication = cmdMount.Flag.String("replication", "000", "replication to create to files")
+	mountOptions.chunkSizeLimitMB = cmdMount.Flag.Int("chunkSizeLimitMB", 16, "local write buffer size, also chunk large files")
 }
 
 var cmdMount = &Command{

weed/command/mount_std.go

@@ -8,9 +8,9 @@ import (
 
 	"bazil.org/fuse"
 	"bazil.org/fuse/fs"
+	"github.com/chrislusf/seaweedfs/weed/filesys"
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/util"
-	"github.com/chrislusf/seaweedfs/weed/filesys"
 )
 
 func runMount(cmd *Command, args []string) bool {
@@ -19,6 +19,10 @@ func runMount(cmd *Command, args []string) bool {
 		fmt.Printf("Please specify the mount directory via \"-dir\"")
 		return false
 	}
+	if *mountOptions.chunkSizeLimitMB <= 0 {
+		fmt.Printf("Please specify a reasonable buffer size.")
+		return false
+	}
 
 	fuse.Unmount(*mountOptions.dir)
 
@@ -47,7 +51,8 @@ func runMount(cmd *Command, args []string) bool {
 		c.Close()
 	})
 
-	err = fs.Serve(c, filesys.NewSeaweedFileSystem(*mountOptions.filer))
+	err = fs.Serve(c, filesys.NewSeaweedFileSystem(
+		*mountOptions.filer, *mountOptions.collection, *mountOptions.replication, *mountOptions.chunkSizeLimitMB))
 	if err != nil {
 		fuse.Unmount(*mountOptions.dir)
 	}

weed/command/server.go

@@ -83,21 +83,13 @@ var (
 
 func init() {
 	serverOptions.cpuprofile = cmdServer.Flag.String("cpuprofile", "", "cpu profile output file")
-	filerOptions.master = cmdServer.Flag.String("filer.master", "", "default to current master server")
 	filerOptions.collection = cmdServer.Flag.String("filer.collection", "", "all data will be stored in this collection")
 	filerOptions.port = cmdServer.Flag.Int("filer.port", 8888, "filer server http listen port")
 	filerOptions.publicPort = cmdServer.Flag.Int("filer.port.public", 0, "filer server public http listen port")
-	filerOptions.dir = cmdServer.Flag.String("filer.dir", "", "directory to store meta data, default to a 'filer' sub directory of what -dir is specified")
 	filerOptions.defaultReplicaPlacement = cmdServer.Flag.String("filer.defaultReplicaPlacement", "", "Default replication type if not specified during runtime.")
 	filerOptions.redirectOnRead = cmdServer.Flag.Bool("filer.redirectOnRead", false, "whether proxy or redirect to volume server during file GET request")
 	filerOptions.disableDirListing = cmdServer.Flag.Bool("filer.disableDirListing", false, "turn off directory listing")
-	filerOptions.confFile = cmdServer.Flag.String("filer.confFile", "", "json encoded filer conf file")
 	filerOptions.maxMB = cmdServer.Flag.Int("filer.maxMB", 32, "split files larger than the limit")
-	filerOptions.cassandra_server = cmdServer.Flag.String("filer.cassandra.server", "", "host[:port] of the cassandra server")
-	filerOptions.cassandra_keyspace = cmdServer.Flag.String("filer.cassandra.keyspace", "seaweed", "keyspace of the cassandra server")
-	filerOptions.redis_server = cmdServer.Flag.String("filer.redis.server", "", "host:port of the redis server, e.g., 127.0.0.1:6379")
-	filerOptions.redis_password = cmdServer.Flag.String("filer.redis.password", "", "redis password in clear text")
-	filerOptions.redis_database = cmdServer.Flag.Int("filer.redis.database", 0, "the database on the redis server")
 }
 
 func runServer(cmd *Command, args []string) bool {
@@ -115,7 +107,7 @@ func runServer(cmd *Command, args []string) bool {
 		*isStartingFiler = true
 	}
 
-	*filerOptions.master = *serverIp + ":" + strconv.Itoa(*masterPort)
+	master := *serverIp + ":" + strconv.Itoa(*masterPort)
 	filerOptions.ip = serverIp
 
 	if *filerOptions.defaultReplicaPlacement == "" {
@@ -157,15 +149,6 @@ func runServer(cmd *Command, args []string) bool {
 	if *masterMetaFolder == "" {
 		*masterMetaFolder = folders[0]
 	}
-	if *isStartingFiler {
-		if *filerOptions.dir == "" {
-			*filerOptions.dir = *masterMetaFolder + "/filer"
-			os.MkdirAll(*filerOptions.dir, 0700)
-		}
-		if err := util.TestFolderWritable(*filerOptions.dir); err != nil {
-			glog.Fatalf("Check Mapping Meta Folder (-filer.dir=\"%s\") Writable: %s", *filerOptions.dir, err)
-		}
-	}
 	if err := util.TestFolderWritable(*masterMetaFolder); err != nil {
 		glog.Fatalf("Check Meta Folder (-mdir=\"%s\") Writable: %s", *masterMetaFolder, err)
 	}
@@ -267,7 +250,7 @@ func runServer(cmd *Command, args []string) bool {
 		*serverIp, *volumePort, *volumeServerPublicUrl,
 		folders, maxCounts,
 		volumeNeedleMapKind,
-		*serverIp+":"+strconv.Itoa(*masterPort), *volumePulse, *serverDataCenter, *serverRack,
+		[]string{master}, *volumePulse, *serverDataCenter, *serverRack,
 		serverWhiteList, *volumeFixJpgOrientation, *volumeReadRedirect,
 	)
 

weed/command/volume.go

@@ -26,7 +26,7 @@ type VolumeServerOptions struct {
 	ip                    *string
 	publicUrl             *string
 	bindIp                *string
-	master                *string
+	masters               *string
 	pulseSeconds          *int
 	idleConnectionTimeout *int
 	maxCpu                *int
@@ -47,7 +47,7 @@ func init() {
 	v.ip = cmdVolume.Flag.String("ip", "", "ip or server name")
 	v.publicUrl = cmdVolume.Flag.String("publicUrl", "", "Publicly accessible address")
 	v.bindIp = cmdVolume.Flag.String("ip.bind", "0.0.0.0", "ip address to bind to")
-	v.master = cmdVolume.Flag.String("mserver", "localhost:9333", "master server location")
+	v.masters = cmdVolume.Flag.String("mserver", "localhost:9333", "comma-separated master servers")
 	v.pulseSeconds = cmdVolume.Flag.Int("pulseSeconds", 5, "number of seconds between heartbeats, must be smaller than or equal to the master's setting")
 	v.idleConnectionTimeout = cmdVolume.Flag.Int("idleTimeout", 30, "connection idle seconds")
 	v.maxCpu = cmdVolume.Flag.Int("maxCpu", 0, "maximum number of CPUs. 0 means all available CPUs")
@@ -132,11 +132,14 @@ func runVolume(cmd *Command, args []string) bool {
 	case "btree":
 		volumeNeedleMapKind = storage.NeedleMapBtree
 	}
+
+	masters := *v.masters
+
 	volumeServer := weed_server.NewVolumeServer(volumeMux, publicVolumeMux,
 		*v.ip, *v.port, *v.publicUrl,
 		v.folders, v.folderMaxLimits,
 		volumeNeedleMapKind,
-		*v.master, *v.pulseSeconds, *v.dataCenter, *v.rack,
+		strings.Split(masters, ","), *v.pulseSeconds, *v.dataCenter, *v.rack,
 		v.whiteList,
 		*v.fixJpgOrientation, *v.readRedirect,
 	)

weed/filer/cassandra_store/cassandra_store.go

@@ -1,96 +0,0 @@
-package cassandra_store
-
-import (
-	"fmt"
-	"strings"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/glog"
-
-	"github.com/gocql/gocql"
-)
-
-/*
-
-Basically you need a table just like this:
-
-CREATE TABLE seaweed_files (
-   path varchar,
-   fids list<varchar>,
-   PRIMARY KEY (path)
-);
-
-Need to match flat_namespace.FlatNamespaceStore interface
-	Put(fullFileName string, fid string) (err error)
-	Get(fullFileName string) (fid string, err error)
-	Delete(fullFileName string) (fid string, err error)
-
-*/
-type CassandraStore struct {
-	cluster *gocql.ClusterConfig
-	session *gocql.Session
-}
-
-func NewCassandraStore(keyspace string, hosts string) (c *CassandraStore, err error) {
-	c = &CassandraStore{}
-	s := strings.Split(hosts, ",")
-	if len(s) == 1 {
-		glog.V(2).Info("Only one cassandra node to connect! A cluster is Recommended! Now using:", string(hosts))
-		c.cluster = gocql.NewCluster(hosts)
-	} else if len(s) > 1 {
-		c.cluster = gocql.NewCluster(s...)
-	}
-	c.cluster.Keyspace = keyspace
-	c.cluster.Consistency = gocql.LocalQuorum
-	c.session, err = c.cluster.CreateSession()
-	if err != nil {
-		glog.V(0).Infof("Failed to open cassandra store, hosts %v, keyspace %s", hosts, keyspace)
-	}
-	return
-}
-
-func (c *CassandraStore) Put(fullFileName string, fid string) (err error) {
-	var input []string
-	input = append(input, fid)
-	if err := c.session.Query(
-		`INSERT INTO seaweed_files (path, fids) VALUES (?, ?)`,
-		fullFileName, input).Exec(); err != nil {
-		glog.V(0).Infof("Failed to save file %s with id %s: %v", fullFileName, fid, err)
-		return err
-	}
-	return nil
-}
-func (c *CassandraStore) Get(fullFileName string) (fid string, err error) {
-	var output []string
-	if err := c.session.Query(
-		`select fids FROM seaweed_files WHERE path = ? LIMIT 1`,
-		fullFileName).Consistency(gocql.One).Scan(&output); err != nil {
-		if err != gocql.ErrNotFound {
-			glog.V(0).Infof("Failed to find file %s: %v", fullFileName, fid, err)
-			return "", filer.ErrNotFound
-		}
-	}
-	if len(output) == 0 {
-		return "", fmt.Errorf("No file id found for %s", fullFileName)
-	}
-	return output[0], nil
-}
-
-// Currently the fid is not returned
-func (c *CassandraStore) Delete(fullFileName string) (err error) {
-	if err := c.session.Query(
-		`DELETE FROM seaweed_files WHERE path = ?`,
-		fullFileName).Exec(); err != nil {
-		if err != gocql.ErrNotFound {
-			glog.V(0).Infof("Failed to delete file %s: %v", fullFileName, err)
-		}
-		return err
-	}
-	return nil
-}
-
-func (c *CassandraStore) Close() {
-	if c.session != nil {
-		c.session.Close()
-	}
-}

weed/filer/cassandra_store/schema.cql

@@ -1,22 +0,0 @@
-/*
-
-Here is the CQL to create the table.CassandraStore
-
-Optionally you can adjust the keyspace name and replication settings.
-
-For production server, very likely you want to set replication_factor to 3
-
-*/
-
-create keyspace seaweed WITH replication = {
-  'class':'SimpleStrategy',
-  'replication_factor':1
-};
-
-use seaweed;
-
-CREATE TABLE seaweed_files (
-   path varchar,
-   fids list<varchar>,
-   PRIMARY KEY (path)
-);

weed/filer/embedded_filer/design.txt

@@ -1,26 +0,0 @@
-Design Assumptions:
-1. the number of directories are magnitudely smaller than the number of files
-2. unlimited number of files under any directories
-Phylosophy:
-  metadata for directories and files should be separated
-Design:
-  Store directories in normal map
-    all of directories hopefully all be in memory
-    efficient to move/rename/list_directories
-  Log directory changes to append only log file
-  Store files in sorted string table in <dir_id/filename> format
-    efficient to list_files, just simple iterator
-    efficient to locate files, binary search
-
-Testing:
-1. starting server, "weed server -filer=true"
-2. posting files to different folders
-curl -F "filename=@design.txt"   "http://localhost:8888/sources/"
-curl -F "filename=@design.txt"   "http://localhost:8888/design/"
-curl -F "filename=@directory.go" "http://localhost:8888/sources/weed/go/"
-curl -F "filename=@directory.go" "http://localhost:8888/sources/testing/go/"
-curl -F "filename=@filer.go"     "http://localhost:8888/sources/weed/go/"
-curl -F "filename=@filer_in_leveldb.go"     "http://localhost:8888/sources/weed/go/"
-curl "http://localhost:8888/?pretty=y"
-curl "http://localhost:8888/sources/weed/go/?pretty=y"
-curl "http://localhost:8888/sources/weed/go/?pretty=y"

weed/filer/embedded_filer/directory.go

@@ -1,15 +0,0 @@
-package embedded_filer
-
-import (
-	"github.com/chrislusf/seaweedfs/weed/filer"
-)
-
-type DirectoryManager interface {
-	FindDirectory(dirPath string) (DirectoryId, error)
-	ListDirectories(dirPath string) (dirs []filer.DirectoryName, err error)
-	MakeDirectory(currentDirPath string, dirName string) (DirectoryId, error)
-	MoveUnderDirectory(oldDirPath string, newParentDirPath string) error
-	DeleteDirectory(dirPath string) error
-	//functions used by FUSE
-	FindDirectoryById(DirectoryId, error)
-}

weed/filer/embedded_filer/directory_in_map.go

@@ -1,312 +0,0 @@
-package embedded_filer
-
-import (
-	"bufio"
-	"fmt"
-	"io"
-	"os"
-	"path/filepath"
-	"strconv"
-	"strings"
-	"sync"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/util"
-)
-
-var writeLock sync.Mutex //serialize changes to dir.log
-
-type DirectoryId int32
-
-type DirectoryEntryInMap struct {
-	sync.Mutex
-	Name           string
-	Parent         *DirectoryEntryInMap
-	subDirectories map[string]*DirectoryEntryInMap
-	Id             DirectoryId
-}
-
-func (de *DirectoryEntryInMap) getChild(dirName string) (*DirectoryEntryInMap, bool) {
-	de.Lock()
-	defer de.Unlock()
-	child, ok := de.subDirectories[dirName]
-	return child, ok
-}
-func (de *DirectoryEntryInMap) addChild(dirName string, child *DirectoryEntryInMap) {
-	de.Lock()
-	defer de.Unlock()
-	de.subDirectories[dirName] = child
-}
-func (de *DirectoryEntryInMap) removeChild(dirName string) {
-	de.Lock()
-	defer de.Unlock()
-	delete(de.subDirectories, dirName)
-}
-func (de *DirectoryEntryInMap) hasChildren() bool {
-	de.Lock()
-	defer de.Unlock()
-	return len(de.subDirectories) > 0
-}
-func (de *DirectoryEntryInMap) children() (dirNames []filer.DirectoryName) {
-	de.Lock()
-	defer de.Unlock()
-	for k, _ := range de.subDirectories {
-		dirNames = append(dirNames, filer.DirectoryName(k))
-	}
-	return dirNames
-}
-
-type DirectoryManagerInMap struct {
-	Root      *DirectoryEntryInMap
-	max       DirectoryId
-	logFile   *os.File
-	isLoading bool
-}
-
-func (dm *DirectoryManagerInMap) newDirectoryEntryInMap(parent *DirectoryEntryInMap, name string) (d *DirectoryEntryInMap, err error) {
-	d = &DirectoryEntryInMap{Name: name, Parent: parent, subDirectories: make(map[string]*DirectoryEntryInMap)}
-	var parts []string
-	for p := d; p != nil && p.Name != ""; p = p.Parent {
-		parts = append(parts, p.Name)
-	}
-	n := len(parts)
-	if n <= 0 {
-		return nil, fmt.Errorf("Failed to create folder %s/%s", parent.Name, name)
-	}
-	for i := 0; i < n/2; i++ {
-		parts[i], parts[n-1-i] = parts[n-1-i], parts[i]
-	}
-	dm.max++
-	d.Id = dm.max
-	dm.log("add", "/"+strings.Join(parts, "/"), strconv.Itoa(int(d.Id)))
-	return d, nil
-}
-
-func (dm *DirectoryManagerInMap) log(words ...string) {
-	if !dm.isLoading {
-		dm.logFile.WriteString(strings.Join(words, "\t") + "\n")
-	}
-}
-
-func NewDirectoryManagerInMap(dirLogFile string) (dm *DirectoryManagerInMap, err error) {
-	dm = &DirectoryManagerInMap{}
-	//dm.Root do not use newDirectoryEntryInMap, since dm.max will be changed
-	dm.Root = &DirectoryEntryInMap{subDirectories: make(map[string]*DirectoryEntryInMap)}
-	if dm.logFile, err = os.OpenFile(dirLogFile, os.O_RDWR|os.O_CREATE, 0644); err != nil {
-		return nil, fmt.Errorf("cannot write directory log file %s: %v", dirLogFile, err)
-	}
-	return dm, dm.load()
-}
-
-func (dm *DirectoryManagerInMap) processEachLine(line string) error {
-	if strings.HasPrefix(line, "#") {
-		return nil
-	}
-	if line == "" {
-		return nil
-	}
-	parts := strings.Split(line, "\t")
-	if len(parts) == 0 {
-		return nil
-	}
-	switch parts[0] {
-	case "add":
-		v, pe := strconv.Atoi(parts[2])
-		if pe != nil {
-			return pe
-		}
-		if e := dm.loadDirectory(parts[1], DirectoryId(v)); e != nil {
-			return e
-		}
-	case "mov":
-		newName := ""
-		if len(parts) >= 4 {
-			newName = parts[3]
-		}
-		if e := dm.MoveUnderDirectory(parts[1], parts[2], newName); e != nil {
-			return e
-		}
-	case "del":
-		if e := dm.DeleteDirectory(parts[1]); e != nil {
-			return e
-		}
-	default:
-		fmt.Printf("line %s has %s!\n", line, parts[0])
-		return nil
-	}
-	return nil
-}
-func (dm *DirectoryManagerInMap) load() error {
-	dm.max = 0
-	lines := bufio.NewReader(dm.logFile)
-	dm.isLoading = true
-	defer func() { dm.isLoading = false }()
-	for {
-		line, err := util.Readln(lines)
-		if err != nil && err != io.EOF {
-			return err
-		}
-		if pe := dm.processEachLine(string(line)); pe != nil {
-			return pe
-		}
-		if err == io.EOF {
-			return nil
-		}
-	}
-}
-
-func (dm *DirectoryManagerInMap) findDirectory(dirPath string) (*DirectoryEntryInMap, error) {
-	if dirPath == "" {
-		return dm.Root, nil
-	}
-	dirPath = CleanFilePath(dirPath)
-	if dirPath == "/" {
-		return dm.Root, nil
-	}
-	parts := strings.Split(dirPath, "/")
-	dir := dm.Root
-	for i := 1; i < len(parts); i++ {
-		if sub, ok := dir.getChild(parts[i]); ok {
-			dir = sub
-		} else {
-			return dm.Root, filer.ErrNotFound
-		}
-	}
-	return dir, nil
-}
-func (dm *DirectoryManagerInMap) findDirectoryId(dirPath string) (DirectoryId, error) {
-	d, e := dm.findDirectory(dirPath)
-	if e == nil {
-		return d.Id, nil
-	}
-	return dm.Root.Id, e
-}
-
-func (dm *DirectoryManagerInMap) loadDirectory(dirPath string, dirId DirectoryId) error {
-	dirPath = CleanFilePath(dirPath)
-	if dirPath == "/" {
-		return nil
-	}
-	parts := strings.Split(dirPath, "/")
-	dir := dm.Root
-	for i := 1; i < len(parts); i++ {
-		sub, ok := dir.getChild(parts[i])
-		if !ok {
-			writeLock.Lock()
-			if sub2, createdByOtherThread := dir.getChild(parts[i]); createdByOtherThread {
-				sub = sub2
-			} else {
-				if i != len(parts)-1 {
-					writeLock.Unlock()
-					return fmt.Errorf("%s should be created after parent %s", dirPath, parts[i])
-				}
-				var err error
-				sub, err = dm.newDirectoryEntryInMap(dir, parts[i])
-				if err != nil {
-					writeLock.Unlock()
-					return err
-				}
-				if sub.Id != dirId {
-					writeLock.Unlock()
-					// the dir.log should be the same order as in-memory directory id
-					return fmt.Errorf("%s should be have id %v instead of %v", dirPath, sub.Id, dirId)
-				}
-				dir.addChild(parts[i], sub)
-			}
-			writeLock.Unlock()
-		}
-		dir = sub
-	}
-	return nil
-}
-
-func (dm *DirectoryManagerInMap) makeDirectory(dirPath string) (dir *DirectoryEntryInMap, created bool) {
-	dirPath = CleanFilePath(dirPath)
-	if dirPath == "/" {
-		return dm.Root, false
-	}
-	parts := strings.Split(dirPath, "/")
-	dir = dm.Root
-	for i := 1; i < len(parts); i++ {
-		sub, ok := dir.getChild(parts[i])
-		if !ok {
-			writeLock.Lock()
-			if sub2, createdByOtherThread := dir.getChild(parts[i]); createdByOtherThread {
-				sub = sub2
-			} else {
-				var err error
-				sub, err = dm.newDirectoryEntryInMap(dir, parts[i])
-				if err != nil {
-					writeLock.Unlock()
-					return nil, false
-				}
-				dir.addChild(parts[i], sub)
-				created = true
-			}
-			writeLock.Unlock()
-		}
-		dir = sub
-	}
-	return dir, created
-}
-
-func (dm *DirectoryManagerInMap) MakeDirectory(dirPath string) (DirectoryId, error) {
-	dir, _ := dm.makeDirectory(dirPath)
-	return dir.Id, nil
-}
-
-func (dm *DirectoryManagerInMap) MoveUnderDirectory(oldDirPath string, newParentDirPath string, newName string) error {
-	writeLock.Lock()
-	defer writeLock.Unlock()
-	oldDir, oe := dm.findDirectory(oldDirPath)
-	if oe != nil {
-		return oe
-	}
-	parentDir, pe := dm.findDirectory(newParentDirPath)
-	if pe != nil {
-		return pe
-	}
-	dm.log("mov", oldDirPath, newParentDirPath, newName)
-	oldDir.Parent.removeChild(oldDir.Name)
-	if newName == "" {
-		newName = oldDir.Name
-	}
-	parentDir.addChild(newName, oldDir)
-	oldDir.Name = newName
-	oldDir.Parent = parentDir
-	return nil
-}
-
-func (dm *DirectoryManagerInMap) ListDirectories(dirPath string) (dirNames []filer.DirectoryName, err error) {
-	d, e := dm.findDirectory(dirPath)
-	if e != nil {
-		return dirNames, e
-	}
-	return d.children(), nil
-}
-func (dm *DirectoryManagerInMap) DeleteDirectory(dirPath string) error {
-	writeLock.Lock()
-	defer writeLock.Unlock()
-	if dirPath == "/" {
-		return fmt.Errorf("Can not delete %s", dirPath)
-	}
-	d, e := dm.findDirectory(dirPath)
-	if e != nil {
-		return e
-	}
-	if d.hasChildren() {
-		return fmt.Errorf("dir %s still has sub directories", dirPath)
-	}
-	d.Parent.removeChild(d.Name)
-	d.Parent = nil
-	dm.log("del", dirPath)
-	return nil
-}
-
-func CleanFilePath(fp string) string {
-	ret := filepath.Clean(fp)
-	if os.PathSeparator == '\\' {
-		return strings.Replace(ret, "\\", "/", -1)
-	}
-	return ret
-}

weed/filer/embedded_filer/directory_test.go

@@ -1,86 +0,0 @@
-package embedded_filer
-
-import (
-	"os"
-	"strings"
-	"testing"
-)
-
-func TestDirectory(t *testing.T) {
-	dm, _ := NewDirectoryManagerInMap("/tmp/dir.log")
-	defer func() {
-		if true {
-			os.Remove("/tmp/dir.log")
-		}
-	}()
-	dm.MakeDirectory("/a/b/c")
-	dm.MakeDirectory("/a/b/d")
-	dm.MakeDirectory("/a/b/e")
-	dm.MakeDirectory("/a/b/e/f")
-	dm.MakeDirectory("/a/b/e/f/g")
-	dm.MoveUnderDirectory("/a/b/e/f/g", "/a/b", "t")
-	if _, err := dm.findDirectoryId("/a/b/e/f/g"); err == nil {
-		t.Fatal("/a/b/e/f/g should not exist any more after moving")
-	}
-	if _, err := dm.findDirectoryId("/a/b/t"); err != nil {
-		t.Fatal("/a/b/t should exist after moving")
-	}
-	if _, err := dm.findDirectoryId("/a/b/g"); err == nil {
-		t.Fatal("/a/b/g should not exist after moving")
-	}
-	dm.MoveUnderDirectory("/a/b/e/f", "/a/b", "")
-	if _, err := dm.findDirectoryId("/a/b/f"); err != nil {
-		t.Fatal("/a/b/g should not exist after moving")
-	}
-	dm.MakeDirectory("/a/b/g/h/i")
-	dm.DeleteDirectory("/a/b/e/f")
-	dm.DeleteDirectory("/a/b/e")
-	dirNames, _ := dm.ListDirectories("/a/b/e")
-	for _, v := range dirNames {
-		println("sub1 dir:", v)
-	}
-	dm.logFile.Close()
-
-	var path []string
-	printTree(dm.Root, path)
-
-	dm2, e := NewDirectoryManagerInMap("/tmp/dir.log")
-	if e != nil {
-		println("load error", e.Error())
-	}
-	if !compare(dm.Root, dm2.Root) {
-		t.Fatal("restored dir not the same!")
-	}
-	printTree(dm2.Root, path)
-}
-
-func printTree(node *DirectoryEntryInMap, path []string) {
-	println(strings.Join(path, "/") + "/" + node.Name)
-	path = append(path, node.Name)
-	for _, v := range node.subDirectories {
-		printTree(v, path)
-	}
-}
-
-func compare(root1 *DirectoryEntryInMap, root2 *DirectoryEntryInMap) bool {
-	if len(root1.subDirectories) != len(root2.subDirectories) {
-		return false
-	}
-	if root1.Name != root2.Name {
-		return false
-	}
-	if root1.Id != root2.Id {
-		return false
-	}
-	if !(root1.Parent == nil && root2.Parent == nil) {
-		if root1.Parent.Id != root2.Parent.Id {
-			return false
-		}
-	}
-	for k, v := range root1.subDirectories {
-		if !compare(v, root2.subDirectories[k]) {
-			return false
-		}
-	}
-	return true
-}

weed/filer/embedded_filer/filer_embedded.go

@@ -1,156 +0,0 @@
-package embedded_filer
-
-import (
-	"errors"
-	"fmt"
-	"path/filepath"
-	"strings"
-	"sync"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/operation"
-)
-
-type FilerEmbedded struct {
-	master      string
-	directories *DirectoryManagerInMap
-	files       *FileListInLevelDb
-	mvMutex     sync.Mutex
-}
-
-func NewFilerEmbedded(master string, dir string) (filer *FilerEmbedded, err error) {
-	dm, de := NewDirectoryManagerInMap(filepath.Join(dir, "dir.log"))
-	if de != nil {
-		return nil, de
-	}
-	fl, fe := NewFileListInLevelDb(dir)
-	if fe != nil {
-		return nil, fe
-	}
-	filer = &FilerEmbedded{
-		master:      master,
-		directories: dm,
-		files:       fl,
-	}
-	return
-}
-
-func (filer *FilerEmbedded) CreateFile(filePath string, fid string) (err error) {
-	dir, file := filepath.Split(filePath)
-	dirId, e := filer.directories.MakeDirectory(dir)
-	if e != nil {
-		return e
-	}
-	return filer.files.CreateFile(dirId, file, fid)
-}
-func (filer *FilerEmbedded) FindFile(filePath string) (fid string, err error) {
-	dir, file := filepath.Split(filePath)
-	return filer.findFileEntry(dir, file)
-}
-func (filer *FilerEmbedded) findFileEntry(parentPath string, fileName string) (fid string, err error) {
-	dirId, e := filer.directories.findDirectoryId(parentPath)
-	if e != nil {
-		return "", e
-	}
-	return filer.files.FindFile(dirId, fileName)
-}
-
-func (filer *FilerEmbedded) LookupDirectoryEntry(dirPath string, name string) (found bool, fileId string, err error) {
-	if _, err = filer.directories.findDirectory(filepath.Join(dirPath, name)); err == nil {
-		return true, "", nil
-	}
-	if fileId, err = filer.findFileEntry(dirPath, name); err == nil {
-		return true, fileId, nil
-	}
-	return false, "", err
-}
-func (filer *FilerEmbedded) ListDirectories(dirPath string) (dirs []filer.DirectoryName, err error) {
-	return filer.directories.ListDirectories(dirPath)
-}
-func (filer *FilerEmbedded) ListFiles(dirPath string, lastFileName string, limit int) (files []filer.FileEntry, err error) {
-	dirId, e := filer.directories.findDirectoryId(dirPath)
-	if e != nil {
-		return nil, e
-	}
-	return filer.files.ListFiles(dirId, lastFileName, limit), nil
-}
-func (filer *FilerEmbedded) DeleteDirectory(dirPath string, recursive bool) (err error) {
-	dirId, e := filer.directories.findDirectoryId(dirPath)
-	if e != nil {
-		return e
-	}
-	if sub_dirs, sub_err := filer.directories.ListDirectories(dirPath); sub_err == nil {
-		if len(sub_dirs) > 0 && !recursive {
-			return fmt.Errorf("Fail to delete directory %s: %d sub directories found!", dirPath, len(sub_dirs))
-		}
-		for _, sub := range sub_dirs {
-			if delete_sub_err := filer.DeleteDirectory(filepath.Join(dirPath, string(sub)), recursive); delete_sub_err != nil {
-				return delete_sub_err
-			}
-		}
-	}
-	list := filer.files.ListFiles(dirId, "", 100)
-	if len(list) != 0 && !recursive {
-		if !recursive {
-			return fmt.Errorf("Fail to delete non-empty directory %s!", dirPath)
-		}
-	}
-	for {
-		if len(list) == 0 {
-			return filer.directories.DeleteDirectory(dirPath)
-		}
-		var fids []string
-		for _, fileEntry := range list {
-			fids = append(fids, string(fileEntry.Id))
-		}
-		if result_list, delete_file_err := operation.DeleteFiles(filer.master, fids); delete_file_err != nil {
-			return delete_file_err
-		} else {
-			if len(result_list.Errors) > 0 {
-				return errors.New(strings.Join(result_list.Errors, "\n"))
-			}
-		}
-		lastFile := list[len(list)-1]
-		list = filer.files.ListFiles(dirId, lastFile.Name, 100)
-	}
-
-}
-
-func (filer *FilerEmbedded) DeleteFile(filePath string) (fid string, err error) {
-	dir, file := filepath.Split(filePath)
-	dirId, e := filer.directories.findDirectoryId(dir)
-	if e != nil {
-		return "", e
-	}
-	return filer.files.DeleteFile(dirId, file)
-}
-
-/*
-Move a folder or a file, with 4 Use cases:
-mv fromDir toNewDir
-mv fromDir toOldDir
-mv fromFile toDir
-mv fromFile toFile
-*/
-func (filer *FilerEmbedded) Move(fromPath string, toPath string) error {
-	filer.mvMutex.Lock()
-	defer filer.mvMutex.Unlock()
-
-	if _, dir_err := filer.directories.findDirectoryId(fromPath); dir_err == nil {
-		if _, err := filer.directories.findDirectoryId(toPath); err == nil {
-			// move folder under an existing folder
-			return filer.directories.MoveUnderDirectory(fromPath, toPath, "")
-		}
-		// move folder to a new folder
-		return filer.directories.MoveUnderDirectory(fromPath, filepath.Dir(toPath), filepath.Base(toPath))
-	}
-	if fid, file_err := filer.DeleteFile(fromPath); file_err == nil {
-		if _, err := filer.directories.findDirectoryId(toPath); err == nil {
-			// move file under an existing folder
-			return filer.CreateFile(filepath.Join(toPath, filepath.Base(fromPath)), fid)
-		}
-		// move to a folder with new name
-		return filer.CreateFile(toPath, fid)
-	}
-	return fmt.Errorf("File %s is not found!", fromPath)
-}

weed/filer/embedded_filer/files_in_leveldb.go

@@ -1,87 +0,0 @@
-package embedded_filer
-
-import (
-	"bytes"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/glog"
-	"github.com/syndtr/goleveldb/leveldb"
-	"github.com/syndtr/goleveldb/leveldb/util"
-)
-
-/*
-The entry in level db has this format:
-  key: genKey(dirId, fileName)
-  value: []byte(fid)
-And genKey(dirId, fileName) use first 4 bytes to store dirId, and rest for fileName
-*/
-
-type FileListInLevelDb struct {
-	db *leveldb.DB
-}
-
-func NewFileListInLevelDb(dir string) (fl *FileListInLevelDb, err error) {
-	fl = &FileListInLevelDb{}
-	if fl.db, err = leveldb.OpenFile(dir, nil); err != nil {
-		return
-	}
-	return
-}
-
-func genKey(dirId DirectoryId, fileName string) []byte {
-	ret := make([]byte, 0, 4+len(fileName))
-	for i := 3; i >= 0; i-- {
-		ret = append(ret, byte(dirId>>(uint(i)*8)))
-	}
-	ret = append(ret, []byte(fileName)...)
-	return ret
-}
-
-func (fl *FileListInLevelDb) CreateFile(dirId DirectoryId, fileName string, fid string) (err error) {
-	glog.V(4).Infoln("directory", dirId, "fileName", fileName, "fid", fid)
-	return fl.db.Put(genKey(dirId, fileName), []byte(fid), nil)
-}
-func (fl *FileListInLevelDb) DeleteFile(dirId DirectoryId, fileName string) (fid string, err error) {
-	if fid, err = fl.FindFile(dirId, fileName); err != nil {
-		if err == leveldb.ErrNotFound {
-			return "", nil
-		}
-		return
-	}
-	err = fl.db.Delete(genKey(dirId, fileName), nil)
-	return fid, err
-}
-func (fl *FileListInLevelDb) FindFile(dirId DirectoryId, fileName string) (fid string, err error) {
-	data, e := fl.db.Get(genKey(dirId, fileName), nil)
-	if e == leveldb.ErrNotFound {
-		return "", filer.ErrNotFound
-	} else if e != nil {
-		return "", e
-	}
-	return string(data), nil
-}
-func (fl *FileListInLevelDb) ListFiles(dirId DirectoryId, lastFileName string, limit int) (files []filer.FileEntry) {
-	glog.V(4).Infoln("directory", dirId, "lastFileName", lastFileName, "limit", limit)
-	dirKey := genKey(dirId, "")
-	iter := fl.db.NewIterator(&util.Range{Start: genKey(dirId, lastFileName)}, nil)
-	limitCounter := 0
-	for iter.Next() {
-		key := iter.Key()
-		if !bytes.HasPrefix(key, dirKey) {
-			break
-		}
-		fileName := string(key[len(dirKey):])
-		if fileName == lastFileName {
-			continue
-		}
-		limitCounter++
-		if limit > 0 {
-			if limitCounter > limit {
-				break
-			}
-		}
-		files = append(files, filer.FileEntry{Name: fileName, Id: filer.FileId(string(iter.Value()))})
-	}
-	iter.Release()
-	return
-}

weed/filer/filer.go

@@ -1,29 +0,0 @@
-package filer
-
-import (
-	"errors"
-)
-
-type FileId string //file id in SeaweedFS
-
-type FileEntry struct {
-	Name string `json:"name,omitempty"` //file name without path
-	Id   FileId `json:"fid,omitempty"`
-}
-
-type DirectoryName string
-
-type Filer interface {
-	CreateFile(fullFileName string, fid string) (err error)
-	FindFile(fullFileName string) (fid string, err error)
-	DeleteFile(fullFileName string) (fid string, err error)
-
-	//Optional functions. embedded filer support these
-	ListDirectories(dirPath string) (dirs []DirectoryName, err error)
-	ListFiles(dirPath string, lastFileName string, limit int) (files []FileEntry, err error)
-	DeleteDirectory(dirPath string, recursive bool) (err error)
-	Move(fromPath string, toPath string) (err error)
-	LookupDirectoryEntry(dirPath string, name string) (found bool, fileId string, err error)
-}
-
-var ErrNotFound = errors.New("filer: no entry is found in filer store")

weed/filer/flat_namespace/flat_namespace_filer.go

@@ -1,66 +0,0 @@
-package flat_namespace
-
-import (
-	"errors"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"path/filepath"
-)
-
-type FlatNamespaceFiler struct {
-	master string
-	store  FlatNamespaceStore
-}
-
-var (
-	ErrNotImplemented = errors.New("Not Implemented for flat namespace meta data store")
-)
-
-func NewFlatNamespaceFiler(master string, store FlatNamespaceStore) *FlatNamespaceFiler {
-	return &FlatNamespaceFiler{
-		master: master,
-		store:  store,
-	}
-}
-
-func (filer *FlatNamespaceFiler) CreateFile(fullFileName string, fid string) (err error) {
-	return filer.store.Put(fullFileName, fid)
-}
-func (filer *FlatNamespaceFiler) FindFile(fullFileName string) (fid string, err error) {
-	return filer.store.Get(fullFileName)
-}
-func (filer *FlatNamespaceFiler) LookupDirectoryEntry(dirPath string, name string) (found bool, fileId string, err error) {
-	if fileId, err = filer.FindFile(filepath.Join(dirPath, name)); err == nil {
-		return true, fileId, nil
-	}
-	return false, "", err
-}
-func (filer *FlatNamespaceFiler) ListDirectories(dirPath string) (dirs []filer.DirectoryName, err error) {
-	return nil, ErrNotImplemented
-}
-func (filer *FlatNamespaceFiler) ListFiles(dirPath string, lastFileName string, limit int) (files []filer.FileEntry, err error) {
-	return nil, ErrNotImplemented
-}
-func (filer *FlatNamespaceFiler) DeleteDirectory(dirPath string, recursive bool) (err error) {
-	return ErrNotImplemented
-}
-
-func (filer *FlatNamespaceFiler) DeleteFile(fullFileName string) (fid string, err error) {
-	fid, err = filer.FindFile(fullFileName)
-	if err != nil {
-		return "", err
-	}
-
-	err = filer.store.Delete(fullFileName)
-	if err != nil {
-		return "", err
-	}
-
-	return fid, nil
-	//return filer.store.Delete(fullFileName)
-	//are you kidding me!!!!
-}
-
-func (filer *FlatNamespaceFiler) Move(fromPath string, toPath string) error {
-	return ErrNotImplemented
-}

weed/filer/flat_namespace/flat_namespace_store.go

@@ -1,9 +0,0 @@
-package flat_namespace
-
-import ()
-
-type FlatNamespaceStore interface {
-	Put(fullFileName string, fid string) (err error)
-	Get(fullFileName string) (fid string, err error)
-	Delete(fullFileName string) (err error)
-}

weed/filer/mysql_store/README.md

@@ -1,67 +0,0 @@
-#MySQL filer mapping store
-
-## Schema format
-
-
-Basically, uriPath and fid are the key elements stored in MySQL. In view of the optimization and user's usage, 
-adding primary key with integer type and involving createTime, updateTime, status fields should be somewhat meaningful.
-Of course, you could customize the schema per your concretely circumstance freely.
-
-<pre><code>
-CREATE TABLE IF NOT EXISTS `filer_mapping` (
-  `id` bigint(20) NOT NULL AUTO_INCREMENT,
-  `uriPath` char(256) NOT NULL DEFAULT "" COMMENT 'http uriPath',
-  `fid` char(36) NOT NULL DEFAULT "" COMMENT 'seaweedfs fid',
-  `createTime` int(10) NOT NULL DEFAULT 0 COMMENT 'createdTime in unix timestamp',
-  `updateTime` int(10) NOT NULL DEFAULT 0 COMMENT 'updatedTime in unix timestamp',
-  `remark` varchar(20) NOT NULL DEFAULT "" COMMENT 'reserverd field',
-  `status` tinyint(2) DEFAULT '1' COMMENT 'resource status',
-  PRIMARY KEY (`id`),
-  UNIQUE KEY `index_uriPath` (`uriPath`)
-) DEFAULT CHARSET=utf8;
-</code></pre>
-
-
-The MySQL 's config params is not added into the weed command option as other stores(redis,cassandra). Instead,
-We created a config file(json format) for them. TOML,YAML or XML also should be OK. But TOML and YAML need import thirdparty package
-while XML is a little bit complex.
-
-The sample config file's  content is below:
-
-<pre><code>
-{
-    "mysql": [
-        {
-            "User": "root",
-            "Password": "root",
-            "HostName": "127.0.0.1",
-            "Port": 3306,
-            "DataBase": "seaweedfs"
-        },
-        {
-            "User": "root",
-            "Password": "root",
-            "HostName": "127.0.0.2",
-            "Port": 3306,
-            "DataBase": "seaweedfs"
-        }
-    ],
-    "IsSharding":true,
-    "ShardCount":1024
-}
-</code></pre>
-
-
-The "mysql" field in above conf file is an array which include all mysql instances you prepared to store sharding data.
-
-1. If one mysql instance is enough, just keep one instance in "mysql" field.
-
-2. If table sharding at a specific mysql instance is needed , mark "IsSharding" field with true and specify total table sharding numbers using "ShardCount" field.
-
-3. If the mysql service could be auto scaled transparently in your environment, just config one mysql instance(usually it's a frondend proxy or VIP),and mark "IsSharding" with false value
-
-4. If you prepare more than one mysql instance and have no plan to use table sharding for any instance(mark isSharding with false), instance sharding will still be done implicitly
-
-
-
-

weed/filer/mysql_store/mysql_store.go

@@ -1,274 +0,0 @@
-package mysql_store
-
-import (
-	"database/sql"
-	"fmt"
-	"hash/crc32"
-	"sync"
-	"time"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-
-	_ "github.com/go-sql-driver/mysql"
-)
-
-const (
-	sqlUrl                     = "%s:%s@tcp(%s:%d)/%s?charset=utf8"
-	default_maxIdleConnections = 100
-	default_maxOpenConnections = 50
-	default_maxTableNums       = 1024
-	tableName                  = "filer_mapping"
-)
-
-var (
-	_init_db        sync.Once
-	_db_connections []*sql.DB
-)
-
-type MySqlConf struct {
-	User               string
-	Password           string
-	HostName           string
-	Port               int
-	DataBase           string
-	MaxIdleConnections int
-	MaxOpenConnections int
-}
-
-type ShardingConf struct {
-	IsSharding bool `json:"isSharding"`
-	ShardCount int  `json:"shardCount"`
-}
-
-type MySqlStore struct {
-	dbs        []*sql.DB
-	isSharding bool
-	shardCount int
-}
-
-func getDbConnection(confs []MySqlConf) []*sql.DB {
-	_init_db.Do(func() {
-		for _, conf := range confs {
-
-			sqlUrl := fmt.Sprintf(sqlUrl, conf.User, conf.Password, conf.HostName, conf.Port, conf.DataBase)
-			var dbErr error
-			_db_connection, dbErr := sql.Open("mysql", sqlUrl)
-			if dbErr != nil {
-				_db_connection.Close()
-				_db_connection = nil
-				panic(dbErr)
-			}
-			var maxIdleConnections, maxOpenConnections int
-
-			if conf.MaxIdleConnections != 0 {
-				maxIdleConnections = conf.MaxIdleConnections
-			} else {
-				maxIdleConnections = default_maxIdleConnections
-			}
-			if conf.MaxOpenConnections != 0 {
-				maxOpenConnections = conf.MaxOpenConnections
-			} else {
-				maxOpenConnections = default_maxOpenConnections
-			}
-
-			_db_connection.SetMaxIdleConns(maxIdleConnections)
-			_db_connection.SetMaxOpenConns(maxOpenConnections)
-			_db_connections = append(_db_connections, _db_connection)
-		}
-	})
-	return _db_connections
-}
-
-func NewMysqlStore(confs []MySqlConf, isSharding bool, shardCount int) *MySqlStore {
-	ms := &MySqlStore{
-		dbs:        getDbConnection(confs),
-		isSharding: isSharding,
-		shardCount: shardCount,
-	}
-
-	for _, db := range ms.dbs {
-		if !isSharding {
-			ms.shardCount = 1
-		} else {
-			if ms.shardCount == 0 {
-				ms.shardCount = default_maxTableNums
-			}
-		}
-		for i := 0; i < ms.shardCount; i++ {
-			if err := ms.createTables(db, tableName, i); err != nil {
-				fmt.Printf("create table failed %v", err)
-			}
-		}
-	}
-
-	return ms
-}
-
-func (s *MySqlStore) hash(fullFileName string) (instance_offset, table_postfix int) {
-	hash_value := crc32.ChecksumIEEE([]byte(fullFileName))
-	instance_offset = int(hash_value) % len(s.dbs)
-	table_postfix = int(hash_value) % s.shardCount
-	return
-}
-
-func (s *MySqlStore) parseFilerMappingInfo(path string) (instanceId int, tableFullName string, err error) {
-	instance_offset, table_postfix := s.hash(path)
-	instanceId = instance_offset
-	if s.isSharding {
-		tableFullName = fmt.Sprintf("%s_%04d", tableName, table_postfix)
-	} else {
-		tableFullName = tableName
-	}
-	return
-}
-
-func (s *MySqlStore) Get(fullFilePath string) (fid string, err error) {
-	instance_offset, tableFullName, err := s.parseFilerMappingInfo(fullFilePath)
-	if err != nil {
-		return "", fmt.Errorf("MySqlStore Get operation can not parse file path %s: err is %v", fullFilePath, err)
-	}
-	fid, err = s.query(fullFilePath, s.dbs[instance_offset], tableFullName)
-	if err == sql.ErrNoRows {
-		//Could not found
-		err = filer.ErrNotFound
-	}
-	return fid, err
-}
-
-func (s *MySqlStore) Put(fullFilePath string, fid string) (err error) {
-	var tableFullName string
-
-	instance_offset, tableFullName, err := s.parseFilerMappingInfo(fullFilePath)
-	if err != nil {
-		return fmt.Errorf("MySqlStore Put operation can not parse file path %s: err is %v", fullFilePath, err)
-	}
-	var old_fid string
-	if old_fid, err = s.query(fullFilePath, s.dbs[instance_offset], tableFullName); err != nil && err != sql.ErrNoRows {
-		return fmt.Errorf("MySqlStore Put operation failed when querying path %s: err is %v", fullFilePath, err)
-	} else {
-		if len(old_fid) == 0 {
-			err = s.insert(fullFilePath, fid, s.dbs[instance_offset], tableFullName)
-			if err != nil {
-				err = fmt.Errorf("MySqlStore Put operation failed when inserting path %s with fid %s : err is %v", fullFilePath, fid, err)
-			}
-		} else {
-			err = s.update(fullFilePath, fid, s.dbs[instance_offset], tableFullName)
-			if err != nil {
-				err = fmt.Errorf("MySqlStore Put operation failed when updating path %s with fid %s : err is %v", fullFilePath, fid, err)
-			}
-		}
-	}
-	return
-}
-
-func (s *MySqlStore) Delete(fullFilePath string) (err error) {
-	var fid string
-	instance_offset, tableFullName, err := s.parseFilerMappingInfo(fullFilePath)
-	if err != nil {
-		return fmt.Errorf("MySqlStore Delete operation can not parse file path %s: err is %v", fullFilePath, err)
-	}
-	if fid, err = s.query(fullFilePath, s.dbs[instance_offset], tableFullName); err != nil {
-		return fmt.Errorf("MySqlStore Delete operation failed when querying path %s: err is %v", fullFilePath, err)
-	} else if fid == "" {
-		return nil
-	}
-	if err = s.delete(fullFilePath, s.dbs[instance_offset], tableFullName); err != nil {
-		return fmt.Errorf("MySqlStore Delete operation failed when deleting path %s: err is %v", fullFilePath, err)
-	} else {
-		return nil
-	}
-}
-
-func (s *MySqlStore) Close() {
-	for _, db := range s.dbs {
-		db.Close()
-	}
-}
-
-var createTable = `
-CREATE TABLE IF NOT EXISTS %s (
-  id bigint(20) NOT NULL AUTO_INCREMENT,
-  uriPath char(255) NOT NULL DEFAULT "" COMMENT 'http uriPath',
-  fid char(36) NOT NULL DEFAULT "" COMMENT 'seaweedfs fid',
-  createTime int(10) NOT NULL DEFAULT 0 COMMENT 'createdTime in unix timestamp',
-  updateTime int(10) NOT NULL DEFAULT 0 COMMENT 'updatedTime in unix timestamp',
-  remark varchar(20) NOT NULL DEFAULT "" COMMENT 'reserverd field',
-  status tinyint(2) DEFAULT '1' COMMENT 'resource status',
-  PRIMARY KEY (id),
-  UNIQUE KEY index_uriPath (uriPath)
-) DEFAULT CHARSET=utf8;
-`
-
-func (s *MySqlStore) createTables(db *sql.DB, tableName string, postfix int) error {
-	var realTableName string
-	if s.isSharding {
-		realTableName = fmt.Sprintf("%s_%04d", tableName, postfix)
-	} else {
-		realTableName = tableName
-	}
-
-	stmt, err := db.Prepare(fmt.Sprintf(createTable, realTableName))
-	if err != nil {
-		return err
-	}
-	defer stmt.Close()
-
-	_, err = stmt.Exec()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *MySqlStore) query(uriPath string, db *sql.DB, tableName string) (string, error) {
-	sqlStatement := "SELECT fid FROM %s WHERE uriPath=?"
-	row := db.QueryRow(fmt.Sprintf(sqlStatement, tableName), uriPath)
-	var fid string
-	err := row.Scan(&fid)
-	if err != nil {
-		return "", err
-	}
-	return fid, nil
-}
-
-func (s *MySqlStore) update(uriPath string, fid string, db *sql.DB, tableName string) error {
-	sqlStatement := "UPDATE %s SET fid=?, updateTime=? WHERE uriPath=?"
-	res, err := db.Exec(fmt.Sprintf(sqlStatement, tableName), fid, time.Now().Unix(), uriPath)
-	if err != nil {
-		return err
-	}
-
-	_, err = res.RowsAffected()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *MySqlStore) insert(uriPath string, fid string, db *sql.DB, tableName string) error {
-	sqlStatement := "INSERT INTO %s (uriPath,fid,createTime) VALUES(?,?,?)"
-	res, err := db.Exec(fmt.Sprintf(sqlStatement, tableName), uriPath, fid, time.Now().Unix())
-	if err != nil {
-		return err
-	}
-
-	_, err = res.RowsAffected()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *MySqlStore) delete(uriPath string, db *sql.DB, tableName string) error {
-	sqlStatement := "DELETE FROM %s WHERE uriPath=?"
-	res, err := db.Exec(fmt.Sprintf(sqlStatement, tableName), uriPath)
-	if err != nil {
-		return err
-	}
-
-	_, err = res.RowsAffected()
-	if err != nil {
-		return err
-	}
-	return nil
-}

weed/filer/mysql_store/mysql_store_test.go

@@ -1,30 +0,0 @@
-package mysql_store
-
-import (
-	"encoding/json"
-	"hash/crc32"
-	"testing"
-)
-
-func TestGenerateMysqlConf(t *testing.T) {
-	var conf []MySqlConf
-	conf = append(conf, MySqlConf{
-		User:     "root",
-		Password: "root",
-		HostName: "localhost",
-		Port:     3306,
-		DataBase: "seaweedfs",
-	})
-	body, err := json.Marshal(conf)
-	if err != nil {
-		t.Errorf("json encoding err %s", err.Error())
-	}
-	t.Logf("json output is %s", string(body))
-}
-
-func TestCRC32FullPathName(t *testing.T) {
-	fullPathName := "/prod-bucket/law632191483895612493300-signed.pdf"
-	hash_value := crc32.ChecksumIEEE([]byte(fullPathName))
-	table_postfix := int(hash_value) % 1024
-	t.Logf("table postfix %d", table_postfix)
-}

weed/filer/postgres_store/postgres_native.go

@@ -1,456 +0,0 @@
-package postgres_store
-
-import (
-	"database/sql"
-	"fmt"
-	"path/filepath"
-	"time"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/glog"
-
-	_ "github.com/lib/pq"
-	_ "path/filepath"
-	"strings"
-)
-
-type DirectoryId int32
-
-func databaseExists(db *sql.DB, databaseName string) (bool, error) {
-	sqlStatement := "SELECT datname from pg_database WHERE datname='%s'"
-	row := db.QueryRow(fmt.Sprintf(sqlStatement, databaseName))
-
-	var dbName string
-	err := row.Scan(&dbName)
-	if err != nil {
-		if err == sql.ErrNoRows {
-			return false, nil
-		}
-		return false, err
-	}
-	return true, nil
-}
-
-func createDatabase(db *sql.DB, databaseName string) error {
-	sqlStatement := "CREATE DATABASE %s ENCODING='UTF8'"
-	_, err := db.Exec(fmt.Sprintf(sqlStatement, databaseName))
-	return err
-}
-
-func getDbConnection(conf PostgresConf) *sql.DB {
-	_init_db.Do(func() {
-
-		sqlUrl := fmt.Sprintf("host=%s port=%d user=%s password=%s dbname=%s sslmode=%s connect_timeout=30", conf.HostName, conf.Port, conf.User, conf.Password, "postgres", conf.SslMode)
-		glog.V(3).Infoln("Opening postgres master database")
-
-		var dbErr error
-		_db_connection, dbErr := sql.Open("postgres", sqlUrl)
-		if dbErr != nil {
-			_db_connection.Close()
-			_db_connection = nil
-			panic(dbErr)
-		}
-
-		pingErr := _db_connection.Ping()
-		if pingErr != nil {
-			_db_connection.Close()
-			_db_connection = nil
-			panic(pingErr)
-		}
-
-		glog.V(3).Infoln("Checking to see if DB exists: ", conf.DataBase)
-		var existsErr error
-		dbExists, existsErr := databaseExists(_db_connection, conf.DataBase)
-		if existsErr != nil {
-			_db_connection.Close()
-			_db_connection = nil
-			panic(existsErr)
-		}
-
-		if !dbExists {
-			glog.V(3).Infoln("Database doesn't exist. Attempting to create one:  ", conf.DataBase)
-			createErr := createDatabase(_db_connection, conf.DataBase)
-			if createErr != nil {
-				_db_connection.Close()
-				_db_connection = nil
-				panic(createErr)
-			}
-		}
-
-		glog.V(3).Infoln("Closing master postgres database and opening configured database: ", conf.DataBase)
-		_db_connection.Close()
-		_db_connection = nil
-
-		sqlUrl = fmt.Sprintf("host=%s port=%d user=%s password=%s dbname=%s sslmode=%s connect_timeout=30", conf.HostName, conf.Port, conf.User, conf.Password, conf.DataBase, conf.SslMode)
-		_db_connection, dbErr = sql.Open("postgres", sqlUrl)
-		if dbErr != nil {
-			_db_connection.Close()
-			_db_connection = nil
-			panic(dbErr)
-		}
-
-		pingErr = _db_connection.Ping()
-		if pingErr != nil {
-			_db_connection.Close()
-			_db_connection = nil
-			panic(pingErr)
-		}
-
-		maxIdleConnections, maxOpenConnections := default_maxIdleConnections, default_maxOpenConnections
-		if conf.MaxIdleConnections != 0 {
-			maxIdleConnections = conf.MaxIdleConnections
-		}
-		if conf.MaxOpenConnections != 0 {
-			maxOpenConnections = conf.MaxOpenConnections
-		}
-
-		_db_connection.SetMaxIdleConns(maxIdleConnections)
-		_db_connection.SetMaxOpenConns(maxOpenConnections)
-	})
-	return _db_connection
-}
-
-var createDirectoryTable = `
-
-CREATE TABLE IF NOT EXISTS %s (
-  id BIGSERIAL NOT NULL,
-  directoryRoot VARCHAR(1024) NOT NULL DEFAULT '',
-  directoryName VARCHAR(1024) NOT NULL DEFAULT '',
-  CONSTRAINT unique_directory UNIQUE (directoryRoot, directoryName)
-);
-`
-
-var createFileTable = `
-
-CREATE TABLE IF NOT EXISTS %s (
-  id BIGSERIAL NOT NULL,
-  directoryPart VARCHAR(1024) NOT NULL DEFAULT '',
-  filePart VARCHAR(1024) NOT NULL DEFAULT '',
-  fid VARCHAR(36) NOT NULL DEFAULT '',
-  createTime BIGINT NOT NULL DEFAULT 0,
-  updateTime BIGINT NOT NULL DEFAULT 0,
-  remark VARCHAR(20) NOT NULL DEFAULT '',
-  status SMALLINT NOT NULL DEFAULT '1',
-  PRIMARY KEY (id),
-  CONSTRAINT %s_unique_file UNIQUE (directoryPart, filePart)
-);
-`
-
-func (s *PostgresStore) createDirectoriesTable() error {
-	glog.V(3).Infoln("Creating postgres table if it doesn't exist: ", directoriesTableName)
-
-	sqlCreate := fmt.Sprintf(createDirectoryTable, directoriesTableName)
-
-	stmt, err := s.db.Prepare(sqlCreate)
-	if err != nil {
-		return err
-	}
-	defer stmt.Close()
-
-	_, err = stmt.Exec()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *PostgresStore) createFilesTable() error {
-
-	glog.V(3).Infoln("Creating postgres table if it doesn't exist: ", filesTableName)
-
-	sqlCreate := fmt.Sprintf(createFileTable, filesTableName, filesTableName)
-
-	stmt, err := s.db.Prepare(sqlCreate)
-	if err != nil {
-		return err
-	}
-	defer stmt.Close()
-
-	_, err = stmt.Exec()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *PostgresStore) query(uriPath string) (string, error) {
-	directoryPart, filePart := filepath.Split(uriPath)
-	sqlStatement := fmt.Sprintf("SELECT fid FROM %s WHERE directoryPart=$1 AND filePart=$2", filesTableName)
-
-	row := s.db.QueryRow(sqlStatement, directoryPart, filePart)
-	var fid string
-	err := row.Scan(&fid)
-
-	glog.V(3).Infof("Postgres query -- looking up path '%s' and found id '%s' ", uriPath, fid)
-
-	if err != nil {
-		return "", err
-	}
-	return fid, nil
-}
-
-func (s *PostgresStore) update(uriPath string, fid string) error {
-	directoryPart, filePart := filepath.Split(uriPath)
-	sqlStatement := fmt.Sprintf("UPDATE %s SET fid=$1, updateTime=$2 WHERE directoryPart=$3 AND filePart=$4", filesTableName)
-
-	glog.V(3).Infof("Postgres query -- updating path '%s' with id '%s'", uriPath, fid)
-
-	res, err := s.db.Exec(sqlStatement, fid, time.Now().Unix(), directoryPart, filePart)
-	if err != nil {
-		return err
-	}
-
-	_, err = res.RowsAffected()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *PostgresStore) insert(uriPath string, fid string) error {
-	directoryPart, filePart := filepath.Split(uriPath)
-
-	existingId, _, _ := s.lookupDirectory(directoryPart)
-	if existingId == 0 {
-		s.recursiveInsertDirectory(directoryPart)
-	}
-
-	sqlStatement := fmt.Sprintf("INSERT INTO %s (directoryPart,filePart,fid,createTime) VALUES($1, $2, $3, $4)", filesTableName)
-	glog.V(3).Infof("Postgres query -- inserting path '%s' with id '%s'", uriPath, fid)
-
-	res, err := s.db.Exec(sqlStatement, directoryPart, filePart, fid, time.Now().Unix())
-
-	if err != nil {
-		return err
-	}
-
-	rows, err := res.RowsAffected()
-	if rows != 1 {
-		return fmt.Errorf("Postgres insert -- rows affected = %d. Expecting 1", rows)
-	}
-	if err != nil {
-		return err
-	}
-
-	return nil
-}
-
-func (s *PostgresStore) recursiveInsertDirectory(dirPath string) {
-	pathParts := strings.Split(dirPath, "/")
-
-	var workingPath string = "/"
-	for _, part := range pathParts {
-		if part == "" {
-			continue
-		}
-		workingPath += (part + "/")
-		existingId, _, _ := s.lookupDirectory(workingPath)
-		if existingId == 0 {
-			s.insertDirectory(workingPath)
-		}
-	}
-}
-
-func (s *PostgresStore) insertDirectory(dirPath string) {
-	pathParts := strings.Split(dirPath, "/")
-
-	directoryRoot := "/"
-	directoryName := ""
-	if len(pathParts) > 1 {
-		directoryRoot = strings.Join(pathParts[0:len(pathParts)-2], "/") + "/"
-		directoryName = strings.Join(pathParts[len(pathParts)-2:], "/")
-	} else if len(pathParts) == 1 {
-		directoryRoot = "/"
-		directoryName = pathParts[0] + "/"
-	}
-	sqlInsertDirectoryStatement := fmt.Sprintf("INSERT INTO %s (directoryroot, directoryname)  "+
-		"SELECT $1, $2 WHERE NOT EXISTS ( SELECT id FROM %s WHERE directoryroot=$3 AND directoryname=$4 )",
-		directoriesTableName, directoriesTableName)
-
-	glog.V(4).Infof("Postgres query -- Inserting directory (if it doesn't exist) - root = %s, name = %s",
-		directoryRoot, directoryName)
-
-	_, err := s.db.Exec(sqlInsertDirectoryStatement, directoryRoot, directoryName, directoryRoot, directoryName)
-	if err != nil {
-		glog.V(0).Infof("Postgres query -- Error inserting directory - root = %s, name = %s: %s",
-			directoryRoot, directoryName, err)
-	}
-}
-
-func (s *PostgresStore) delete(uriPath string) error {
-	directoryPart, filePart := filepath.Split(uriPath)
-	sqlStatement := fmt.Sprintf("DELETE FROM %s WHERE directoryPart=$1 AND filePart=$2", filesTableName)
-
-	glog.V(3).Infof("Postgres query -- deleting path '%s'", uriPath)
-
-	res, err := s.db.Exec(sqlStatement, directoryPart, filePart)
-	if err != nil {
-		return err
-	}
-
-	_, err = res.RowsAffected()
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (s *PostgresStore) lookupDirectory(dirPath string) (DirectoryId, string, error) {
-	directoryRoot, directoryName := s.mySplitPath(dirPath)
-
-	sqlStatement := fmt.Sprintf("SELECT id, directoryroot, directoryname FROM %s WHERE directoryRoot=$1 AND directoryName=$2", directoriesTableName)
-
-	row := s.db.QueryRow(sqlStatement, directoryRoot, directoryName)
-	var id DirectoryId
-	var dirRoot string
-	var dirName string
-	err := row.Scan(&id, &dirRoot, &dirName)
-
-	glog.V(3).Infof("Postgres lookupDirectory -- looking up directory '%s' and found id '%d', root '%s', name '%s' ", dirPath, id, dirRoot, dirName)
-
-	if err != nil {
-		return 0, "", err
-	}
-	return id, filepath.Join(dirRoot, dirName), err
-}
-
-func (s *PostgresStore) findDirectories(dirPath string, limit int) (dirs []filer.DirectoryName, err error) {
-	sqlStatement := fmt.Sprintf("SELECT id, directoryroot, directoryname FROM %s WHERE directoryRoot=$1 AND directoryName != '' ORDER BY id LIMIT $2", directoriesTableName)
-	rows, err := s.db.Query(sqlStatement, dirPath, limit)
-
-	if err != nil {
-		glog.V(0).Infof("Postgres findDirectories error: %s", err)
-	}
-
-	if rows != nil {
-		defer rows.Close()
-		for rows.Next() {
-			var id DirectoryId
-			var directoryRoot string
-			var directoryName string
-
-			scanErr := rows.Scan(&id, &directoryRoot, &directoryName)
-			if scanErr != nil {
-				err = scanErr
-			}
-			dirs = append(dirs, filer.DirectoryName(directoryName))
-		}
-	}
-	return
-}
-
-func (s *PostgresStore) safeToDeleteDirectory(dirPath string, recursive bool) bool {
-	if recursive {
-		return true
-	}
-	sqlStatement := fmt.Sprintf("SELECT id FROM %s WHERE directoryRoot LIKE $1 LIMIT 1", directoriesTableName)
-	row := s.db.QueryRow(sqlStatement, dirPath+"%")
-
-	var id DirectoryId
-	err := row.Scan(&id)
-	if err != nil {
-		if err == sql.ErrNoRows {
-			return true
-		}
-	}
-	return false
-}
-
-func (s *PostgresStore) mySplitPath(dirPath string) (directoryRoot string, directoryName string) {
-	pathParts := strings.Split(dirPath, "/")
-	directoryRoot = "/"
-	directoryName = ""
-	if len(pathParts) > 1 {
-		directoryRoot = strings.Join(pathParts[0:len(pathParts)-2], "/") + "/"
-		directoryName = strings.Join(pathParts[len(pathParts)-2:], "/")
-	} else if len(pathParts) == 1 {
-		directoryRoot = "/"
-		directoryName = pathParts[0] + "/"
-	}
-	return directoryRoot, directoryName
-}
-
-func (s *PostgresStore) deleteDirectory(dirPath string, recursive bool) (err error) {
-	directoryRoot, directoryName := s.mySplitPath(dirPath)
-
-	// delete files
-	sqlStatement := fmt.Sprintf("DELETE FROM %s WHERE directorypart=$1", filesTableName)
-	_, err = s.db.Exec(sqlStatement, dirPath)
-	if err != nil {
-		return err
-	}
-
-	// delete specific directory if it is empty or recursive delete was requested
-	safeToDelete := s.safeToDeleteDirectory(dirPath, recursive)
-	if safeToDelete {
-		sqlStatement = fmt.Sprintf("DELETE FROM %s WHERE directoryRoot=$1 AND directoryName=$2", directoriesTableName)
-		_, err = s.db.Exec(sqlStatement, directoryRoot, directoryName)
-		if err != nil {
-			return err
-		}
-	}
-
-	if recursive {
-		// delete descendant files
-		sqlStatement = fmt.Sprintf("DELETE FROM %s WHERE directorypart LIKE $1", filesTableName)
-		_, err = s.db.Exec(sqlStatement, dirPath+"%")
-		if err != nil {
-			return err
-		}
-
-		// delete descendant directories
-		sqlStatement = fmt.Sprintf("DELETE FROM %s WHERE directoryRoot LIKE $1", directoriesTableName)
-		_, err = s.db.Exec(sqlStatement, dirPath+"%")
-		if err != nil {
-			return err
-		}
-	}
-
-	return err
-}
-
-func (s *PostgresStore) findFiles(dirPath string, lastFileName string, limit int) (files []filer.FileEntry, err error) {
-	var rows *sql.Rows = nil
-
-	if lastFileName == "" {
-		sqlStatement :=
-			fmt.Sprintf("SELECT fid, directorypart, filepart FROM %s WHERE directorypart=$1 ORDER BY id LIMIT $2", filesTableName)
-		rows, err = s.db.Query(sqlStatement, dirPath, limit)
-	} else {
-		sqlStatement :=
-			fmt.Sprintf("SELECT fid, directorypart, filepart FROM %s WHERE directorypart=$1 "+
-				"AND id > (SELECT id FROM %s WHERE directoryPart=$2 AND filepart=$3)  ORDER BY id LIMIT $4",
-				filesTableName, filesTableName)
-		_, lastFileNameName := filepath.Split(lastFileName)
-		rows, err = s.db.Query(sqlStatement, dirPath, dirPath, lastFileNameName, limit)
-	}
-
-	if err != nil {
-		glog.V(0).Infof("Postgres find files error: %s", err)
-	}
-
-	if rows != nil {
-		defer rows.Close()
-
-		for rows.Next() {
-			var fid filer.FileId
-			var directoryPart string
-			var filePart string
-
-			scanErr := rows.Scan(&fid, &directoryPart, &filePart)
-			if scanErr != nil {
-				err = scanErr
-			}
-
-			files = append(files, filer.FileEntry{Name: filepath.Join(directoryPart, filePart), Id: fid})
-			if len(files) >= limit {
-				break
-			}
-		}
-	}
-
-	glog.V(3).Infof("Postgres findFiles -- looking up files under '%s' and found %d files. Limit=%d, lastFileName=%s",
-		dirPath, len(files), limit, lastFileName)
-
-	return files, err
-}

weed/filer/postgres_store/postgres_store.go

@@ -1,149 +0,0 @@
-package postgres_store
-
-import (
-	"database/sql"
-	"errors"
-	"fmt"
-	"sync"
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/glog"
-
-	_ "github.com/lib/pq"
-	_ "path/filepath"
-	"path/filepath"
-)
-
-const (
-	default_maxIdleConnections = 100
-	default_maxOpenConnections = 50
-	filesTableName             = "files"
-	directoriesTableName       = "directories"
-)
-
-var (
-	_init_db       sync.Once
-	_db_connection *sql.DB
-)
-
-type PostgresConf struct {
-	User               string
-	Password           string
-	HostName           string
-	Port               int
-	DataBase           string
-	SslMode            string
-	MaxIdleConnections int
-	MaxOpenConnections int
-}
-
-type PostgresStore struct {
-	db       *sql.DB
-	server   string
-	user     string
-	password string
-}
-
-func (s *PostgresStore) CreateFile(fullFilePath string, fid string) (err error) {
-
-	var old_fid string
-	if old_fid, err = s.query(fullFilePath); err != nil && err != sql.ErrNoRows {
-		return fmt.Errorf("PostgresStore Put operation failed when querying path %s: err is %v", fullFilePath, err)
-	} else {
-		if len(old_fid) == 0 {
-			err = s.insert(fullFilePath, fid)
-			if err != nil {
-				return fmt.Errorf("PostgresStore Put operation failed when inserting path %s with fid %s : err is %v", fullFilePath, fid, err)
-			}
-		} else {
-			err = s.update(fullFilePath, fid)
-			if err != nil {
-				return fmt.Errorf("PostgresStore Put operation failed when updating path %s with fid %s : err is %v", fullFilePath, fid, err)
-			}
-		}
-	}
-	return
-
-}
-
-func (s *PostgresStore) FindFile(fullFilePath string) (fid string, err error) {
-
-	if err != nil {
-		return "", fmt.Errorf("PostgresStore Get operation can not parse file path %s: err is %v", fullFilePath, err)
-	}
-	fid, err = s.query(fullFilePath)
-
-	return fid, err
-}
-
-func (s *PostgresStore) LookupDirectoryEntry(dirPath string, name string) (found bool, fileId string, err error) {
-	fullPath := filepath.Join(dirPath, name)
-	if fileId, err = s.FindFile(fullPath); err == nil {
-		return true, fileId, nil
-	}
-	if _, _, err := s.lookupDirectory(fullPath); err == nil {
-		return true, "", err
-	}
-	return false, "", err
-}
-
-func (s *PostgresStore) DeleteFile(fullFilePath string) (fid string, err error) {
-	if err != nil {
-		return "", fmt.Errorf("PostgresStore Delete operation can not parse file path %s: err is %v", fullFilePath, err)
-	}
-	if fid, err = s.query(fullFilePath); err != nil {
-		return "", fmt.Errorf("PostgresStore Delete operation failed when querying path %s: err is %v", fullFilePath, err)
-	} else if fid == "" {
-		return "", nil
-	}
-	if err = s.delete(fullFilePath); err != nil {
-		return "", fmt.Errorf("PostgresStore Delete operation failed when deleting path %s: err is %v", fullFilePath, err)
-	} else {
-		return "", nil
-	}
-}
-
-func (s *PostgresStore) ListDirectories(dirPath string) (dirs []filer.DirectoryName, err error) {
-
-	dirs, err = s.findDirectories(dirPath, 1000)
-
-	glog.V(3).Infof("Postgres ListDirs = found %d directories under %s", len(dirs), dirPath)
-
-	return dirs, err
-}
-
-func (s *PostgresStore) ListFiles(dirPath string, lastFileName string, limit int) (files []filer.FileEntry, err error) {
-	files, err = s.findFiles(dirPath, lastFileName, limit)
-	return files, err
-}
-
-func (s *PostgresStore) DeleteDirectory(dirPath string, recursive bool) (err error) {
-	err = s.deleteDirectory(dirPath, recursive)
-	if err != nil {
-		glog.V(0).Infof("Error in Postgres DeleteDir '%s' (recursive = '%t'): %s", err)
-	}
-	return err
-}
-
-func (s *PostgresStore) Move(fromPath string, toPath string) (err error) {
-	glog.V(3).Infoln("Calling posgres_store Move")
-	return errors.New("Move is not yet implemented for the PostgreSQL store.")
-}
-
-//func NewPostgresStore(master string, confs []PostgresConf, isSharding bool, shardCount int) *PostgresStore {
-func NewPostgresStore(master string, conf PostgresConf) *PostgresStore {
-	pg := &PostgresStore{
-		db: getDbConnection(conf),
-	}
-
-	pg.createDirectoriesTable()
-
-	if err := pg.createFilesTable(); err != nil {
-		fmt.Printf("create table failed %v", err)
-	}
-
-	return pg
-}
-
-func (s *PostgresStore) Close() {
-	s.db.Close()
-}

weed/filer/redis_store/redis_store.go

@@ -1,50 +0,0 @@
-package redis_store
-
-import (
-	"github.com/chrislusf/seaweedfs/weed/filer"
-
-	"github.com/go-redis/redis"
-)
-
-type RedisStore struct {
-	Client *redis.Client
-}
-
-func NewRedisStore(hostPort string, password string, database int) *RedisStore {
-	client := redis.NewClient(&redis.Options{
-		Addr:     hostPort,
-		Password: password,
-		DB:       database,
-	})
-	return &RedisStore{Client: client}
-}
-
-func (s *RedisStore) Get(fullFileName string) (fid string, err error) {
-	fid, err = s.Client.Get(fullFileName).Result()
-	if err == redis.Nil {
-		err = filer.ErrNotFound
-	}
-	return fid, err
-}
-func (s *RedisStore) Put(fullFileName string, fid string) (err error) {
-	_, err = s.Client.Set(fullFileName, fid, 0).Result()
-	if err == redis.Nil {
-		err = nil
-	}
-	return err
-}
-
-// Currently the fid is not returned
-func (s *RedisStore) Delete(fullFileName string) (err error) {
-	_, err = s.Client.Del(fullFileName).Result()
-	if err == redis.Nil {
-		err = nil
-	}
-	return err
-}
-
-func (s *RedisStore) Close() {
-	if s.Client != nil {
-		s.Client.Close()
-	}
-}

weed/filer/vasto_store/design.txt

@@ -1,45 +0,0 @@
-There are two main components of a filer: directories and files.
-
-My previous approach was to use some sequance number to generate directoryId.
-However, this is not scalable. The id generation itself is a bottleneck.
-It needs careful locking and deduplication checking to get a directoryId.
-
-In a second design, each directory is deterministically mapped to UUID version 3,
-which uses MD5 to map a tuple of <uuid, name> to a version 3 UUID.
-However, this UUID3 approach is logically the same as storing the full path.
-
-Storing the full path is the simplest design.
-
-separator is a special byte, 0x00.
-
-When writing a file:
-  <file parent full path, separator, file name> => fildId, file properties
-For folders:
-  The filer breaks the directory path into folders.
-  for each folder:
-    if it is not in cache:
-	  check whether the folder is created in the KVS, if not:
-        set <folder parent full path, separator, folder name> => directory properties
-	if no permission for the folder:
-	  break
-	  
-
-The filer caches the most recently used folder permissions with a TTL.
-  So any folder permission change needs to wait TTL interval to take effect.
-
-
-
-When listing the directory:
-  prefix scan of using (the folder full path + separator) as the prefix
-
-The downside:
-	1. Rename a folder will need to recursively process all sub folders and files.
-	2. Move   a folder will need to recursively process all sub folders and files.
-So these operations are not allowed if the folder is not empty.
-
-Allowing:
-	1. Rename a file
-	2. Move a file to a different folder
-	3. Delete an empty folder
-
-

weed/filer2/abstract_sql/abstract_sql_store.go

@@ -0,0 +1,130 @@
+package abstract_sql
+
+import (
+	"database/sql"
+	"fmt"
+
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+)
+
+type AbstractSqlStore struct {
+	DB               *sql.DB
+	SqlInsert        string
+	SqlUpdate        string
+	SqlFind          string
+	SqlDelete        string
+	SqlListExclusive string
+	SqlListInclusive string
+}
+
+func (store *AbstractSqlStore) InsertEntry(entry *filer2.Entry) (err error) {
+
+	dir, name := entry.FullPath.DirAndName()
+	meta, err := entry.EncodeAttributesAndChunks()
+	if err != nil {
+		return fmt.Errorf("encode %s: %s", entry.FullPath, err)
+	}
+
+	res, err := store.DB.Exec(store.SqlInsert, hashToLong(dir), name, dir, meta)
+	if err != nil {
+		return fmt.Errorf("insert %s: %s", entry.FullPath, err)
+	}
+
+	_, err = res.RowsAffected()
+	if err != nil {
+		return fmt.Errorf("insert %s but no rows affected: %s", entry.FullPath, err)
+	}
+	return nil
+}
+
+func (store *AbstractSqlStore) UpdateEntry(entry *filer2.Entry) (err error) {
+
+	dir, name := entry.FullPath.DirAndName()
+	meta, err := entry.EncodeAttributesAndChunks()
+	if err != nil {
+		return fmt.Errorf("encode %s: %s", entry.FullPath, err)
+	}
+
+	res, err := store.DB.Exec(store.SqlUpdate, meta, hashToLong(dir), name, dir)
+	if err != nil {
+		return fmt.Errorf("update %s: %s", entry.FullPath, err)
+	}
+
+	_, err = res.RowsAffected()
+	if err != nil {
+		return fmt.Errorf("update %s but no rows affected: %s", entry.FullPath, err)
+	}
+	return nil
+}
+
+func (store *AbstractSqlStore) FindEntry(fullpath filer2.FullPath) (*filer2.Entry, error) {
+
+	dir, name := fullpath.DirAndName()
+	row := store.DB.QueryRow(store.SqlFind, hashToLong(dir), name, dir)
+	var data []byte
+	if err := row.Scan(&data); err != nil {
+		return nil, fmt.Errorf("read entry %s: %v", fullpath, err)
+	}
+
+	entry := &filer2.Entry{
+		FullPath: fullpath,
+	}
+	if err := entry.DecodeAttributesAndChunks(data); err != nil {
+		return entry, fmt.Errorf("decode %s : %v", entry.FullPath, err)
+	}
+
+	return entry, nil
+}
+
+func (store *AbstractSqlStore) DeleteEntry(fullpath filer2.FullPath) (error) {
+
+	dir, name := fullpath.DirAndName()
+
+	res, err := store.DB.Exec(store.SqlDelete, hashToLong(dir), name, dir)
+	if err != nil {
+		return fmt.Errorf("delete %s: %s", fullpath, err)
+	}
+
+	_, err = res.RowsAffected()
+	if err != nil {
+		return fmt.Errorf("delete %s but no rows affected: %s", fullpath, err)
+	}
+
+	return nil
+}
+
+func (store *AbstractSqlStore) ListDirectoryEntries(fullpath filer2.FullPath, startFileName string, inclusive bool, limit int) (entries []*filer2.Entry, err error) {
+
+	sqlText := store.SqlListExclusive
+	if inclusive {
+		sqlText = store.SqlListInclusive
+	}
+
+	rows, err := store.DB.Query(sqlText, hashToLong(string(fullpath)), startFileName, string(fullpath), limit)
+	if err != nil {
+		return nil, fmt.Errorf("list %s : %v", fullpath, err)
+	}
+	defer rows.Close()
+
+	for rows.Next() {
+		var name string
+		var data []byte
+		if err = rows.Scan(&name, &data); err != nil {
+			glog.V(0).Infof("scan %s : %v", fullpath, err)
+			return nil, fmt.Errorf("scan %s: %v", fullpath, err)
+		}
+
+		entry := &filer2.Entry{
+			FullPath: filer2.NewFullPath(string(fullpath), name),
+		}
+		if err = entry.DecodeAttributesAndChunks(data); err != nil {
+			glog.V(0).Infof("scan decode %s : %v", entry.FullPath, err)
+			return nil, fmt.Errorf("scan decode %s : %v", entry.FullPath, err)
+		}
+
+		entries = append(entries, entry)
+	}
+
+	return entries, nil
+}

weed/filer2/abstract_sql/hashing.go

@@ -0,0 +1,32 @@
+package abstract_sql
+
+import (
+	"crypto/md5"
+	"io"
+)
+
+// returns a 64 bit big int
+func hashToLong(dir string) (v int64) {
+	h := md5.New()
+	io.WriteString(h, dir)
+
+	b := h.Sum(nil)
+
+	v += int64(b[0])
+	v <<= 8
+	v += int64(b[1])
+	v <<= 8
+	v += int64(b[2])
+	v <<= 8
+	v += int64(b[3])
+	v <<= 8
+	v += int64(b[4])
+	v <<= 8
+	v += int64(b[5])
+	v <<= 8
+	v += int64(b[6])
+	v <<= 8
+	v += int64(b[7])
+
+	return
+}

weed/filer2/cassandra/README.txt

@@ -0,0 +1,14 @@
+1. create a keyspace
+
+CREATE KEYSPACE seaweedfs WITH replication = {'class':'SimpleStrategy', 'replication_factor' : 1};
+
+2. create filemeta table
+
+ USE seaweedfs;
+
+ CREATE TABLE filemeta (
+    directory varchar,
+    name varchar,
+    meta blob,
+    PRIMARY KEY (directory, name)
+ ) WITH CLUSTERING ORDER BY (name ASC);

weed/filer2/cassandra/cassandra_store.go

@@ -0,0 +1,131 @@
+package cassandra
+
+import (
+	"fmt"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/gocql/gocql"
+	"github.com/spf13/viper"
+)
+
+func init() {
+	filer2.Stores = append(filer2.Stores, &CassandraStore{})
+}
+
+type CassandraStore struct {
+	cluster *gocql.ClusterConfig
+	session *gocql.Session
+}
+
+func (store *CassandraStore) GetName() string {
+	return "cassandra"
+}
+
+func (store *CassandraStore) Initialize(viper *viper.Viper) (err error) {
+	return store.initialize(
+		viper.GetString("keyspace"),
+		viper.GetStringSlice("hosts"),
+	)
+}
+
+func (store *CassandraStore) initialize(keyspace string, hosts []string) (err error) {
+	store.cluster = gocql.NewCluster(hosts...)
+	store.cluster.Keyspace = keyspace
+	store.cluster.Consistency = gocql.LocalQuorum
+	store.session, err = store.cluster.CreateSession()
+	if err != nil {
+		glog.V(0).Infof("Failed to open cassandra store, hosts %v, keyspace %s", hosts, keyspace)
+	}
+	return
+}
+
+func (store *CassandraStore) InsertEntry(entry *filer2.Entry) (err error) {
+
+	dir, name := entry.FullPath.DirAndName()
+	meta, err := entry.EncodeAttributesAndChunks()
+	if err != nil {
+		return fmt.Errorf("encode %s: %s", entry.FullPath, err)
+	}
+
+	if err := store.session.Query(
+		"INSERT INTO filemeta (directory,name,meta) VALUES(?,?,?)",
+		dir, name, meta).Exec(); err != nil {
+		return fmt.Errorf("insert %s: %s", entry.FullPath, err)
+	}
+
+	return nil
+}
+
+func (store *CassandraStore) UpdateEntry(entry *filer2.Entry) (err error) {
+
+	return store.InsertEntry(entry)
+}
+
+func (store *CassandraStore) FindEntry(fullpath filer2.FullPath) (entry *filer2.Entry, err error) {
+
+	dir, name := fullpath.DirAndName()
+	var data []byte
+	if err := store.session.Query(
+		"SELECT meta FROM filemeta WHERE directory=? AND name=?",
+		dir, name).Consistency(gocql.One).Scan(&data); err != nil {
+		if err != gocql.ErrNotFound {
+			return nil, fmt.Errorf("read entry %s: %v", fullpath, err)
+		}
+	}
+
+	if len(data) == 0 {
+		return nil, fmt.Errorf("not found: %s", fullpath)
+	}
+
+	entry = &filer2.Entry{
+		FullPath: fullpath,
+	}
+	err = entry.DecodeAttributesAndChunks(data)
+	if err != nil {
+		return entry, fmt.Errorf("decode %s : %v", entry.FullPath, err)
+	}
+
+	return entry, nil
+}
+
+func (store *CassandraStore) DeleteEntry(fullpath filer2.FullPath) error {
+
+	dir, name := fullpath.DirAndName()
+
+	if err := store.session.Query(
+		"DELETE FROM filemeta WHERE directory=? AND name=?",
+		dir, name).Exec(); err != nil {
+		return fmt.Errorf("delete %s : %v", fullpath, err)
+	}
+
+	return nil
+}
+
+func (store *CassandraStore) ListDirectoryEntries(fullpath filer2.FullPath, startFileName string, inclusive bool,
+	limit int) (entries []*filer2.Entry, err error) {
+
+	cqlStr := "SELECT NAME, meta FROM filemeta WHERE directory=? AND name>? ORDER BY NAME ASC LIMIT ?"
+	if inclusive {
+		cqlStr = "SELECT NAME, meta FROM filemeta WHERE directory=? AND name>=? ORDER BY NAME ASC LIMIT ?"
+	}
+
+	var data []byte
+	var name string
+	iter := store.session.Query(cqlStr, string(fullpath), startFileName, limit).Iter()
+	for iter.Scan(&name, &data) {
+		entry := &filer2.Entry{
+			FullPath: filer2.NewFullPath(string(fullpath), name),
+		}
+		if decodeErr := entry.DecodeAttributesAndChunks(data); decodeErr != nil {
+			err = decodeErr
+			glog.V(0).Infof("list %s : %v", entry.FullPath, err)
+			break
+		}
+		entries = append(entries, entry)
+	}
+	if err := iter.Close(); err != nil {
+		glog.V(0).Infof("list iterator close: %v", err)
+	}
+
+	return entries, err
+}

weed/filer2/configuration.go

@@ -0,0 +1,126 @@
+package filer2
+
+import (
+	"os"
+
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/spf13/viper"
+)
+
+const (
+	FILER_TOML_EXAMPLE = `
+# A sample TOML config file for SeaweedFS filer store
+
+[memory]
+# local in memory, mostly for testing purpose
+enabled = false
+
+[leveldb]
+# local on disk, mostly for simple single-machine setup, fairly scalable
+enabled = false
+dir = "."					# directory to store level db files
+
+####################################################
+# multiple filers on shared storage, fairly scalable
+####################################################
+
+[mysql]
+# CREATE TABLE IF NOT EXISTS filemeta (
+#   dirhash     BIGINT        COMMENT 'first 64 bits of MD5 hash value of directory field',
+#   name        VARCHAR(1000) COMMENT 'directory or file name',
+#   directory   VARCHAR(4096) COMMENT 'full path to parent directory',
+#   meta        BLOB,
+#   PRIMARY KEY (dirhash, name)
+# ) DEFAULT CHARSET=utf8;
+enabled = true
+hostname = "localhost"
+port = 3306
+username = "root"
+password = ""
+database = ""              # create or use an existing database
+connection_max_idle = 2
+connection_max_open = 100
+
+[postgres]
+# CREATE TABLE IF NOT EXISTS filemeta (
+#   dirhash     BIGINT,
+#   name        VARCHAR(1000),
+#   directory   VARCHAR(4096),
+#   meta        bytea,
+#   PRIMARY KEY (dirhash, name)
+# );
+enabled = false
+hostname = "localhost"
+port = 5432
+username = "postgres"
+password = ""
+database = ""              # create or use an existing database
+sslmode = "disable"
+connection_max_idle = 100
+connection_max_open = 100
+
+[cassandra]
+# CREATE TABLE filemeta (
+#    directory varchar,
+#    name varchar,
+#    meta blob,
+#    PRIMARY KEY (directory, name)
+# ) WITH CLUSTERING ORDER BY (name ASC);
+enabled = false
+keyspace="seaweedfs"
+hosts=[
+	"localhost:9042",
+]
+
+[redis]
+enabled = true
+address  = "localhost:6379"
+password = ""
+db = 0
+
+`
+)
+
+var (
+	Stores []FilerStore
+)
+
+func (f *Filer) LoadConfiguration() {
+
+	// find a filer store
+	viper.SetConfigName("filer")                 // name of config file (without extension)
+	viper.AddConfigPath(".")                     // optionally look for config in the working directory
+	viper.AddConfigPath("$HOME/.seaweedfs")      // call multiple times to add many search paths
+	viper.AddConfigPath("/etc/seaweedfs/")       // path to look for the config file in
+	if err := viper.ReadInConfig(); err != nil { // Handle errors reading the config file
+		glog.Fatalf("Failed to load filer.toml file from current directory, or $HOME/.seaweedfs/, or /etc/seaweedfs/" +
+			"\n\nPlease follow this example and add a filer.toml file to " +
+			"current directory, or $HOME/.seaweedfs/, or /etc/seaweedfs/:\n" + FILER_TOML_EXAMPLE)
+	}
+
+	glog.V(0).Infof("Reading filer configuration from %s", viper.ConfigFileUsed())
+	for _, store := range Stores {
+		if viper.GetBool(store.GetName() + ".enabled") {
+			viperSub := viper.Sub(store.GetName())
+			if err := store.Initialize(viperSub); err != nil {
+				glog.Fatalf("Failed to initialize store for %s: %+v",
+					store.GetName(), err)
+			}
+			f.SetStore(store)
+			glog.V(0).Infof("Configure filer for %s from %s", store.GetName(), viper.ConfigFileUsed())
+			return
+		}
+	}
+
+	println()
+	println("Supported filer stores are:")
+	for _, store := range Stores {
+		println("    " + store.GetName())
+	}
+
+	println()
+	println("Please configure a supported filer store in", viper.ConfigFileUsed())
+	println()
+
+	os.Exit(-1)
+}

weed/filer2/embedded/embedded_store.go

@@ -1,42 +0,0 @@
-package embedded
-
-import (
-	"github.com/syndtr/goleveldb/leveldb"
-	"github.com/chrislusf/seaweedfs/weed/filer2"
-)
-
-type EmbeddedStore struct {
-	db *leveldb.DB
-}
-
-func NewEmbeddedStore(dir string) (filer *EmbeddedStore, err error) {
-	filer = &EmbeddedStore{}
-	if filer.db, err = leveldb.OpenFile(dir, nil); err != nil {
-		return
-	}
-	return
-}
-
-func (filer *EmbeddedStore) InsertEntry(entry *filer2.Entry) (err error) {
-	return nil
-}
-
-func (filer *EmbeddedStore) AddDirectoryLink(directory *filer2.Entry, delta int32) (err error) {
-	return nil
-}
-
-func (filer *EmbeddedStore) AppendFileChunk(fullpath filer2.FullPath, fileChunk filer2.FileChunk) (err error) {
-	return nil
-}
-
-func (filer *EmbeddedStore) FindEntry(fullpath filer2.FullPath) (found bool, entry *filer2.Entry, err error) {
-	return false, nil, nil
-}
-
-func (filer *EmbeddedStore) DeleteEntry(fullpath filer2.FullPath) (entry *filer2.Entry, err error) {
-	return nil, nil
-}
-
-func (filer *EmbeddedStore) ListDirectoryEntries(fullpath filer2.FullPath) (entries []*filer2.Entry, err error) {
-	return nil, nil
-}

weed/filer2/entry.go

@@ -0,0 +1,42 @@
+package filer2
+
+import (
+	"os"
+	"time"
+
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+)
+
+type Attr struct {
+	Mtime  time.Time   // time of last modification
+	Crtime time.Time   // time of creation (OS X only)
+	Mode   os.FileMode // file mode
+	Uid    uint32      // owner uid
+	Gid    uint32      // group gid
+	Mime   string
+}
+
+func (attr Attr) IsDirectory() bool {
+	return attr.Mode&os.ModeDir > 0
+}
+
+type Entry struct {
+	FullPath
+
+	Attr
+
+	// the following is for files
+	Chunks []*filer_pb.FileChunk `json:"chunks,omitempty"`
+}
+
+func (entry *Entry) Size() uint64 {
+	return TotalSize(entry.Chunks)
+}
+
+func (entry *Entry) Timestamp() time.Time {
+	if entry.IsDirectory() {
+		return entry.Crtime
+	} else {
+		return entry.Mtime
+	}
+}

weed/filer2/entry_codec.go

@@ -0,0 +1,45 @@
+package filer2
+
+import (
+	"os"
+	"time"
+
+	"fmt"
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"github.com/gogo/protobuf/proto"
+)
+
+func (entry *Entry) EncodeAttributesAndChunks() ([]byte, error) {
+	message := &filer_pb.Entry{
+		Attributes: &filer_pb.FuseAttributes{
+			Crtime:   entry.Attr.Crtime.Unix(),
+			Mtime:    entry.Attr.Mtime.Unix(),
+			FileMode: uint32(entry.Attr.Mode),
+			Uid:      entry.Uid,
+			Gid:      entry.Gid,
+			Mime:     entry.Mime,
+		},
+		Chunks: entry.Chunks,
+	}
+	return proto.Marshal(message)
+}
+
+func (entry *Entry) DecodeAttributesAndChunks(blob []byte) error {
+
+	message := &filer_pb.Entry{}
+
+	if err := proto.UnmarshalMerge(blob, message); err != nil {
+		return fmt.Errorf("decoding value blob for %s: %v", entry.FullPath, err)
+	}
+
+	entry.Attr.Crtime = time.Unix(message.Attributes.Crtime, 0)
+	entry.Attr.Mtime = time.Unix(message.Attributes.Mtime, 0)
+	entry.Attr.Mode = os.FileMode(message.Attributes.FileMode)
+	entry.Attr.Uid = message.Attributes.Uid
+	entry.Attr.Gid = message.Attributes.Gid
+	entry.Attr.Mime = message.Attributes.Mime
+
+	entry.Chunks = message.Chunks
+
+	return nil
+}

weed/filer2/filechunks.go

@@ -1,8 +1,13 @@
 package filer2
 
-type Chunks []FileChunk
+import (
+	"math"
+	"sort"
 
-func (chunks Chunks) TotalSize() (size uint64) {
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+)
+
+func TotalSize(chunks []*filer_pb.FileChunk) (size uint64) {
 	for _, c := range chunks {
 		t := uint64(c.Offset + int64(c.Size))
 		if size < t {
@@ -12,12 +17,236 @@ func (chunks Chunks) TotalSize() (size uint64) {
 	return
 }
 
-func (chunks Chunks) Len() int {
-	return len(chunks)
+func CompactFileChunks(chunks []*filer_pb.FileChunk) (compacted, garbage []*filer_pb.FileChunk) {
+
+	visibles := nonOverlappingVisibleIntervals(chunks)
+
+	fileIds := make(map[string]bool)
+	for _, interval := range visibles {
+		fileIds[interval.fileId] = true
+	}
+	for _, chunk := range chunks {
+		if found := fileIds[chunk.FileId]; found {
+			compacted = append(compacted, chunk)
+		} else {
+			garbage = append(garbage, chunk)
+		}
+	}
+
+	return
+}
+
+func FindUnusedFileChunks(oldChunks, newChunks []*filer_pb.FileChunk) (unused []*filer_pb.FileChunk) {
+
+	fileIds := make(map[string]bool)
+	for _, interval := range newChunks {
+		fileIds[interval.FileId] = true
+	}
+	for _, chunk := range oldChunks {
+		if found := fileIds[chunk.FileId]; !found {
+			unused = append(unused, chunk)
+		}
+	}
+
+	return
+}
+
+type ChunkView struct {
+	FileId      string
+	Offset      int64
+	Size        uint64
+	LogicOffset int64
+}
+
+func ViewFromChunks(chunks []*filer_pb.FileChunk, offset int64, size int) (views []*ChunkView) {
+
+	visibles := nonOverlappingVisibleIntervals(chunks)
+
+	stop := offset + int64(size)
+
+	for _, chunk := range visibles {
+		if chunk.start <= offset && offset < chunk.stop && offset < stop {
+			views = append(views, &ChunkView{
+				FileId:      chunk.fileId,
+				Offset:      offset - chunk.start, // offset is the data starting location in this file id
+				Size:        uint64(min(chunk.stop, stop) - offset),
+				LogicOffset: offset,
+			})
+			offset = min(chunk.stop, stop)
+		}
+	}
+
+	return views
+
+}
+
+func logPrintf(name string, visibles []*visibleInterval) {
+	/*
+		log.Printf("%s len %d", name, len(visibles))
+		for _, v := range visibles {
+			log.Printf("%s:  => %+v", name, v)
+		}
+	*/
+}
+
+func nonOverlappingVisibleIntervals(chunks []*filer_pb.FileChunk) (visibles []*visibleInterval) {
+
+	sort.Slice(chunks, func(i, j int) bool {
+		if chunks[i].Offset < chunks[j].Offset {
+			return true
+		}
+		if chunks[i].Offset == chunks[j].Offset {
+			return chunks[i].Mtime < chunks[j].Mtime
+		}
+		return false
+	})
+
+	if len(chunks) == 0 {
+		return
+	}
+
+	var parallelIntervals, intervals []*visibleInterval
+	var minStopInterval, upToDateInterval *visibleInterval
+	watermarkStart := chunks[0].Offset
+	for _, chunk := range chunks {
+		// log.Printf("checking chunk: [%d,%d)", chunk.Offset, chunk.Offset+int64(chunk.Size))
+		logPrintf("parallelIntervals", parallelIntervals)
+		for len(parallelIntervals) > 0 && watermarkStart < chunk.Offset {
+			logPrintf("parallelIntervals loop 1", parallelIntervals)
+			logPrintf("parallelIntervals loop 1 intervals", intervals)
+			minStopInterval, upToDateInterval = findMinStopInterval(parallelIntervals)
+			nextStop := min(minStopInterval.stop, chunk.Offset)
+			intervals = append(intervals, newVisibleInterval(
+				max(watermarkStart, minStopInterval.start),
+				nextStop,
+				upToDateInterval.fileId,
+				upToDateInterval.modifiedTime,
+			))
+			watermarkStart = nextStop
+			logPrintf("parallelIntervals loop intervals =>", intervals)
+
+			// remove processed intervals, possibly multiple
+			var remaining []*visibleInterval
+			for _, interval := range parallelIntervals {
+				if interval.stop != watermarkStart {
+					remaining = append(remaining, interval)
+				}
+			}
+			parallelIntervals = remaining
+			logPrintf("parallelIntervals loop 2", parallelIntervals)
+			logPrintf("parallelIntervals loop 2 intervals", intervals)
+		}
+		parallelIntervals = append(parallelIntervals, newVisibleInterval(
+			chunk.Offset,
+			chunk.Offset+int64(chunk.Size),
+			chunk.FileId,
+			chunk.Mtime,
+		))
+	}
+
+	logPrintf("parallelIntervals loop 3", parallelIntervals)
+	logPrintf("parallelIntervals loop 3 intervals", intervals)
+	for len(parallelIntervals) > 0 {
+		minStopInterval, upToDateInterval = findMinStopInterval(parallelIntervals)
+		intervals = append(intervals, newVisibleInterval(
+			max(watermarkStart, minStopInterval.start),
+			minStopInterval.stop,
+			upToDateInterval.fileId,
+			upToDateInterval.modifiedTime,
+		))
+		watermarkStart = minStopInterval.stop
+
+		// remove processed intervals, possibly multiple
+		var remaining []*visibleInterval
+		for _, interval := range parallelIntervals {
+			if interval.stop != watermarkStart {
+				remaining = append(remaining, interval)
+			}
+		}
+		parallelIntervals = remaining
+	}
+	logPrintf("parallelIntervals loop 4", parallelIntervals)
+	logPrintf("intervals", intervals)
+
+	// merge connected intervals, now the intervals are non-intersecting
+	var lastIntervalIndex int
+	var prevIntervalIndex int
+	for i, interval := range intervals {
+		if i == 0 {
+			prevIntervalIndex = i
+			lastIntervalIndex = i
+			continue
+		}
+		if intervals[i-1].fileId != interval.fileId ||
+			intervals[i-1].stop < intervals[i].start {
+			visibles = append(visibles, newVisibleInterval(
+				intervals[prevIntervalIndex].start,
+				intervals[i-1].stop,
+				intervals[prevIntervalIndex].fileId,
+				intervals[prevIntervalIndex].modifiedTime,
+			))
+			prevIntervalIndex = i
+		}
+		lastIntervalIndex = i
+		logPrintf("intervals loop 1 visibles", visibles)
+	}
+
+	visibles = append(visibles, newVisibleInterval(
+		intervals[prevIntervalIndex].start,
+		intervals[lastIntervalIndex].stop,
+		intervals[prevIntervalIndex].fileId,
+		intervals[prevIntervalIndex].modifiedTime,
+	))
+
+	logPrintf("visibles", visibles)
+
+	return
+}
+
+func findMinStopInterval(intervals []*visibleInterval) (minStopInterval, upToDateInterval *visibleInterval) {
+	var latestMtime int64
+	latestIntervalIndex := 0
+	minStop := int64(math.MaxInt64)
+	minIntervalIndex := 0
+	for i, interval := range intervals {
+		if minStop > interval.stop {
+			minIntervalIndex = i
+			minStop = interval.stop
+		}
+		if latestMtime < interval.modifiedTime {
+			latestMtime = interval.modifiedTime
+			latestIntervalIndex = i
+		}
+	}
+	minStopInterval = intervals[minIntervalIndex]
+	upToDateInterval = intervals[latestIntervalIndex]
+	return
+}
+
+// find non-overlapping visible intervals
+// visible interval map to one file chunk
+
+type visibleInterval struct {
+	start        int64
+	stop         int64
+	modifiedTime int64
+	fileId       string
 }
-func (chunks Chunks) Swap(i, j int) {
-	chunks[i], chunks[j] = chunks[j], chunks[i]
+
+func newVisibleInterval(start, stop int64, fileId string, modifiedTime int64) *visibleInterval {
+	return &visibleInterval{start: start, stop: stop, fileId: fileId, modifiedTime: modifiedTime}
 }
-func (chunks Chunks) Less(i, j int) bool {
-	return chunks[i].Offset < chunks[j].Offset
+
+func min(x, y int64) int64 {
+	if x <= y {
+		return x
+	}
+	return y
+}
+
+func max(x, y int64) int64 {
+	if x > y {
+		return x
+	}
+	return y
 }

weed/filer2/filechunks_test.go

@@ -0,0 +1,316 @@
+package filer2
+
+import (
+	"log"
+	"testing"
+
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+)
+
+func TestCompactFileChunks(t *testing.T) {
+	chunks := []*filer_pb.FileChunk{
+		{Offset: 10, Size: 100, FileId: "abc", Mtime: 50},
+		{Offset: 100, Size: 100, FileId: "def", Mtime: 100},
+		{Offset: 200, Size: 100, FileId: "ghi", Mtime: 200},
+		{Offset: 110, Size: 200, FileId: "jkl", Mtime: 300},
+	}
+
+	compacted, garbarge := CompactFileChunks(chunks)
+
+	log.Printf("Compacted: %+v", compacted)
+	log.Printf("Garbage  : %+v", garbarge)
+
+	if len(compacted) != 3 {
+		t.Fatalf("unexpected compacted: %d", len(compacted))
+	}
+	if len(garbarge) != 1 {
+		t.Fatalf("unexpected garbarge: %d", len(garbarge))
+	}
+
+}
+
+func TestIntervalMerging(t *testing.T) {
+
+	testcases := []struct {
+		Chunks   []*filer_pb.FileChunk
+		Expected []*visibleInterval
+	}{
+		// case 0: normal
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 100, Size: 100, FileId: "asdf", Mtime: 134},
+				{Offset: 200, Size: 100, FileId: "fsad", Mtime: 353},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 100, fileId: "abc"},
+				{start: 100, stop: 200, fileId: "asdf"},
+				{start: 200, stop: 300, fileId: "fsad"},
+			},
+		},
+		// case 1: updates overwrite full chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 134},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 200, fileId: "asdf"},
+			},
+		},
+		// case 2: updates overwrite part of previous chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 50, FileId: "asdf", Mtime: 134},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 50, fileId: "asdf"},
+				{start: 50, stop: 100, fileId: "abc"},
+			},
+		},
+		// case 3: updates overwrite full chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 134},
+				{Offset: 50, Size: 250, FileId: "xxxx", Mtime: 154},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 50, fileId: "asdf"},
+				{start: 50, stop: 300, fileId: "xxxx"},
+			},
+		},
+		// case 4: updates far away from prev chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 134},
+				{Offset: 250, Size: 250, FileId: "xxxx", Mtime: 154},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 200, fileId: "asdf"},
+				{start: 250, stop: 500, fileId: "xxxx"},
+			},
+		},
+		// case 5: updates overwrite full chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 184},
+				{Offset: 70, Size: 150, FileId: "abc", Mtime: 143},
+				{Offset: 80, Size: 100, FileId: "xxxx", Mtime: 134},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 200, fileId: "asdf"},
+				{start: 200, stop: 220, fileId: "abc"},
+			},
+		},
+		// case 6: same updates
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 100, fileId: "abc"},
+			},
+		},
+		// case 7: real updates
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 2097152, FileId: "7,0294cbb9892b", Mtime: 123},
+				{Offset: 0, Size: 3145728, FileId: "3,029565bf3092", Mtime: 130},
+				{Offset: 2097152, Size: 3145728, FileId: "6,029632f47ae2", Mtime: 140},
+				{Offset: 5242880, Size: 3145728, FileId: "2,029734c5aa10", Mtime: 150},
+				{Offset: 8388608, Size: 3145728, FileId: "5,02982f80de50", Mtime: 160},
+				{Offset: 11534336, Size: 2842193, FileId: "7,0299ad723803", Mtime: 170},
+			},
+			Expected: []*visibleInterval{
+				{start: 0, stop: 2097152, fileId: "3,029565bf3092"},
+				{start: 2097152, stop: 5242880, fileId: "6,029632f47ae2"},
+				{start: 5242880, stop: 8388608, fileId: "2,029734c5aa10"},
+				{start: 8388608, stop: 11534336, fileId: "5,02982f80de50"},
+				{start: 11534336, stop: 14376529, fileId: "7,0299ad723803"},
+			},
+		},
+	}
+
+	for i, testcase := range testcases {
+		log.Printf("++++++++++ merged test case %d ++++++++++++++++++++", i)
+		intervals := nonOverlappingVisibleIntervals(testcase.Chunks)
+		for x, interval := range intervals {
+			log.Printf("test case %d, interval %d, start=%d, stop=%d, fileId=%s",
+				i, x, interval.start, interval.stop, interval.fileId)
+		}
+		for x, interval := range intervals {
+			if interval.start != testcase.Expected[x].start {
+				t.Fatalf("failed on test case %d, interval %d, start %d, expect %d",
+					i, x, interval.start, testcase.Expected[x].start)
+			}
+			if interval.stop != testcase.Expected[x].stop {
+				t.Fatalf("failed on test case %d, interval %d, stop %d, expect %d",
+					i, x, interval.stop, testcase.Expected[x].stop)
+			}
+			if interval.fileId != testcase.Expected[x].fileId {
+				t.Fatalf("failed on test case %d, interval %d, chunkId %s, expect %s",
+					i, x, interval.fileId, testcase.Expected[x].fileId)
+			}
+		}
+		if len(intervals) != len(testcase.Expected) {
+			t.Fatalf("failed to compact test case %d, len %d expected %d", i, len(intervals), len(testcase.Expected))
+		}
+	}
+
+}
+
+func TestChunksReading(t *testing.T) {
+
+	testcases := []struct {
+		Chunks   []*filer_pb.FileChunk
+		Offset   int64
+		Size     int
+		Expected []*ChunkView
+	}{
+		// case 0: normal
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 100, Size: 100, FileId: "asdf", Mtime: 134},
+				{Offset: 200, Size: 100, FileId: "fsad", Mtime: 353},
+			},
+			Offset: 0,
+			Size:   250,
+			Expected: []*ChunkView{
+				{Offset: 0, Size: 100, FileId: "abc", LogicOffset: 0},
+				{Offset: 0, Size: 100, FileId: "asdf", LogicOffset: 100},
+				{Offset: 0, Size: 50, FileId: "fsad", LogicOffset: 200},
+			},
+		},
+		// case 1: updates overwrite full chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 134},
+			},
+			Offset: 50,
+			Size:   100,
+			Expected: []*ChunkView{
+				{Offset: 50, Size: 100, FileId: "asdf", LogicOffset: 50},
+			},
+		},
+		// case 2: updates overwrite part of previous chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 50, FileId: "asdf", Mtime: 134},
+			},
+			Offset: 25,
+			Size:   50,
+			Expected: []*ChunkView{
+				{Offset: 25, Size: 25, FileId: "asdf", LogicOffset: 25},
+				{Offset: 0, Size: 25, FileId: "abc", LogicOffset: 50},
+			},
+		},
+		// case 3: updates overwrite full chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 134},
+				{Offset: 50, Size: 250, FileId: "xxxx", Mtime: 154},
+			},
+			Offset: 0,
+			Size:   200,
+			Expected: []*ChunkView{
+				{Offset: 0, Size: 50, FileId: "asdf", LogicOffset: 0},
+				{Offset: 0, Size: 150, FileId: "xxxx", LogicOffset: 50},
+			},
+		},
+		// case 4: updates far away from prev chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 134},
+				{Offset: 250, Size: 250, FileId: "xxxx", Mtime: 154},
+			},
+			Offset: 0,
+			Size:   400,
+			Expected: []*ChunkView{
+				{Offset: 0, Size: 200, FileId: "asdf", LogicOffset: 0},
+				// {Offset: 0, Size: 150, FileId: "xxxx"}, // missing intervals should not happen
+			},
+		},
+		// case 5: updates overwrite full chunks
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 200, FileId: "asdf", Mtime: 184},
+				{Offset: 70, Size: 150, FileId: "abc", Mtime: 143},
+				{Offset: 80, Size: 100, FileId: "xxxx", Mtime: 134},
+			},
+			Offset: 0,
+			Size:   220,
+			Expected: []*ChunkView{
+				{Offset: 0, Size: 200, FileId: "asdf", LogicOffset: 0},
+				{Offset: 0, Size: 20, FileId: "abc", LogicOffset: 200},
+			},
+		},
+		// case 6: same updates
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+			},
+			Offset: 0,
+			Size:   100,
+			Expected: []*ChunkView{
+				{Offset: 0, Size: 100, FileId: "abc", LogicOffset: 0},
+			},
+		},
+		// case 7: edge cases
+		{
+			Chunks: []*filer_pb.FileChunk{
+				{Offset: 0, Size: 100, FileId: "abc", Mtime: 123},
+				{Offset: 100, Size: 100, FileId: "asdf", Mtime: 134},
+				{Offset: 200, Size: 100, FileId: "fsad", Mtime: 353},
+			},
+			Offset: 0,
+			Size:   200,
+			Expected: []*ChunkView{
+				{Offset: 0, Size: 100, FileId: "abc", LogicOffset: 0},
+				{Offset: 0, Size: 100, FileId: "asdf", LogicOffset: 100},
+			},
+		},
+	}
+
+	for i, testcase := range testcases {
+		log.Printf("++++++++++ read test case %d ++++++++++++++++++++", i)
+		chunks := ViewFromChunks(testcase.Chunks, testcase.Offset, testcase.Size)
+		for x, chunk := range chunks {
+			log.Printf("read case %d, chunk %d, offset=%d, size=%d, fileId=%s",
+				i, x, chunk.Offset, chunk.Size, chunk.FileId)
+			if chunk.Offset != testcase.Expected[x].Offset {
+				t.Fatalf("failed on read case %d, chunk %d, Offset %d, expect %d",
+					i, x, chunk.Offset, testcase.Expected[x].Offset)
+			}
+			if chunk.Size != testcase.Expected[x].Size {
+				t.Fatalf("failed on read case %d, chunk %d, Size %d, expect %d",
+					i, x, chunk.Size, testcase.Expected[x].Size)
+			}
+			if chunk.FileId != testcase.Expected[x].FileId {
+				t.Fatalf("failed on read case %d, chunk %d, FileId %s, expect %s",
+					i, x, chunk.FileId, testcase.Expected[x].FileId)
+			}
+			if chunk.LogicOffset != testcase.Expected[x].LogicOffset {
+				t.Fatalf("failed on read case %d, chunk %d, LogicOffset %d, expect %d",
+					i, x, chunk.LogicOffset, testcase.Expected[x].LogicOffset)
+			}
+		}
+		if len(chunks) != len(testcase.Expected) {
+			t.Fatalf("failed to read test case %d, len %d expected %d", i, len(chunks), len(testcase.Expected))
+		}
+	}
+
+}

weed/filer2/filer.go

@@ -3,35 +3,39 @@ package filer2
 import (
 	"fmt"
 
+	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/karlseguin/ccache"
-	"strings"
+	"os"
 	"path/filepath"
+	"strings"
 	"time"
-	"os"
+	"github.com/chrislusf/seaweedfs/weed/operation"
 )
 
 type Filer struct {
-	master         string
+	masters        []string
 	store          FilerStore
 	directoryCache *ccache.Cache
+
+	currentMaster string
 }
 
-func NewFiler(master string) *Filer {
+func NewFiler(masters []string) *Filer {
 	return &Filer{
-		master:         master,
+		masters:        masters,
 		directoryCache: ccache.New(ccache.Configure().MaxSize(1000).ItemsToPrune(100)),
 	}
 }
 
-func (f *Filer) SetStore(store FilerStore) () {
+func (f *Filer) SetStore(store FilerStore) {
 	f.store = store
 }
 
-func (f *Filer) DisableDirectoryCache() () {
+func (f *Filer) DisableDirectoryCache() {
 	f.directoryCache = nil
 }
 
-func (f *Filer) CreateEntry(entry *Entry) (error) {
+func (f *Filer) CreateEntry(entry *Entry) error {
 
 	dirParts := strings.Split(string(entry.FullPath), "/")
 
@@ -43,22 +47,19 @@ func (f *Filer) CreateEntry(entry *Entry) (error) {
 		dirPath := "/" + filepath.Join(dirParts[:i]...)
 		// fmt.Printf("%d directory: %+v\n", i, dirPath)
 
-		dirFound := false
-
 		// first check local cache
 		dirEntry := f.cacheGetDirectory(dirPath)
 
 		// not found, check the store directly
 		if dirEntry == nil {
-			var dirFindErr error
-			dirFound, dirEntry, dirFindErr = f.FindEntry(FullPath(dirPath))
-			if dirFindErr != nil {
-				return fmt.Errorf("findDirectory %s: %v", dirPath, dirFindErr)
-			}
+			glog.V(4).Infof("find uncached directory: %s", dirPath)
+			dirEntry, _ = f.FindEntry(FullPath(dirPath))
+		} else {
+			glog.V(4).Infof("found cached directory: %s", dirPath)
 		}
 
 		// no such existing directory
-		if !dirFound {
+		if dirEntry == nil {
 
 			// create the directory
 			now := time.Now()
@@ -68,12 +69,13 @@ func (f *Filer) CreateEntry(entry *Entry) (error) {
 				Attr: Attr{
 					Mtime:  now,
 					Crtime: now,
-					Mode:   os.ModeDir | 0660,
+					Mode:   os.ModeDir | 0770,
 					Uid:    entry.Uid,
 					Gid:    entry.Gid,
 				},
 			}
 
+			glog.V(2).Infof("create directory: %s %v", dirPath, dirEntry.Mode)
 			mkdirErr := f.store.InsertEntry(dirEntry)
 			if mkdirErr != nil {
 				return fmt.Errorf("mkdir %s: %v", dirPath, mkdirErr)
@@ -94,9 +96,17 @@ func (f *Filer) CreateEntry(entry *Entry) (error) {
 		return fmt.Errorf("parent folder not found: %v", entry.FullPath)
 	}
 
+	/*
 		if !hasWritePermission(lastDirectoryEntry, entry) {
+			glog.V(0).Infof("directory %s: %v, entry: uid=%d gid=%d",
+				lastDirectoryEntry.FullPath, lastDirectoryEntry.Attr, entry.Uid, entry.Gid)
 			return fmt.Errorf("no write permission in folder %v", lastDirectoryEntry.FullPath)
 		}
+	*/
+
+	if oldEntry, err := f.FindEntry(entry.FullPath); err == nil {
+		f.deleteChunks(oldEntry)
+	}
 
 	if err := f.store.InsertEntry(entry); err != nil {
 		return fmt.Errorf("insert entry %s: %v", entry.FullPath, err)
@@ -105,26 +115,43 @@ func (f *Filer) CreateEntry(entry *Entry) (error) {
 	return nil
 }
 
-func (f *Filer) AppendFileChunk(p FullPath, c FileChunk) (err error) {
-	return f.store.AppendFileChunk(p, c)
+func (f *Filer) UpdateEntry(entry *Entry) (err error) {
+	return f.store.UpdateEntry(entry)
 }
 
-func (f *Filer) FindEntry(p FullPath) (found bool, entry *Entry, err error) {
+func (f *Filer) FindEntry(p FullPath) (entry *Entry, err error) {
 	return f.store.FindEntry(p)
 }
 
-func (f *Filer) DeleteEntry(p FullPath) (fileEntry *Entry, err error) {
+func (f *Filer) DeleteEntryMetaAndData(p FullPath) (err error) {
+	entry, err := f.FindEntry(p)
+	if err != nil {
+		return err
+	}
+
+	if entry.IsDirectory() {
+		entries, err := f.ListDirectoryEntries(p, "", false, 1)
+		if err != nil {
+			return fmt.Errorf("list folder %s: %v", p, err)
+		}
+		if len(entries) > 0 {
+			return fmt.Errorf("folder %s is not empty", p)
+		}
+	}
+
+	f.deleteChunks(entry)
+
 	return f.store.DeleteEntry(p)
 }
 
-func (f *Filer) ListDirectoryEntries(p FullPath) ([]*Entry, error) {
-	if strings.HasSuffix(string(p), "/") {
-		p = p[0:len(p)-1]
+func (f *Filer) ListDirectoryEntries(p FullPath, startFileName string, inclusive bool, limit int) ([]*Entry, error) {
+	if strings.HasSuffix(string(p), "/") && len(p) > 1 {
+		p = p[0: len(p)-1]
 	}
-	return f.store.ListDirectoryEntries(p)
+	return f.store.ListDirectoryEntries(p, startFileName, inclusive, limit)
 }
 
-func (f *Filer) cacheGetDirectory(dirpath string) (*Entry) {
+func (f *Filer) cacheGetDirectory(dirpath string) *Entry {
 	if f.directoryCache == nil {
 		return nil
 	}
@@ -148,3 +175,15 @@ func (f *Filer) cacheSetDirectory(dirpath string, dirEntry *Entry, level int) {
 
 	f.directoryCache.Set(dirpath, dirEntry, time.Duration(minutes)*time.Minute)
 }
+
+func (f *Filer) deleteChunks(entry *Entry) {
+
+	if entry == nil {
+		return
+	}
+	for _, chunk := range entry.Chunks {
+		if err := operation.DeleteFile(f.GetMaster(), chunk.FileId, ""); err != nil {
+			glog.V(0).Infof("deleting file %s: %v", chunk.FileId, err)
+		}
+	}
+}

weed/filer2/filer_master.go

@@ -0,0 +1,60 @@
+package filer2
+
+import (
+	"fmt"
+	"context"
+	"time"
+
+	"github.com/chrislusf/seaweedfs/weed/pb/master_pb"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"google.golang.org/grpc"
+)
+
+func (fs *Filer) GetMaster() string {
+	return fs.currentMaster
+}
+
+func (fs *Filer) KeepConnectedToMaster() {
+	glog.V(0).Infof("Filer bootstraps with masters %v", fs.masters)
+	for _, master := range fs.masters {
+		glog.V(0).Infof("Connecting to %v", master)
+		withMasterClient(master, func(client master_pb.SeaweedClient) error {
+			stream, err := client.KeepConnected(context.Background())
+			if err != nil {
+				glog.V(0).Infof("failed to keep connected to %s: %v", master, err)
+				return err
+			}
+
+			glog.V(0).Infof("Connected to %v", master)
+			fs.currentMaster = master
+
+			for {
+				time.Sleep(time.Duration(float32(10*1e3)*0.25) * time.Millisecond)
+
+				if err = stream.Send(&master_pb.Empty{}); err != nil {
+					glog.V(0).Infof("failed to send to %s: %v", master, err)
+					return err
+				}
+
+				if _, err = stream.Recv(); err != nil {
+					glog.V(0).Infof("failed to receive from %s: %v", master, err)
+					return err
+				}
+			}
+		})
+		fs.currentMaster = ""
+	}
+}
+
+func withMasterClient(master string, fn func(client master_pb.SeaweedClient) error) error {
+
+	grpcConnection, err := grpc.Dial(master, grpc.WithInsecure())
+	if err != nil {
+		return fmt.Errorf("fail to dial %s: %v", master, err)
+	}
+	defer grpcConnection.Close()
+
+	client := master_pb.NewSeaweedClient(grpcConnection)
+
+	return fn(client)
+}

weed/filer2/filer_structure.go

@@ -1,66 +0,0 @@
-package filer2
-
-import (
-	"errors"
-	"os"
-	"time"
-	"path/filepath"
-)
-
-type FileId string //file id in SeaweedFS
-type FullPath string
-
-func (fp FullPath) DirAndName() (string, string) {
-	dir, name := filepath.Split(string(fp))
-	if dir == "/" {
-		return dir, name
-	}
-	if len(dir) < 1 {
-		return "/", ""
-	}
-	return dir[:len(dir)-1], name
-}
-
-type Attr struct {
-	Mtime  time.Time   // time of last modification
-	Crtime time.Time   // time of creation (OS X only)
-	Mode   os.FileMode // file mode
-	Uid    uint32      // owner uid
-	Gid    uint32      // group gid
-}
-
-type Entry struct {
-	FullPath
-
-	Attr
-
-	// the following is for files
-	Chunks []FileChunk `json:"chunks,omitempty"`
-}
-
-type FileChunk struct {
-	Fid    FileId `json:"fid,omitempty"`
-	Offset int64  `json:"offset,omitempty"`
-	Size   uint64 `json:"size,omitempty"` // size in bytes
-}
-
-type AbstractFiler interface {
-	CreateEntry(*Entry) (error)
-	AppendFileChunk(FullPath, FileChunk) (err error)
-	FindEntry(FullPath) (found bool, fileEntry *Entry, err error)
-	DeleteEntry(FullPath) (fileEntry *Entry, err error)
-
-	ListDirectoryEntries(dirPath FullPath) ([]*Entry, error)
-	UpdateEntry(*Entry) (error)
-}
-
-var ErrNotFound = errors.New("filer: no entry is found in filer store")
-
-type FilerStore interface {
-	InsertEntry(*Entry) (error)
-	AppendFileChunk(FullPath, FileChunk) (err error)
-	FindEntry(FullPath) (found bool, entry *Entry, err error)
-	DeleteEntry(FullPath) (fileEntry *Entry, err error)
-
-	ListDirectoryEntries(dirPath FullPath) ([]*Entry, error)
-}

weed/filer2/filerstore.go

@@ -0,0 +1,18 @@
+package filer2
+
+import (
+	"errors"
+	"github.com/spf13/viper"
+)
+
+type FilerStore interface {
+	GetName() string
+	Initialize(viper *viper.Viper) error
+	InsertEntry(*Entry) error
+	UpdateEntry(*Entry) (err error)
+	FindEntry(FullPath) (entry *Entry, err error)
+	DeleteEntry(FullPath) (err error)
+	ListDirectoryEntries(dirPath FullPath, startFileName string, inclusive bool, limit int) ([]*Entry, error)
+}
+
+var ErrNotFound = errors.New("filer: no entry is found in filer store")

weed/filer2/fullpath.go

@@ -0,0 +1,31 @@
+package filer2
+
+import (
+	"path/filepath"
+	"strings"
+)
+
+type FullPath string
+
+func NewFullPath(dir, name string) FullPath {
+	if strings.HasSuffix(dir, "/") {
+		return FullPath(dir + name)
+	}
+	return FullPath(dir + "/" + name)
+}
+
+func (fp FullPath) DirAndName() (string, string) {
+	dir, name := filepath.Split(string(fp))
+	if dir == "/" {
+		return dir, name
+	}
+	if len(dir) < 1 {
+		return "/", ""
+	}
+	return dir[:len(dir)-1], name
+}
+
+func (fp FullPath) Name() string {
+	_, name := filepath.Split(string(fp))
+	return name
+}

weed/filer2/leveldb/leveldb_store.go

@@ -0,0 +1,169 @@
+package leveldb
+
+import (
+	"bytes"
+	"fmt"
+
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	weed_util "github.com/chrislusf/seaweedfs/weed/util"
+	"github.com/spf13/viper"
+	"github.com/syndtr/goleveldb/leveldb"
+	leveldb_util "github.com/syndtr/goleveldb/leveldb/util"
+)
+
+const (
+	DIR_FILE_SEPARATOR = byte(0x00)
+)
+
+func init() {
+	filer2.Stores = append(filer2.Stores, &LevelDBStore{})
+}
+
+type LevelDBStore struct {
+	db *leveldb.DB
+}
+
+func (store *LevelDBStore) GetName() string {
+	return "leveldb"
+}
+
+func (store *LevelDBStore) Initialize(viper *viper.Viper) (err error) {
+	dir := viper.GetString("dir")
+	return store.initialize(dir)
+}
+
+func (store *LevelDBStore) initialize(dir string) (err error) {
+	if err := weed_util.TestFolderWritable(dir); err != nil {
+		return fmt.Errorf("Check Level Folder %s Writable: %s", dir, err)
+	}
+
+	if store.db, err = leveldb.OpenFile(dir, nil); err != nil {
+		return
+	}
+	return
+}
+
+func (store *LevelDBStore) InsertEntry(entry *filer2.Entry) (err error) {
+	key := genKey(entry.DirAndName())
+
+	value, err := entry.EncodeAttributesAndChunks()
+	if err != nil {
+		return fmt.Errorf("encoding %s %+v: %v", entry.FullPath, entry.Attr, err)
+	}
+
+	err = store.db.Put(key, value, nil)
+
+	if err != nil {
+		return fmt.Errorf("persisting %s : %v", entry.FullPath, err)
+	}
+
+	// println("saved", entry.FullPath, "chunks", len(entry.Chunks))
+
+	return nil
+}
+
+func (store *LevelDBStore) UpdateEntry(entry *filer2.Entry) (err error) {
+
+	return store.InsertEntry(entry)
+}
+
+func (store *LevelDBStore) FindEntry(fullpath filer2.FullPath) (entry *filer2.Entry, err error) {
+	key := genKey(fullpath.DirAndName())
+
+	data, err := store.db.Get(key, nil)
+
+	if err == leveldb.ErrNotFound {
+		return nil, filer2.ErrNotFound
+	}
+	if err != nil {
+		return nil, fmt.Errorf("get %s : %v", entry.FullPath, err)
+	}
+
+	entry = &filer2.Entry{
+		FullPath: fullpath,
+	}
+	err = entry.DecodeAttributesAndChunks(data)
+	if err != nil {
+		return entry, fmt.Errorf("decode %s : %v", entry.FullPath, err)
+	}
+
+	// println("read", entry.FullPath, "chunks", len(entry.Chunks), "data", len(data), string(data))
+
+	return entry, nil
+}
+
+func (store *LevelDBStore) DeleteEntry(fullpath filer2.FullPath) (err error) {
+	key := genKey(fullpath.DirAndName())
+
+	err = store.db.Delete(key, nil)
+	if err != nil {
+		return fmt.Errorf("delete %s : %v", fullpath, err)
+	}
+
+	return nil
+}
+
+func (store *LevelDBStore) ListDirectoryEntries(fullpath filer2.FullPath, startFileName string, inclusive bool,
+	limit int) (entries []*filer2.Entry, err error) {
+
+	directoryPrefix := genDirectoryKeyPrefix(fullpath, "")
+
+	iter := store.db.NewIterator(&leveldb_util.Range{Start: genDirectoryKeyPrefix(fullpath, startFileName)}, nil)
+	for iter.Next() {
+		key := iter.Key()
+		if !bytes.HasPrefix(key, directoryPrefix) {
+			break
+		}
+		fileName := getNameFromKey(key)
+		if fileName == "" {
+			continue
+		}
+		if fileName == startFileName && !inclusive {
+			continue
+		}
+		limit--
+		if limit < 0 {
+			break
+		}
+		entry := &filer2.Entry{
+			FullPath: filer2.NewFullPath(string(fullpath), fileName),
+		}
+		if decodeErr := entry.DecodeAttributesAndChunks(iter.Value()); decodeErr != nil {
+			err = decodeErr
+			glog.V(0).Infof("list %s : %v", entry.FullPath, err)
+			break
+		}
+		entries = append(entries, entry)
+	}
+	iter.Release()
+
+	return entries, err
+}
+
+func genKey(dirPath, fileName string) (key []byte) {
+	key = []byte(dirPath)
+	key = append(key, DIR_FILE_SEPARATOR)
+	key = append(key, []byte(fileName)...)
+	return key
+}
+
+func genDirectoryKeyPrefix(fullpath filer2.FullPath, startFileName string) (keyPrefix []byte) {
+	keyPrefix = []byte(string(fullpath))
+	keyPrefix = append(keyPrefix, DIR_FILE_SEPARATOR)
+	if len(startFileName) > 0 {
+		keyPrefix = append(keyPrefix, []byte(startFileName)...)
+	}
+	return keyPrefix
+}
+
+func getNameFromKey(key []byte) string {
+
+	sepIndex := len(key) - 1
+	for sepIndex >= 0 && key[sepIndex] != DIR_FILE_SEPARATOR {
+		sepIndex--
+	}
+
+	return string(key[sepIndex+1:])
+
+}

weed/filer2/leveldb/leveldb_store_test.go

@@ -0,0 +1,61 @@
+package leveldb
+
+import (
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"io/ioutil"
+	"os"
+	"testing"
+)
+
+func TestCreateAndFind(t *testing.T) {
+	filer := filer2.NewFiler(nil)
+	dir, _ := ioutil.TempDir("", "seaweedfs_filer_test")
+	defer os.RemoveAll(dir)
+	store := &LevelDBStore{}
+	store.initialize(dir)
+	filer.SetStore(store)
+	filer.DisableDirectoryCache()
+
+	fullpath := filer2.FullPath("/home/chris/this/is/one/file1.jpg")
+
+	entry1 := &filer2.Entry{
+		FullPath: fullpath,
+		Attr: filer2.Attr{
+			Mode: 0440,
+			Uid:  1234,
+			Gid:  5678,
+		},
+	}
+
+	if err := filer.CreateEntry(entry1); err != nil {
+		t.Errorf("create entry %v: %v", entry1.FullPath, err)
+		return
+	}
+
+	entry, err := filer.FindEntry(fullpath)
+
+	if err != nil {
+		t.Errorf("find entry: %v", err)
+		return
+	}
+
+	if entry.FullPath != entry1.FullPath {
+		t.Errorf("find wrong entry: %v", entry.FullPath)
+		return
+	}
+
+	// checking one upper directory
+	entries, _ := filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is/one"), "", false, 100)
+	if len(entries) != 1 {
+		t.Errorf("list entries count: %v", len(entries))
+		return
+	}
+
+	// checking one upper directory
+	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/"), "", false, 100)
+	if len(entries) != 1 {
+		t.Errorf("list entries count: %v", len(entries))
+		return
+	}
+
+}

weed/filer2/memdb/memdb_store.go

@@ -1,81 +1,102 @@
 package memdb
 
 import (
+	"fmt"
 	"github.com/chrislusf/seaweedfs/weed/filer2"
 	"github.com/google/btree"
+	"github.com/spf13/viper"
 	"strings"
-	"fmt"
-	"time"
 )
 
+func init() {
+	filer2.Stores = append(filer2.Stores, &MemDbStore{})
+}
+
 type MemDbStore struct {
 	tree *btree.BTree
 }
 
-type Entry struct {
+type entryItem struct {
 	*filer2.Entry
 }
 
-func (a Entry) Less(b btree.Item) bool {
-	return strings.Compare(string(a.FullPath), string(b.(Entry).FullPath)) < 0
+func (a entryItem) Less(b btree.Item) bool {
+	return strings.Compare(string(a.FullPath), string(b.(entryItem).FullPath)) < 0
 }
 
-func NewMemDbStore() (filer *MemDbStore) {
-	filer = &MemDbStore{}
-	filer.tree = btree.New(8)
-	return
+func (store *MemDbStore) GetName() string {
+	return "memory"
 }
 
-func (filer *MemDbStore) InsertEntry(entry *filer2.Entry) (err error) {
-	// println("inserting", entry.FullPath)
-	filer.tree.ReplaceOrInsert(Entry{entry})
+func (store *MemDbStore) Initialize(viper *viper.Viper) (err error) {
+	store.tree = btree.New(8)
 	return nil
 }
 
-func (filer *MemDbStore) AppendFileChunk(fullpath filer2.FullPath, fileChunk filer2.FileChunk) (err error) {
-	found, entry, err := filer.FindEntry(fullpath)
-	if !found {
-		return fmt.Errorf("No such file: %s", fullpath)
-	}
-	entry.Chunks = append(entry.Chunks, fileChunk)
-	entry.Mtime = time.Now()
+func (store *MemDbStore) InsertEntry(entry *filer2.Entry) (err error) {
+	// println("inserting", entry.FullPath)
+	store.tree.ReplaceOrInsert(entryItem{entry})
 	return nil
 }
 
-func (filer *MemDbStore) FindEntry(fullpath filer2.FullPath) (found bool, entry *filer2.Entry, err error) {
-	item := filer.tree.Get(Entry{&filer2.Entry{FullPath: fullpath}})
-	if item == nil {
-		return false, nil, nil
+func (store *MemDbStore) UpdateEntry(entry *filer2.Entry) (err error) {
+	if _, err = store.FindEntry(entry.FullPath); err != nil {
+		return fmt.Errorf("no such file %s : %v", entry.FullPath, err)
 	}
-	entry = item.(Entry).Entry
-	return true, entry, nil
+	store.tree.ReplaceOrInsert(entryItem{entry})
+	return nil
 }
 
-func (filer *MemDbStore) DeleteEntry(fullpath filer2.FullPath) (entry *filer2.Entry, err error) {
-	item := filer.tree.Delete(Entry{&filer2.Entry{FullPath: fullpath}})
+func (store *MemDbStore) FindEntry(fullpath filer2.FullPath) (entry *filer2.Entry, err error) {
+	item := store.tree.Get(entryItem{&filer2.Entry{FullPath: fullpath}})
 	if item == nil {
 		return nil, nil
 	}
-	entry = item.(Entry).Entry
+	entry = item.(entryItem).Entry
 	return entry, nil
 }
 
-func (filer *MemDbStore) ListDirectoryEntries(fullpath filer2.FullPath) (entries []*filer2.Entry, err error) {
-	filer.tree.AscendGreaterOrEqual(Entry{&filer2.Entry{FullPath: fullpath}},
+func (store *MemDbStore) DeleteEntry(fullpath filer2.FullPath) (err error) {
+	store.tree.Delete(entryItem{&filer2.Entry{FullPath: fullpath}})
+	return nil
+}
+
+func (store *MemDbStore) ListDirectoryEntries(fullpath filer2.FullPath, startFileName string, inclusive bool, limit int) (entries []*filer2.Entry, err error) {
+
+	startFrom := string(fullpath)
+	if startFileName != "" {
+		startFrom = startFrom + "/" + startFileName
+	}
+
+	store.tree.AscendGreaterOrEqual(entryItem{&filer2.Entry{FullPath: filer2.FullPath(startFrom)}},
 		func(item btree.Item) bool {
-			entry := item.(Entry).Entry
+			if limit <= 0 {
+				return false
+			}
+			entry := item.(entryItem).Entry
 			// println("checking", entry.FullPath)
+
 			if entry.FullPath == fullpath {
 				// skipping the current directory
 				// println("skipping the folder", entry.FullPath)
 				return true
 			}
-			dir, _ := entry.FullPath.DirAndName()
-			if !strings.HasPrefix(dir, string(fullpath)) {
-				// println("directory is:", dir, "fullpath:", fullpath)
+
+			dir, name := entry.FullPath.DirAndName()
+			if name == startFileName {
+				if inclusive {
+					limit--
+					entries = append(entries, entry)
+				}
+				return true
+			}
+
+			// only iterate the same prefix
+			if !strings.HasPrefix(string(entry.FullPath), string(fullpath)) {
 				// println("breaking from", entry.FullPath)
 				return false
 			}
+
 			if dir != string(fullpath) {
 				// this could be items in deeper directories
 				// println("skipping deeper folder", entry.FullPath)
@@ -83,6 +104,7 @@ func (filer *MemDbStore) ListDirectoryEntries(fullpath filer2.FullPath) (entries
 			}
 			// now process the directory items
 			// println("adding entry", entry.FullPath)
+			limit--
 			entries = append(entries, entry)
 			return true
 		},

weed/filer2/memdb/memdb_store_test.go

@@ -1,13 +1,15 @@
 package memdb
 
 import (
-	"testing"
 	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"testing"
 )
 
 func TestCreateAndFind(t *testing.T) {
-	filer := filer2.NewFiler("")
-	filer.SetStore(NewMemDbStore())
+	filer := filer2.NewFiler(nil)
+	store := &MemDbStore{}
+	store.Initialize(nil)
+	filer.SetStore(store)
 	filer.DisableDirectoryCache()
 
 	fullpath := filer2.FullPath("/home/chris/this/is/one/file1.jpg")
@@ -26,18 +28,13 @@ func TestCreateAndFind(t *testing.T) {
 		return
 	}
 
-	found, entry, err := filer.FindEntry(fullpath)
+	entry, err := filer.FindEntry(fullpath)
 
 	if err != nil {
 		t.Errorf("find entry: %v", err)
 		return
 	}
 
-	if !found {
-		t.Errorf("Failed to find newly created file")
-		return
-	}
-
 	if entry.FullPath != entry1.FullPath {
 		t.Errorf("find wrong entry: %v", entry.FullPath)
 		return
@@ -46,8 +43,10 @@ func TestCreateAndFind(t *testing.T) {
 }
 
 func TestCreateFileAndList(t *testing.T) {
-	filer := filer2.NewFiler("")
-	filer.SetStore(NewMemDbStore())
+	filer := filer2.NewFiler(nil)
+	store := &MemDbStore{}
+	store.Initialize(nil)
+	filer.SetStore(store)
 	filer.DisableDirectoryCache()
 
 	entry1 := &filer2.Entry{
@@ -72,7 +71,7 @@ func TestCreateFileAndList(t *testing.T) {
 	filer.CreateEntry(entry2)
 
 	// checking the 2 files
-	entries, err := filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is/one/"))
+	entries, err := filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is/one/"), "", false, 100)
 
 	if err != nil {
 		t.Errorf("list entries: %v", err)
@@ -94,13 +93,28 @@ func TestCreateFileAndList(t *testing.T) {
 		return
 	}
 
+	// checking the offset
+	entries, err = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is/one/"), "file1.jpg", false, 100)
+	if len(entries) != 1 {
+		t.Errorf("list entries count: %v", len(entries))
+		return
+	}
+
 	// checking one upper directory
-	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is"))
+	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is"), "", false, 100)
+	if len(entries) != 1 {
+		t.Errorf("list entries count: %v", len(entries))
+		return
+	}
+
+	// checking root directory
+	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/"), "", false, 100)
 	if len(entries) != 1 {
 		t.Errorf("list entries count: %v", len(entries))
 		return
 	}
 
+	// add file3
 	file3Path := filer2.FullPath("/home/chris/this/is/file3.jpg")
 	entry3 := &filer2.Entry{
 		FullPath: file3Path,
@@ -113,15 +127,15 @@ func TestCreateFileAndList(t *testing.T) {
 	filer.CreateEntry(entry3)
 
 	// checking one upper directory
-	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is"))
+	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is"), "", false, 100)
 	if len(entries) != 2 {
 		t.Errorf("list entries count: %v", len(entries))
 		return
 	}
 
 	// delete file and count
-	filer.DeleteEntry(file3Path)
-	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is"))
+	filer.DeleteEntryMetaAndData(file3Path)
+	entries, _ = filer.ListDirectoryEntries(filer2.FullPath("/home/chris/this/is"), "", false, 100)
 	if len(entries) != 1 {
 		t.Errorf("list entries count: %v", len(entries))
 		return

weed/filer2/mysql/mysql_store.go

@@ -0,0 +1,67 @@
+package mysql
+
+import (
+	"database/sql"
+	"fmt"
+
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/filer2/abstract_sql"
+	_ "github.com/go-sql-driver/mysql"
+	"github.com/spf13/viper"
+)
+
+const (
+	CONNECTION_URL_PATTERN = "%s:%s@tcp(%s:%d)/%s?charset=utf8"
+)
+
+func init() {
+	filer2.Stores = append(filer2.Stores, &MysqlStore{})
+}
+
+type MysqlStore struct {
+	abstract_sql.AbstractSqlStore
+}
+
+func (store *MysqlStore) GetName() string {
+	return "mysql"
+}
+
+func (store *MysqlStore) Initialize(viper *viper.Viper) (err error) {
+	return store.initialize(
+		viper.GetString("username"),
+		viper.GetString("password"),
+		viper.GetString("hostname"),
+		viper.GetInt("port"),
+		viper.GetString("database"),
+		viper.GetInt("connection_max_idle"),
+		viper.GetInt("connection_max_open"),
+	)
+}
+
+func (store *MysqlStore) initialize(user, password, hostname string, port int, database string, maxIdle, maxOpen int) (err error) {
+
+	store.SqlInsert = "INSERT INTO filemeta (dirhash,name,directory,meta) VALUES(?,?,?,?)"
+	store.SqlUpdate = "UPDATE filemeta SET meta=? WHERE dirhash=? AND name=? AND directory=?"
+	store.SqlFind = "SELECT meta FROM filemeta WHERE dirhash=? AND name=? AND directory=?"
+	store.SqlDelete = "DELETE FROM filemeta WHERE dirhash=? AND name=? AND directory=?"
+	store.SqlListExclusive = "SELECT NAME, meta FROM filemeta WHERE dirhash=? AND name>? AND directory=? ORDER BY NAME ASC LIMIT ?"
+	store.SqlListInclusive = "SELECT NAME, meta FROM filemeta WHERE dirhash=? AND name>=? AND directory=? ORDER BY NAME ASC LIMIT ?"
+
+	sqlUrl := fmt.Sprintf(CONNECTION_URL_PATTERN, user, password, hostname, port, database)
+	var dbErr error
+	store.DB, dbErr = sql.Open("mysql", sqlUrl)
+	if dbErr != nil {
+		store.DB.Close()
+		store.DB = nil
+		return fmt.Errorf("can not connect to %s error:%v", sqlUrl, err)
+	}
+
+	store.DB.SetMaxIdleConns(maxIdle)
+	store.DB.SetMaxOpenConns(maxOpen)
+
+	if err = store.DB.Ping(); err != nil {
+		return fmt.Errorf("connect to %s error:%v", sqlUrl, err)
+	}
+
+	return nil
+}

weed/filer2/postgres/README.txt

@@ -0,0 +1,17 @@
+
+1. create "seaweedfs" database
+
+export PGHOME=/Library/PostgreSQL/10
+$PGHOME/bin/createdb --username=postgres --password seaweedfs
+
+2. create "filemeta" table
+$PGHOME/bin/psql --username=postgres --password seaweedfs
+
+CREATE TABLE IF NOT EXISTS filemeta (
+  dirhash     BIGINT,
+  name        VARCHAR(1000),
+  directory   VARCHAR(4096),
+  meta        bytea,
+  PRIMARY KEY (dirhash, name)
+);
+

weed/filer2/postgres/postgres_store.go

@@ -0,0 +1,68 @@
+package postgres
+
+import (
+	"database/sql"
+	"fmt"
+
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/filer2/abstract_sql"
+	_ "github.com/lib/pq"
+	"github.com/spf13/viper"
+)
+
+const (
+	CONNECTION_URL_PATTERN = "host=%s port=%d user=%s password=%s dbname=%s sslmode=%s connect_timeout=30"
+)
+
+func init() {
+	filer2.Stores = append(filer2.Stores, &PostgresStore{})
+}
+
+type PostgresStore struct {
+	abstract_sql.AbstractSqlStore
+}
+
+func (store *PostgresStore) GetName() string {
+	return "postgres"
+}
+
+func (store *PostgresStore) Initialize(viper *viper.Viper) (err error) {
+	return store.initialize(
+		viper.GetString("username"),
+		viper.GetString("password"),
+		viper.GetString("hostname"),
+		viper.GetInt("port"),
+		viper.GetString("database"),
+		viper.GetString("sslmode"),
+		viper.GetInt("connection_max_idle"),
+		viper.GetInt("connection_max_open"),
+	)
+}
+
+func (store *PostgresStore) initialize(user, password, hostname string, port int, database, sslmode string, maxIdle, maxOpen int) (err error) {
+
+	store.SqlInsert = "INSERT INTO filemeta (dirhash,name,directory,meta) VALUES($1,$2,$3,$4)"
+	store.SqlUpdate = "UPDATE filemeta SET meta=$1 WHERE dirhash=$2 AND name=$3 AND directory=$4"
+	store.SqlFind = "SELECT meta FROM filemeta WHERE dirhash=$1 AND name=$2 AND directory=$3"
+	store.SqlDelete = "DELETE FROM filemeta WHERE dirhash=$1 AND name=$2 AND directory=$3"
+	store.SqlListExclusive = "SELECT NAME, meta FROM filemeta WHERE dirhash=$1 AND name>$2 AND directory=$3 ORDER BY NAME ASC LIMIT $4"
+	store.SqlListInclusive = "SELECT NAME, meta FROM filemeta WHERE dirhash=$1 AND name>=$2 AND directory=$3 ORDER BY NAME ASC LIMIT $4"
+
+	sqlUrl := fmt.Sprintf(CONNECTION_URL_PATTERN, hostname, port, user, password, database, sslmode)
+	var dbErr error
+	store.DB, dbErr = sql.Open("postgres", sqlUrl)
+	if dbErr != nil {
+		store.DB.Close()
+		store.DB = nil
+		return fmt.Errorf("can not connect to %s error:%v", sqlUrl, err)
+	}
+
+	store.DB.SetMaxIdleConns(maxIdle)
+	store.DB.SetMaxOpenConns(maxOpen)
+
+	if err = store.DB.Ping(); err != nil {
+		return fmt.Errorf("connect to %s error:%v", sqlUrl, err)
+	}
+
+	return nil
+}

weed/filer2/redis/redis_store.go

@@ -0,0 +1,167 @@
+package redis
+
+import (
+	"fmt"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/go-redis/redis"
+	"github.com/spf13/viper"
+	"sort"
+	"strings"
+)
+
+const (
+	DIR_LIST_MARKER = "\x00"
+)
+
+func init() {
+	filer2.Stores = append(filer2.Stores, &RedisStore{})
+}
+
+type RedisStore struct {
+	Client *redis.Client
+}
+
+func (store *RedisStore) GetName() string {
+	return "redis"
+}
+
+func (store *RedisStore) Initialize(viper *viper.Viper) (err error) {
+	return store.initialize(
+		viper.GetString("address"),
+		viper.GetString("password"),
+		viper.GetInt("database"),
+	)
+}
+
+func (store *RedisStore) initialize(hostPort string, password string, database int) (err error) {
+	store.Client = redis.NewClient(&redis.Options{
+		Addr:     hostPort,
+		Password: password,
+		DB:       database,
+	})
+	return
+}
+
+func (store *RedisStore) InsertEntry(entry *filer2.Entry) (err error) {
+
+	value, err := entry.EncodeAttributesAndChunks()
+	if err != nil {
+		return fmt.Errorf("encoding %s %+v: %v", entry.FullPath, entry.Attr, err)
+	}
+
+	_, err = store.Client.Set(string(entry.FullPath), value, 0).Result()
+
+	if err != nil {
+		return fmt.Errorf("persisting %s : %v", entry.FullPath, err)
+	}
+
+	dir, name := entry.FullPath.DirAndName()
+	if name != "" {
+		_, err = store.Client.SAdd(genDirectoryListKey(dir), name).Result()
+		if err != nil {
+			return fmt.Errorf("persisting %s in parent dir: %v", entry.FullPath, err)
+		}
+	}
+
+	return nil
+}
+
+func (store *RedisStore) UpdateEntry(entry *filer2.Entry) (err error) {
+
+	return store.InsertEntry(entry)
+}
+
+func (store *RedisStore) FindEntry(fullpath filer2.FullPath) (entry *filer2.Entry, err error) {
+
+	data, err := store.Client.Get(string(fullpath)).Result()
+	if err == redis.Nil {
+		return nil, filer2.ErrNotFound
+	}
+
+	if err != nil {
+		return nil, fmt.Errorf("get %s : %v", entry.FullPath, err)
+	}
+
+	entry = &filer2.Entry{
+		FullPath: fullpath,
+	}
+	err = entry.DecodeAttributesAndChunks([]byte(data))
+	if err != nil {
+		return entry, fmt.Errorf("decode %s : %v", entry.FullPath, err)
+	}
+
+	return entry, nil
+}
+
+func (store *RedisStore) DeleteEntry(fullpath filer2.FullPath) (err error) {
+
+	_, err = store.Client.Del(string(fullpath)).Result()
+
+	if err != nil {
+		return fmt.Errorf("delete %s : %v", fullpath, err)
+	}
+
+	dir, name := fullpath.DirAndName()
+	if name != "" {
+		_, err = store.Client.SRem(genDirectoryListKey(dir), name).Result()
+		if err != nil {
+			return fmt.Errorf("delete %s in parent dir: %v", fullpath, err)
+		}
+	}
+
+	return nil
+}
+
+func (store *RedisStore) ListDirectoryEntries(fullpath filer2.FullPath, startFileName string, inclusive bool,
+	limit int) (entries []*filer2.Entry, err error) {
+
+	members, err := store.Client.SMembers(genDirectoryListKey(string(fullpath))).Result()
+	if err != nil {
+		return nil, fmt.Errorf("list %s : %v", fullpath, err)
+	}
+
+	// skip
+	if startFileName != "" {
+		var t []string
+		for _, m := range members {
+			if strings.Compare(m, startFileName) >= 0 {
+				if m == startFileName {
+					if inclusive {
+						t = append(t, m)
+					}
+				} else {
+					t = append(t, m)
+				}
+			}
+		}
+		members = t
+	}
+
+	// sort
+	sort.Slice(members, func(i, j int) bool {
+		return strings.Compare(members[i], members[j]) < 0
+	})
+
+	// limit
+	if limit < len(members) {
+		members = members[:limit]
+	}
+
+	// fetch entry meta
+	for _, fileName := range members {
+		path := filer2.NewFullPath(string(fullpath), fileName)
+		entry, err := store.FindEntry(path)
+		if err != nil {
+			glog.V(0).Infof("list %s : %v", path, err)
+		} else {
+			entries = append(entries, entry)
+		}
+	}
+
+	return entries, err
+}
+
+func genDirectoryListKey(dir string) (dirList string) {
+	return dir + DIR_LIST_MARKER
+}

weed/filesys/dir.go

@@ -7,11 +7,12 @@ import (
 	"path"
 	"sync"
 
-	"bazil.org/fuse/fs"
 	"bazil.org/fuse"
-	"github.com/chrislusf/seaweedfs/weed/filer"
+	"bazil.org/fuse/fs"
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"path/filepath"
+	"time"
 )
 
 type Dir struct {
@@ -21,21 +22,156 @@ type Dir struct {
 	wfs         *WFS
 }
 
+var _ = fs.Node(&Dir{})
+var _ = fs.NodeCreater(&Dir{})
+var _ = fs.NodeMkdirer(&Dir{})
+var _ = fs.NodeStringLookuper(&Dir{})
+var _ = fs.HandleReadDirAller(&Dir{})
+var _ = fs.NodeRemover(&Dir{})
+
 func (dir *Dir) Attr(context context.Context, attr *fuse.Attr) error {
+
+	if dir.Path == "/" {
+		attr.Valid = time.Second
 		attr.Mode = os.ModeDir | 0777
 		return nil
+	}
+
+	parent, name := filepath.Split(dir.Path)
+
+	var attributes *filer_pb.FuseAttributes
+
+	err := dir.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+		request := &filer_pb.GetEntryAttributesRequest{
+			Name:      name,
+			ParentDir: parent,
+		}
+
+		glog.V(1).Infof("read dir attr: %v", request)
+		resp, err := client.GetEntryAttributes(context, request)
+		if err != nil {
+			glog.V(0).Infof("read dir attr %v: %v", request, err)
+			return err
+		}
+
+		attributes = resp.Attributes
+
+		return nil
+	})
+
+	if err != nil {
+		return err
+	}
+
+	// glog.V(1).Infof("dir %s: %v", dir.Path, attributes)
+	// glog.V(1).Infof("dir %s permission: %v", dir.Path, os.FileMode(attributes.FileMode))
+
+	attr.Mode = os.FileMode(attributes.FileMode) | os.ModeDir
+	if dir.Path == "/" && attributes.FileMode == 0 {
+		attr.Valid = time.Second
+	}
+
+	attr.Mtime = time.Unix(attributes.Mtime, 0)
+	attr.Ctime = time.Unix(attributes.Mtime, 0)
+	attr.Gid = attributes.Gid
+	attr.Uid = attributes.Uid
+
+	return nil
+}
+
+func (dir *Dir) newFile(name string, chunks []*filer_pb.FileChunk) *File {
+	return &File{
+		Name: name,
+		dir:  dir,
+		wfs:  dir.wfs,
+		// attributes: &filer_pb.FuseAttributes{},
+		Chunks: chunks,
+	}
+}
+
+func (dir *Dir) Create(ctx context.Context, req *fuse.CreateRequest,
+	resp *fuse.CreateResponse) (fs.Node, fs.Handle, error) {
+
+	err := dir.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+		request := &filer_pb.CreateEntryRequest{
+			Directory: dir.Path,
+			Entry: &filer_pb.Entry{
+				Name:        req.Name,
+				IsDirectory: req.Mode&os.ModeDir > 0,
+				Attributes: &filer_pb.FuseAttributes{
+					Mtime:    time.Now().Unix(),
+					Crtime:   time.Now().Unix(),
+					FileMode: uint32(req.Mode),
+					Uid:      req.Uid,
+					Gid:      req.Gid,
+				},
+			},
+		}
+
+		glog.V(1).Infof("create: %v", request)
+		if _, err := client.CreateEntry(ctx, request); err != nil {
+			return fmt.Errorf("create file: %v", err)
+		}
+
+		return nil
+	})
+
+	if err == nil {
+		file := dir.newFile(req.Name, nil)
+		dir.NodeMap[req.Name] = file
+		file.isOpen = true
+		return file, &FileHandle{
+			f:          file,
+			dirtyPages: newDirtyPages(file),
+			RequestId:  req.Header.ID,
+			NodeId:     req.Header.Node,
+			Uid:        req.Uid,
+			Gid:        req.Gid,
+		}, nil
+	}
+
+	return nil, nil, err
 }
 
 func (dir *Dir) Mkdir(ctx context.Context, req *fuse.MkdirRequest) (fs.Node, error) {
 	dir.NodeMapLock.Lock()
 	defer dir.NodeMapLock.Unlock()
 
-	fmt.Printf("mkdir %+v\n", req)
+	err := dir.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+		request := &filer_pb.CreateEntryRequest{
+			Directory: dir.Path,
+			Entry: &filer_pb.Entry{
+				Name:        req.Name,
+				IsDirectory: true,
+				Attributes: &filer_pb.FuseAttributes{
+					Mtime:    time.Now().Unix(),
+					Crtime:   time.Now().Unix(),
+					FileMode: uint32(req.Mode),
+					Uid:      req.Uid,
+					Gid:      req.Gid,
+				},
+			},
+		}
+
+		glog.V(1).Infof("mkdir: %v", request)
+		if _, err := client.CreateEntry(ctx, request); err != nil {
+			glog.V(0).Infof("mkdir %v: %v", request, err)
+			return fmt.Errorf("make dir: %v", err)
+		}
 
+		return nil
+	})
+
+	if err == nil {
 		node := &Dir{Path: path.Join(dir.Path, req.Name), wfs: dir.wfs}
 		dir.NodeMap[req.Name] = node
-
 		return node, nil
+	}
+
+	return nil, err
 }
 
 func (dir *Dir) Lookup(ctx context.Context, name string) (node fs.Node, err error) {
@@ -59,7 +195,7 @@ func (dir *Dir) Lookup(ctx context.Context, name string) (node fs.Node, err erro
 			Name:      name,
 		}
 
-		glog.V(1).Infof("lookup directory entry: %v", request)
+		glog.V(4).Infof("lookup directory entry: %v", request)
 		resp, err := client.LookupDirectoryEntry(ctx, request)
 		if err != nil {
 			return err
@@ -74,13 +210,13 @@ func (dir *Dir) Lookup(ctx context.Context, name string) (node fs.Node, err erro
 		if entry.IsDirectory {
 			node = &Dir{Path: path.Join(dir.Path, name), wfs: dir.wfs}
 		} else {
-			node = &File{FileId: filer.FileId(entry.FileId), Name: name, wfs: dir.wfs}
+			node = dir.newFile(name, entry.Chunks)
 		}
 		dir.NodeMap[name] = node
 		return node, nil
 	}
 
-	return nil, err
+	return nil, fuse.ENOENT
 }
 
 func (dir *Dir) ReadDirAll(ctx context.Context) (ret []fuse.Dirent, err error) {
@@ -91,7 +227,7 @@ func (dir *Dir) ReadDirAll(ctx context.Context) (ret []fuse.Dirent, err error) {
 			Directory: dir.Path,
 		}
 
-		glog.V(1).Infof("read directory: %v", request)
+		glog.V(4).Infof("read directory: %v", request)
 		resp, err := client.ListEntries(ctx, request)
 		if err != nil {
 			return err
@@ -104,6 +240,7 @@ func (dir *Dir) ReadDirAll(ctx context.Context) (ret []fuse.Dirent, err error) {
 			} else {
 				dirent := fuse.Dirent{Name: entry.Name, Type: fuse.DT_File}
 				ret = append(ret, dirent)
+				dir.wfs.listDirectoryEntriesCache.Set(dir.Path+"/"+entry.Name, entry, 300*time.Millisecond)
 			}
 		}
 

weed/filesys/dirty_page.go

@@ -0,0 +1,165 @@
+package filesys
+
+import (
+	"fmt"
+	"bytes"
+	"time"
+	"context"
+
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"github.com/chrislusf/seaweedfs/weed/operation"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+)
+
+type ContinuousDirtyPages struct {
+	hasData bool
+	Offset  int64
+	Size    int64
+	Data    []byte
+	f       *File
+}
+
+func newDirtyPages(file *File) *ContinuousDirtyPages {
+	return &ContinuousDirtyPages{
+		Data: make([]byte, file.wfs.chunkSizeLimit),
+		f:    file,
+	}
+}
+
+func (pages *ContinuousDirtyPages) AddPage(ctx context.Context, offset int64, data []byte) (chunks []*filer_pb.FileChunk, err error) {
+
+	var chunk *filer_pb.FileChunk
+
+	if len(data) > len(pages.Data) {
+		// this is more than what buffer can hold.
+
+		// flush existing
+		if chunk, err = pages.saveExistingPagesToStorage(ctx); err == nil {
+			if chunk != nil {
+				glog.V(4).Infof("%s/%s flush existing [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
+			}
+			chunks = append(chunks, chunk)
+		} else {
+			glog.V(0).Infof("%s/%s failed to flush1 [%d,%d): %v", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size), err)
+			return
+		}
+		pages.Size = 0
+
+		// flush the big page
+		if chunk, err = pages.saveToStorage(ctx, data, offset); err == nil {
+			if chunk != nil {
+				glog.V(4).Infof("%s/%s flush big request [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
+				chunks = append(chunks, chunk)
+			}
+		} else {
+			glog.V(0).Infof("%s/%s failed to flush2 [%d,%d): %v", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size), err)
+			return
+		}
+
+		return
+	}
+
+	if offset < pages.Offset || offset >= pages.Offset+int64(len(pages.Data)) ||
+		pages.Offset+int64(len(pages.Data)) < offset+int64(len(data)) {
+		// if the data is out of range,
+		// or buffer is full if adding new data,
+		// flush current buffer and add new data
+
+		// println("offset", offset, "size", len(data), "existing offset", pages.Offset, "size", pages.Size)
+
+		if chunk, err = pages.saveExistingPagesToStorage(ctx); err == nil {
+			if chunk != nil {
+				glog.V(4).Infof("%s/%s add save [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
+				chunks = append(chunks, chunk)
+			}
+		} else {
+			glog.V(0).Infof("%s/%s add save [%d,%d): %v", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size), err)
+			return
+		}
+		pages.Offset = offset
+		pages.Size = int64(len(data))
+		copy(pages.Data, data)
+		return
+	}
+
+	copy(pages.Data[offset-pages.Offset:], data)
+	pages.Size = max(pages.Size, offset+int64(len(data))-pages.Offset)
+
+	return
+}
+
+func (pages *ContinuousDirtyPages) FlushToStorage(ctx context.Context) (chunk *filer_pb.FileChunk, err error) {
+
+	if pages.Size == 0 {
+		return nil, nil
+	}
+
+	if chunk, err = pages.saveExistingPagesToStorage(ctx); err == nil {
+		pages.Size = 0
+		if chunk != nil {
+			glog.V(4).Infof("%s/%s flush [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
+		}
+	}
+	return
+}
+
+func (pages *ContinuousDirtyPages) saveExistingPagesToStorage(ctx context.Context) (*filer_pb.FileChunk, error) {
+	return pages.saveToStorage(ctx, pages.Data[:pages.Size], pages.Offset)
+}
+
+func (pages *ContinuousDirtyPages) saveToStorage(ctx context.Context, buf []byte, offset int64) (*filer_pb.FileChunk, error) {
+
+	if pages.Size == 0 {
+		return nil, nil
+	}
+
+	var fileId, host string
+
+	if err := pages.f.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+		request := &filer_pb.AssignVolumeRequest{
+			Count:       1,
+			Replication: pages.f.wfs.replication,
+			Collection:  pages.f.wfs.collection,
+		}
+
+		resp, err := client.AssignVolume(ctx, request)
+		if err != nil {
+			glog.V(0).Infof("assign volume failure %v: %v", request, err)
+			return err
+		}
+
+		fileId, host = resp.FileId, resp.Url
+
+		return nil
+	}); err != nil {
+		return nil, fmt.Errorf("filer assign volume: %v", err)
+	}
+
+	fileUrl := fmt.Sprintf("http://%s/%s", host, fileId)
+	bufReader := bytes.NewReader(pages.Data[:pages.Size])
+	uploadResult, err := operation.Upload(fileUrl, pages.f.Name, bufReader, false, "application/octet-stream", nil, "")
+	if err != nil {
+		glog.V(0).Infof("upload data %v to %s: %v", pages.f.Name, fileUrl, err)
+		return nil, fmt.Errorf("upload data: %v", err)
+	}
+	if uploadResult.Error != "" {
+		glog.V(0).Infof("upload failure %v to %s: %v", pages.f.Name, fileUrl, err)
+		return nil, fmt.Errorf("upload result: %v", uploadResult.Error)
+	}
+
+	return &filer_pb.FileChunk{
+		FileId: fileId,
+		Offset: offset,
+		Size:   uint64(len(buf)),
+		Mtime:  time.Now().UnixNano(),
+	}, nil
+
+}
+
+func max(x, y int64) int64 {
+	if x > y {
+		return x
+	}
+	return y
+}

weed/filesys/file.go

@@ -1,75 +1,136 @@
 package filesys
 
 import (
-	"context"
-	"fmt"
-
 	"bazil.org/fuse"
-	"github.com/chrislusf/seaweedfs/weed/filer"
 	"bazil.org/fuse/fs"
+	"context"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"os"
+	"path/filepath"
+	"time"
 )
 
 var _ = fs.Node(&File{})
-// var _ = fs.NodeOpener(&File{})
-// var _ = fs.NodeFsyncer(&File{})
-var _ = fs.Handle(&File{})
-var _ = fs.HandleReadAller(&File{})
-// var _ = fs.HandleReader(&File{})
-var _ = fs.HandleWriter(&File{})
+var _ = fs.NodeOpener(&File{})
+var _ = fs.NodeFsyncer(&File{})
+var _ = fs.NodeSetattrer(&File{})
 
 type File struct {
-	FileId filer.FileId
+	Chunks     []*filer_pb.FileChunk
 	Name       string
+	dir        *Dir
 	wfs        *WFS
+	attributes *filer_pb.FuseAttributes
+	isOpen     bool
 }
 
 func (file *File) Attr(context context.Context, attr *fuse.Attr) error {
-	attr.Mode = 0444
-	return file.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
 
-		request := &filer_pb.GetFileAttributesRequest{
+	fullPath := filepath.Join(file.dir.Path, file.Name)
+
+	if file.attributes == nil || !file.isOpen {
+		item := file.wfs.listDirectoryEntriesCache.Get(fullPath)
+		if item != nil {
+			entry := item.Value().(*filer_pb.Entry)
+			file.Chunks = entry.Chunks
+			file.attributes = entry.Attributes
+			glog.V(1).Infof("file attr read cached %v attributes", file.Name)
+		} else {
+			err := file.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+				request := &filer_pb.GetEntryAttributesRequest{
 					Name:      file.Name,
-			ParentDir: "", //TODO add parent folder
-			FileId:    string(file.FileId),
+					ParentDir: file.dir.Path,
 				}
 
-		glog.V(1).Infof("read file size: %v", request)
-		resp, err := client.GetFileAttributes(context, request)
+				resp, err := client.GetEntryAttributes(context, request)
 				if err != nil {
+					glog.V(0).Infof("file attr read file %v: %v", request, err)
 					return err
 				}
 
-		attr.Size = resp.Attributes.FileSize
+				file.attributes = resp.Attributes
+				file.Chunks = resp.Chunks
+
+				glog.V(1).Infof("file attr %v %+v: %d", fullPath, file.attributes, filer2.TotalSize(file.Chunks))
 
 				return nil
 			})
+
+			if err != nil {
+				return err
+			}
+		}
+	}
+
+	attr.Mode = os.FileMode(file.attributes.FileMode)
+	attr.Size = filer2.TotalSize(file.Chunks)
+	attr.Mtime = time.Unix(file.attributes.Mtime, 0)
+	attr.Gid = file.attributes.Gid
+	attr.Uid = file.attributes.Uid
+
+	return nil
+
 }
 
-func (file *File) ReadAll(ctx context.Context) (content []byte, err error) {
+func (file *File) Open(ctx context.Context, req *fuse.OpenRequest, resp *fuse.OpenResponse) (fs.Handle, error) {
 
-	err = file.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+	fullPath := filepath.Join(file.dir.Path, file.Name)
 
-		request := &filer_pb.GetFileContentRequest{
-			FileId: string(file.FileId),
+	glog.V(3).Infof("%v file open %+v", fullPath, req)
+
+	file.isOpen = true
+
+	return &FileHandle{
+		f:          file,
+		dirtyPages: newDirtyPages(file),
+		RequestId:  req.Header.ID,
+		NodeId:     req.Header.Node,
+		Uid:        req.Uid,
+		Gid:        req.Gid,
+	}, nil
+
+}
+
+func (file *File) Setattr(ctx context.Context, req *fuse.SetattrRequest, resp *fuse.SetattrResponse) error {
+	fullPath := filepath.Join(file.dir.Path, file.Name)
+
+	glog.V(3).Infof("%v file setattr %+v", fullPath, req)
+	if req.Valid.Size() {
+
+		glog.V(3).Infof("%v file setattr set size=%v", fullPath, req.Size)
+		if req.Size == 0 {
+			// fmt.Printf("truncate %v \n", fullPath)
+			file.Chunks = nil
+		}
+		file.attributes.FileSize = req.Size
+	}
+	if req.Valid.Mode() {
+		file.attributes.FileMode = uint32(req.Mode)
 	}
 
-		glog.V(1).Infof("read file content: %v", request)
-		resp, err := client.GetFileContent(ctx, request)
-		if err != nil {
-			return err
+	if req.Valid.Uid() {
+		file.attributes.Uid = req.Uid
+	}
+
+	if req.Valid.Gid() {
+		file.attributes.Gid = req.Gid
 	}
 
-		content = resp.Content
+	if req.Valid.Mtime() {
+		file.attributes.Mtime = req.Mtime.Unix()
+	}
 
 	return nil
-	})
 
-	return content, err
 }
 
-func (file *File) Write(ctx context.Context, req *fuse.WriteRequest, resp *fuse.WriteResponse) error {
-	fmt.Printf("write file %+v\n", req)
+func (file *File) Fsync(ctx context.Context, req *fuse.FsyncRequest) error {
+	// fsync works at OS level
+	// write the file chunks to the filer
+	glog.V(3).Infof("%s/%s fsync file %+v", file.dir.Path, file.Name, req)
+
 	return nil
 }

weed/filesys/filehandle.go

@@ -0,0 +1,219 @@
+package filesys
+
+import (
+	"bazil.org/fuse"
+	"bazil.org/fuse/fs"
+	"context"
+	"fmt"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"github.com/chrislusf/seaweedfs/weed/util"
+	"strings"
+	"sync"
+	"net/http"
+)
+
+type FileHandle struct {
+	// cache file has been written to
+	dirtyPages    *ContinuousDirtyPages
+	dirtyMetadata bool
+
+	f         *File
+	RequestId fuse.RequestID // unique ID for request
+	NodeId    fuse.NodeID    // file or directory the request is about
+	Uid       uint32         // user ID of process making request
+	Gid       uint32         // group ID of process making request
+}
+
+var _ = fs.Handle(&FileHandle{})
+
+// var _ = fs.HandleReadAller(&FileHandle{})
+var _ = fs.HandleReader(&FileHandle{})
+var _ = fs.HandleFlusher(&FileHandle{})
+var _ = fs.HandleWriter(&FileHandle{})
+var _ = fs.HandleReleaser(&FileHandle{})
+
+func (fh *FileHandle) Read(ctx context.Context, req *fuse.ReadRequest, resp *fuse.ReadResponse) error {
+
+	glog.V(4).Infof("%v/%v read fh: [%d,%d)", fh.f.dir.Path, fh.f.Name, req.Offset, req.Offset+int64(req.Size))
+
+	if len(fh.f.Chunks) == 0 {
+		glog.V(0).Infof("empty fh %v/%v", fh.f.dir.Path, fh.f.Name)
+		return fmt.Errorf("empty file %v/%v", fh.f.dir.Path, fh.f.Name)
+	}
+
+	buff := make([]byte, req.Size)
+
+	chunkViews := filer2.ViewFromChunks(fh.f.Chunks, req.Offset, req.Size)
+
+	var vids []string
+	for _, chunkView := range chunkViews {
+		vids = append(vids, volumeId(chunkView.FileId))
+	}
+
+	vid2Locations := make(map[string]*filer_pb.Locations)
+
+	err := fh.f.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+		glog.V(4).Infof("read fh lookup volume id locations: %v", vids)
+		resp, err := client.LookupVolume(ctx, &filer_pb.LookupVolumeRequest{
+			VolumeIds: vids,
+		})
+		if err != nil {
+			return err
+		}
+
+		vid2Locations = resp.LocationsMap
+
+		return nil
+	})
+
+	if err != nil {
+		glog.V(4).Infof("%v/%v read fh lookup volume ids: %v", fh.f.dir.Path, fh.f.Name, err)
+		return fmt.Errorf("failed to lookup volume ids %v: %v", vids, err)
+	}
+
+	var totalRead int64
+	var wg sync.WaitGroup
+	for _, chunkView := range chunkViews {
+		wg.Add(1)
+		go func(chunkView *filer2.ChunkView) {
+			defer wg.Done()
+
+			glog.V(4).Infof("read fh reading chunk: %+v", chunkView)
+
+			locations := vid2Locations[volumeId(chunkView.FileId)]
+			if locations == nil || len(locations.Locations) == 0 {
+				glog.V(0).Infof("failed to locate %s", chunkView.FileId)
+				err = fmt.Errorf("failed to locate %s", chunkView.FileId)
+				return
+			}
+
+			var n int64
+			n, err = util.ReadUrl(
+				fmt.Sprintf("http://%s/%s", locations.Locations[0].Url, chunkView.FileId),
+				chunkView.Offset,
+				int(chunkView.Size),
+				buff[chunkView.LogicOffset-req.Offset:chunkView.LogicOffset-req.Offset+int64(chunkView.Size)])
+
+			if err != nil {
+
+				glog.V(0).Infof("%v/%v read http://%s/%v %v bytes: %v", fh.f.dir.Path, fh.f.Name, locations.Locations[0].Url, chunkView.FileId, n, err)
+
+				err = fmt.Errorf("failed to read http://%s/%s: %v",
+					locations.Locations[0].Url, chunkView.FileId, err)
+				return
+			}
+
+			glog.V(4).Infof("read fh read %d bytes: %+v", n, chunkView)
+			totalRead += n
+
+		}(chunkView)
+	}
+	wg.Wait()
+
+	resp.Data = buff[:totalRead]
+
+	return err
+}
+
+// Write to the file handle
+func (fh *FileHandle) Write(ctx context.Context, req *fuse.WriteRequest, resp *fuse.WriteResponse) error {
+
+	// write the request to volume servers
+
+	glog.V(4).Infof("%+v/%v write fh: [%d,%d)", fh.f.dir.Path, fh.f.Name, req.Offset, req.Offset+int64(len(req.Data)))
+
+	chunks, err := fh.dirtyPages.AddPage(ctx, req.Offset, req.Data)
+	if err != nil {
+		return fmt.Errorf("write %s/%s at [%d,%d): %v", fh.f.dir.Path, fh.f.Name, req.Offset, req.Offset+int64(len(req.Data)), err)
+	}
+
+	resp.Size = len(req.Data)
+
+	if req.Offset == 0 {
+		fh.f.attributes.Mime = http.DetectContentType(req.Data)
+		fh.dirtyMetadata = true
+	}
+
+	for _, chunk := range chunks {
+		fh.f.Chunks = append(fh.f.Chunks, chunk)
+		glog.V(1).Infof("uploaded %s/%s to %s [%d,%d)", fh.f.dir.Path, fh.f.Name, chunk.FileId, chunk.Offset, chunk.Offset+int64(chunk.Size))
+		fh.dirtyMetadata = true
+	}
+
+	return nil
+}
+
+func (fh *FileHandle) Release(ctx context.Context, req *fuse.ReleaseRequest) error {
+
+	glog.V(4).Infof("%+v/%v release fh", fh.f.dir.Path, fh.f.Name)
+
+	fh.f.isOpen = false
+
+	return nil
+}
+
+// Flush - experimenting with uploading at flush, this slows operations down till it has been
+// completely flushed
+func (fh *FileHandle) Flush(ctx context.Context, req *fuse.FlushRequest) error {
+	// fflush works at fh level
+	// send the data to the OS
+	glog.V(4).Infof("%s/%s fh flush %v", fh.f.dir.Path, fh.f.Name, req)
+
+	chunk, err := fh.dirtyPages.FlushToStorage(ctx)
+	if err != nil {
+		glog.V(0).Infof("flush %s/%s to %s [%d,%d): %v", fh.f.dir.Path, fh.f.Name, chunk.FileId, chunk.Offset, chunk.Offset+int64(chunk.Size), err)
+		return fmt.Errorf("flush %s/%s to %s [%d,%d): %v", fh.f.dir.Path, fh.f.Name, chunk.FileId, chunk.Offset, chunk.Offset+int64(chunk.Size), err)
+	}
+	if chunk != nil {
+		fh.f.Chunks = append(fh.f.Chunks, chunk)
+		fh.dirtyMetadata = true
+	}
+
+	if !fh.dirtyMetadata {
+		return nil
+	}
+
+	if len(fh.f.Chunks) == 0 {
+		glog.V(2).Infof("fh %s/%s flush skipping empty: %v", fh.f.dir.Path, fh.f.Name, req)
+		return nil
+	}
+
+	err = fh.f.wfs.withFilerClient(func(client filer_pb.SeaweedFilerClient) error {
+
+		request := &filer_pb.UpdateEntryRequest{
+			Directory: fh.f.dir.Path,
+			Entry: &filer_pb.Entry{
+				Name:       fh.f.Name,
+				Attributes: fh.f.attributes,
+				Chunks:     fh.f.Chunks,
+			},
+		}
+
+		glog.V(1).Infof("%s/%s set chunks: %v", fh.f.dir.Path, fh.f.Name, len(fh.f.Chunks))
+		for i, chunk := range fh.f.Chunks {
+			glog.V(1).Infof("%s/%s chunks %d: %v [%d,%d)", fh.f.dir.Path, fh.f.Name, i, chunk.FileId, chunk.Offset, chunk.Offset+int64(chunk.Size))
+		}
+		if _, err := client.UpdateEntry(ctx, request); err != nil {
+			return fmt.Errorf("update fh: %v", err)
+		}
+
+		return nil
+	})
+
+	if err == nil {
+		fh.dirtyMetadata = false
+	}
+
+	return err
+}
+
+func volumeId(fileId string) string {
+	lastCommaIndex := strings.LastIndex(fileId, ",")
+	if lastCommaIndex > 0 {
+		return fileId[:lastCommaIndex]
+	}
+	return fileId
+}

weed/filesys/wfs.go

@@ -3,17 +3,26 @@ package filesys
 import (
 	"bazil.org/fuse/fs"
 	"fmt"
-	"google.golang.org/grpc"
 	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+	"github.com/karlseguin/ccache"
+	"google.golang.org/grpc"
 )
 
 type WFS struct {
 	filer                     string
+	listDirectoryEntriesCache *ccache.Cache
+	collection                string
+	replication               string
+	chunkSizeLimit            int64
 }
 
-func NewSeaweedFileSystem(filer string) *WFS {
+func NewSeaweedFileSystem(filer string, collection string, replication string, chunkSizeLimitMB int) *WFS {
 	return &WFS{
 		filer:                     filer,
+		listDirectoryEntriesCache: ccache.New(ccache.Configure().MaxSize(6000).ItemsToPrune(100)),
+		collection:                collection,
+		replication:               replication,
+		chunkSizeLimit:            int64(chunkSizeLimitMB) * 1024 * 1024,
 	}
 }
 

weed/images/favicon.go

@@ -181,6 +181,7 @@ type bintree struct {
 	Func     func() (*asset, error)
 	Children map[string]*bintree
 }
+
 var _bintree = &bintree{nil, map[string]*bintree{
 	"favicon": &bintree{nil, map[string]*bintree{
 		"favicon.ico": &bintree{favicon, map[string]*bintree{}},
@@ -233,4 +234,3 @@ func _filePath(dir, name string) string {
 	cannonicalName := strings.Replace(name, "\\", "/", -1)
 	return filepath.Join(append([]string{dir}, strings.Split(cannonicalName, "/")...)...)
 }
-

weed/operation/assign_file_id.go

@@ -52,7 +52,7 @@ func Assign(server string, r *VolumeAssignRequest) (*AssignResult, error) {
 	}
 
 	jsonBlob, err := util.Post("http://"+server+"/dir/assign", values)
-	glog.V(2).Info("assign result :", string(jsonBlob))
+	glog.V(2).Infof("assign result from %s : %s", server, string(jsonBlob))
 	if err != nil {
 		return nil, err
 	}

weed/operation/filer/register.go

@@ -1,31 +0,0 @@
-package filer
-
-import (
-	"fmt"
-	"net/url"
-
-	"github.com/chrislusf/seaweedfs/weed/security"
-	"github.com/chrislusf/seaweedfs/weed/util"
-)
-
-type SubmitResult struct {
-	FileName string `json:"fileName,omitempty"`
-	FileUrl  string `json:"fileUrl,omitempty"`
-	Fid      string `json:"fid,omitempty"`
-	Size     uint32 `json:"size,omitempty"`
-	Error    string `json:"error,omitempty"`
-}
-
-func RegisterFile(filer string, path string, fileId string, secret security.Secret) error {
-	// TODO: jwt need to be used
-	_ = security.GenJwt(secret, fileId)
-
-	values := make(url.Values)
-	values.Add("path", path)
-	values.Add("fileId", fileId)
-	_, err := util.Post("http://"+filer+"/admin/register", values)
-	if err != nil {
-		return fmt.Errorf("Failed to register path %s on filer %s to file id %s : %v", path, filer, fileId, err)
-	}
-	return nil
-}

weed/pb/filer.proto

@@ -12,15 +12,24 @@ service SeaweedFiler {
     rpc ListEntries (ListEntriesRequest) returns (ListEntriesResponse) {
     }
 
-    rpc GetFileAttributes (GetFileAttributesRequest) returns (GetFileAttributesResponse) {
+    rpc GetEntryAttributes (GetEntryAttributesRequest) returns (GetEntryAttributesResponse) {
     }
 
-    rpc GetFileContent (GetFileContentRequest) returns (GetFileContentResponse) {
+    rpc CreateEntry (CreateEntryRequest) returns (CreateEntryResponse) {
+    }
+
+    rpc UpdateEntry (UpdateEntryRequest) returns (UpdateEntryResponse) {
     }
 
     rpc DeleteEntry (DeleteEntryRequest) returns (DeleteEntryResponse) {
     }
 
+    rpc AssignVolume (AssignVolumeRequest) returns (AssignVolumeResponse) {
+    }
+
+    rpc LookupVolume (LookupVolumeRequest) returns (LookupVolumeResponse) {
+    }
+
 }
 
 //////////////////////////////////////////////////
@@ -45,26 +54,36 @@ message ListEntriesResponse {
 message Entry {
     string name = 1;
     bool is_directory = 2;
-    string file_id = 3;
+    repeated FileChunk chunks = 3;
     FuseAttributes attributes = 4;
 }
 
+message FileChunk {
+    string file_id = 1;
+    int64 offset = 2;
+    uint64 size = 3;
+    int64 mtime = 4;
+}
+
 message FuseAttributes {
     uint64 file_size = 1;
     int64 mtime = 2;
     uint32 file_mode = 3;
     uint32 uid = 4;
     uint32 gid = 5;
+    int64 crtime = 6;
+    string mime = 7;
 }
 
-message GetFileAttributesRequest {
+message GetEntryAttributesRequest {
     string name = 1;
     string parent_dir = 2;
     string file_id = 3;
 }
 
-message GetFileAttributesResponse {
+message GetEntryAttributesResponse {
     FuseAttributes attributes = 1;
+    repeated FileChunk chunks = 2;
 }
 
 message GetFileContentRequest {
@@ -75,6 +94,21 @@ message GetFileContentResponse {
     bytes content = 1;
 }
 
+message CreateEntryRequest {
+    string directory = 1;
+    Entry entry = 2;
+}
+
+message CreateEntryResponse {
+}
+
+message UpdateEntryRequest {
+    string directory = 1;
+    Entry entry = 2;
+}
+message UpdateEntryResponse {
+}
+
 message DeleteEntryRequest {
     string directory = 1;
     string name = 2;
@@ -83,3 +117,32 @@ message DeleteEntryRequest {
 
 message DeleteEntryResponse {
 }
+
+message AssignVolumeRequest {
+    int32 count = 1;
+    string collection = 2;
+    string replication = 3;
+}
+
+message AssignVolumeResponse {
+    string file_id = 1;
+    string url = 2;
+    string public_url = 3;
+    int32 count = 4;
+}
+
+message LookupVolumeRequest {
+    repeated string volume_ids = 1;
+}
+
+message Locations {
+    repeated Location locations = 1;
+}
+
+message Location {
+    string url = 1;
+    string public_url = 2;
+}
+message LookupVolumeResponse {
+    map<string, Locations> locations_map = 1;
+}

weed/pb/filer_pb/filer.pb.go

@@ -14,13 +14,24 @@ It has these top-level messages:
 	ListEntriesRequest
 	ListEntriesResponse
 	Entry
+	FileChunk
 	FuseAttributes
-	GetFileAttributesRequest
-	GetFileAttributesResponse
+	GetEntryAttributesRequest
+	GetEntryAttributesResponse
 	GetFileContentRequest
 	GetFileContentResponse
+	CreateEntryRequest
+	CreateEntryResponse
+	UpdateEntryRequest
+	UpdateEntryResponse
 	DeleteEntryRequest
 	DeleteEntryResponse
+	AssignVolumeRequest
+	AssignVolumeResponse
+	LookupVolumeRequest
+	Locations
+	Location
+	LookupVolumeResponse
 */
 package filer_pb
 
@@ -119,7 +130,7 @@ func (m *ListEntriesResponse) GetEntries() []*Entry {
 type Entry struct {
 	Name        string          `protobuf:"bytes,1,opt,name=name" json:"name,omitempty"`
 	IsDirectory bool            `protobuf:"varint,2,opt,name=is_directory,json=isDirectory" json:"is_directory,omitempty"`
-	FileId      string          `protobuf:"bytes,3,opt,name=file_id,json=fileId" json:"file_id,omitempty"`
+	Chunks      []*FileChunk    `protobuf:"bytes,3,rep,name=chunks" json:"chunks,omitempty"`
 	Attributes  *FuseAttributes `protobuf:"bytes,4,opt,name=attributes" json:"attributes,omitempty"`
 }
 
@@ -142,11 +153,11 @@ func (m *Entry) GetIsDirectory() bool {
 	return false
 }
 
-func (m *Entry) GetFileId() string {
+func (m *Entry) GetChunks() []*FileChunk {
 	if m != nil {
-		return m.FileId
+		return m.Chunks
 	}
-	return ""
+	return nil
 }
 
 func (m *Entry) GetAttributes() *FuseAttributes {
@@ -156,18 +167,60 @@ func (m *Entry) GetAttributes() *FuseAttributes {
 	return nil
 }
 
+type FileChunk struct {
+	FileId string `protobuf:"bytes,1,opt,name=file_id,json=fileId" json:"file_id,omitempty"`
+	Offset int64  `protobuf:"varint,2,opt,name=offset" json:"offset,omitempty"`
+	Size   uint64 `protobuf:"varint,3,opt,name=size" json:"size,omitempty"`
+	Mtime  int64  `protobuf:"varint,4,opt,name=mtime" json:"mtime,omitempty"`
+}
+
+func (m *FileChunk) Reset()                    { *m = FileChunk{} }
+func (m *FileChunk) String() string            { return proto.CompactTextString(m) }
+func (*FileChunk) ProtoMessage()               {}
+func (*FileChunk) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{5} }
+
+func (m *FileChunk) GetFileId() string {
+	if m != nil {
+		return m.FileId
+	}
+	return ""
+}
+
+func (m *FileChunk) GetOffset() int64 {
+	if m != nil {
+		return m.Offset
+	}
+	return 0
+}
+
+func (m *FileChunk) GetSize() uint64 {
+	if m != nil {
+		return m.Size
+	}
+	return 0
+}
+
+func (m *FileChunk) GetMtime() int64 {
+	if m != nil {
+		return m.Mtime
+	}
+	return 0
+}
+
 type FuseAttributes struct {
 	FileSize uint64 `protobuf:"varint,1,opt,name=file_size,json=fileSize" json:"file_size,omitempty"`
 	Mtime    int64  `protobuf:"varint,2,opt,name=mtime" json:"mtime,omitempty"`
 	FileMode uint32 `protobuf:"varint,3,opt,name=file_mode,json=fileMode" json:"file_mode,omitempty"`
 	Uid      uint32 `protobuf:"varint,4,opt,name=uid" json:"uid,omitempty"`
 	Gid      uint32 `protobuf:"varint,5,opt,name=gid" json:"gid,omitempty"`
+	Crtime   int64  `protobuf:"varint,6,opt,name=crtime" json:"crtime,omitempty"`
+	Mime     string `protobuf:"bytes,7,opt,name=mime" json:"mime,omitempty"`
 }
 
 func (m *FuseAttributes) Reset()                    { *m = FuseAttributes{} }
 func (m *FuseAttributes) String() string            { return proto.CompactTextString(m) }
 func (*FuseAttributes) ProtoMessage()               {}
-func (*FuseAttributes) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{5} }
+func (*FuseAttributes) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{6} }
 
 func (m *FuseAttributes) GetFileSize() uint64 {
 	if m != nil {
@@ -204,54 +257,76 @@ func (m *FuseAttributes) GetGid() uint32 {
 	return 0
 }
 
-type GetFileAttributesRequest struct {
+func (m *FuseAttributes) GetCrtime() int64 {
+	if m != nil {
+		return m.Crtime
+	}
+	return 0
+}
+
+func (m *FuseAttributes) GetMime() string {
+	if m != nil {
+		return m.Mime
+	}
+	return ""
+}
+
+type GetEntryAttributesRequest struct {
 	Name      string `protobuf:"bytes,1,opt,name=name" json:"name,omitempty"`
 	ParentDir string `protobuf:"bytes,2,opt,name=parent_dir,json=parentDir" json:"parent_dir,omitempty"`
 	FileId    string `protobuf:"bytes,3,opt,name=file_id,json=fileId" json:"file_id,omitempty"`
 }
 
-func (m *GetFileAttributesRequest) Reset()                    { *m = GetFileAttributesRequest{} }
-func (m *GetFileAttributesRequest) String() string            { return proto.CompactTextString(m) }
-func (*GetFileAttributesRequest) ProtoMessage()               {}
-func (*GetFileAttributesRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{6} }
+func (m *GetEntryAttributesRequest) Reset()                    { *m = GetEntryAttributesRequest{} }
+func (m *GetEntryAttributesRequest) String() string            { return proto.CompactTextString(m) }
+func (*GetEntryAttributesRequest) ProtoMessage()               {}
+func (*GetEntryAttributesRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{7} }
 
-func (m *GetFileAttributesRequest) GetName() string {
+func (m *GetEntryAttributesRequest) GetName() string {
 	if m != nil {
 		return m.Name
 	}
 	return ""
 }
 
-func (m *GetFileAttributesRequest) GetParentDir() string {
+func (m *GetEntryAttributesRequest) GetParentDir() string {
 	if m != nil {
 		return m.ParentDir
 	}
 	return ""
 }
 
-func (m *GetFileAttributesRequest) GetFileId() string {
+func (m *GetEntryAttributesRequest) GetFileId() string {
 	if m != nil {
 		return m.FileId
 	}
 	return ""
 }
 
-type GetFileAttributesResponse struct {
+type GetEntryAttributesResponse struct {
 	Attributes *FuseAttributes `protobuf:"bytes,1,opt,name=attributes" json:"attributes,omitempty"`
+	Chunks     []*FileChunk    `protobuf:"bytes,2,rep,name=chunks" json:"chunks,omitempty"`
 }
 
-func (m *GetFileAttributesResponse) Reset()                    { *m = GetFileAttributesResponse{} }
-func (m *GetFileAttributesResponse) String() string            { return proto.CompactTextString(m) }
-func (*GetFileAttributesResponse) ProtoMessage()               {}
-func (*GetFileAttributesResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{7} }
+func (m *GetEntryAttributesResponse) Reset()                    { *m = GetEntryAttributesResponse{} }
+func (m *GetEntryAttributesResponse) String() string            { return proto.CompactTextString(m) }
+func (*GetEntryAttributesResponse) ProtoMessage()               {}
+func (*GetEntryAttributesResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{8} }
 
-func (m *GetFileAttributesResponse) GetAttributes() *FuseAttributes {
+func (m *GetEntryAttributesResponse) GetAttributes() *FuseAttributes {
 	if m != nil {
 		return m.Attributes
 	}
 	return nil
 }
 
+func (m *GetEntryAttributesResponse) GetChunks() []*FileChunk {
+	if m != nil {
+		return m.Chunks
+	}
+	return nil
+}
+
 type GetFileContentRequest struct {
 	FileId string `protobuf:"bytes,1,opt,name=file_id,json=fileId" json:"file_id,omitempty"`
 }
@@ -259,7 +334,7 @@ type GetFileContentRequest struct {
 func (m *GetFileContentRequest) Reset()                    { *m = GetFileContentRequest{} }
 func (m *GetFileContentRequest) String() string            { return proto.CompactTextString(m) }
 func (*GetFileContentRequest) ProtoMessage()               {}
-func (*GetFileContentRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{8} }
+func (*GetFileContentRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{9} }
 
 func (m *GetFileContentRequest) GetFileId() string {
 	if m != nil {
@@ -275,7 +350,7 @@ type GetFileContentResponse struct {
 func (m *GetFileContentResponse) Reset()                    { *m = GetFileContentResponse{} }
 func (m *GetFileContentResponse) String() string            { return proto.CompactTextString(m) }
 func (*GetFileContentResponse) ProtoMessage()               {}
-func (*GetFileContentResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{9} }
+func (*GetFileContentResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{10} }
 
 func (m *GetFileContentResponse) GetContent() []byte {
 	if m != nil {
@@ -284,6 +359,70 @@ func (m *GetFileContentResponse) GetContent() []byte {
 	return nil
 }
 
+type CreateEntryRequest struct {
+	Directory string `protobuf:"bytes,1,opt,name=directory" json:"directory,omitempty"`
+	Entry     *Entry `protobuf:"bytes,2,opt,name=entry" json:"entry,omitempty"`
+}
+
+func (m *CreateEntryRequest) Reset()                    { *m = CreateEntryRequest{} }
+func (m *CreateEntryRequest) String() string            { return proto.CompactTextString(m) }
+func (*CreateEntryRequest) ProtoMessage()               {}
+func (*CreateEntryRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{11} }
+
+func (m *CreateEntryRequest) GetDirectory() string {
+	if m != nil {
+		return m.Directory
+	}
+	return ""
+}
+
+func (m *CreateEntryRequest) GetEntry() *Entry {
+	if m != nil {
+		return m.Entry
+	}
+	return nil
+}
+
+type CreateEntryResponse struct {
+}
+
+func (m *CreateEntryResponse) Reset()                    { *m = CreateEntryResponse{} }
+func (m *CreateEntryResponse) String() string            { return proto.CompactTextString(m) }
+func (*CreateEntryResponse) ProtoMessage()               {}
+func (*CreateEntryResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{12} }
+
+type UpdateEntryRequest struct {
+	Directory string `protobuf:"bytes,1,opt,name=directory" json:"directory,omitempty"`
+	Entry     *Entry `protobuf:"bytes,2,opt,name=entry" json:"entry,omitempty"`
+}
+
+func (m *UpdateEntryRequest) Reset()                    { *m = UpdateEntryRequest{} }
+func (m *UpdateEntryRequest) String() string            { return proto.CompactTextString(m) }
+func (*UpdateEntryRequest) ProtoMessage()               {}
+func (*UpdateEntryRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{13} }
+
+func (m *UpdateEntryRequest) GetDirectory() string {
+	if m != nil {
+		return m.Directory
+	}
+	return ""
+}
+
+func (m *UpdateEntryRequest) GetEntry() *Entry {
+	if m != nil {
+		return m.Entry
+	}
+	return nil
+}
+
+type UpdateEntryResponse struct {
+}
+
+func (m *UpdateEntryResponse) Reset()                    { *m = UpdateEntryResponse{} }
+func (m *UpdateEntryResponse) String() string            { return proto.CompactTextString(m) }
+func (*UpdateEntryResponse) ProtoMessage()               {}
+func (*UpdateEntryResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{14} }
+
 type DeleteEntryRequest struct {
 	Directory   string `protobuf:"bytes,1,opt,name=directory" json:"directory,omitempty"`
 	Name        string `protobuf:"bytes,2,opt,name=name" json:"name,omitempty"`
@@ -293,7 +432,7 @@ type DeleteEntryRequest struct {
 func (m *DeleteEntryRequest) Reset()                    { *m = DeleteEntryRequest{} }
 func (m *DeleteEntryRequest) String() string            { return proto.CompactTextString(m) }
 func (*DeleteEntryRequest) ProtoMessage()               {}
-func (*DeleteEntryRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{10} }
+func (*DeleteEntryRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{15} }
 
 func (m *DeleteEntryRequest) GetDirectory() string {
 	if m != nil {
@@ -322,7 +461,151 @@ type DeleteEntryResponse struct {
 func (m *DeleteEntryResponse) Reset()                    { *m = DeleteEntryResponse{} }
 func (m *DeleteEntryResponse) String() string            { return proto.CompactTextString(m) }
 func (*DeleteEntryResponse) ProtoMessage()               {}
-func (*DeleteEntryResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{11} }
+func (*DeleteEntryResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{16} }
+
+type AssignVolumeRequest struct {
+	Count       int32  `protobuf:"varint,1,opt,name=count" json:"count,omitempty"`
+	Collection  string `protobuf:"bytes,2,opt,name=collection" json:"collection,omitempty"`
+	Replication string `protobuf:"bytes,3,opt,name=replication" json:"replication,omitempty"`
+}
+
+func (m *AssignVolumeRequest) Reset()                    { *m = AssignVolumeRequest{} }
+func (m *AssignVolumeRequest) String() string            { return proto.CompactTextString(m) }
+func (*AssignVolumeRequest) ProtoMessage()               {}
+func (*AssignVolumeRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{17} }
+
+func (m *AssignVolumeRequest) GetCount() int32 {
+	if m != nil {
+		return m.Count
+	}
+	return 0
+}
+
+func (m *AssignVolumeRequest) GetCollection() string {
+	if m != nil {
+		return m.Collection
+	}
+	return ""
+}
+
+func (m *AssignVolumeRequest) GetReplication() string {
+	if m != nil {
+		return m.Replication
+	}
+	return ""
+}
+
+type AssignVolumeResponse struct {
+	FileId    string `protobuf:"bytes,1,opt,name=file_id,json=fileId" json:"file_id,omitempty"`
+	Url       string `protobuf:"bytes,2,opt,name=url" json:"url,omitempty"`
+	PublicUrl string `protobuf:"bytes,3,opt,name=public_url,json=publicUrl" json:"public_url,omitempty"`
+	Count     int32  `protobuf:"varint,4,opt,name=count" json:"count,omitempty"`
+}
+
+func (m *AssignVolumeResponse) Reset()                    { *m = AssignVolumeResponse{} }
+func (m *AssignVolumeResponse) String() string            { return proto.CompactTextString(m) }
+func (*AssignVolumeResponse) ProtoMessage()               {}
+func (*AssignVolumeResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{18} }
+
+func (m *AssignVolumeResponse) GetFileId() string {
+	if m != nil {
+		return m.FileId
+	}
+	return ""
+}
+
+func (m *AssignVolumeResponse) GetUrl() string {
+	if m != nil {
+		return m.Url
+	}
+	return ""
+}
+
+func (m *AssignVolumeResponse) GetPublicUrl() string {
+	if m != nil {
+		return m.PublicUrl
+	}
+	return ""
+}
+
+func (m *AssignVolumeResponse) GetCount() int32 {
+	if m != nil {
+		return m.Count
+	}
+	return 0
+}
+
+type LookupVolumeRequest struct {
+	VolumeIds []string `protobuf:"bytes,1,rep,name=volume_ids,json=volumeIds" json:"volume_ids,omitempty"`
+}
+
+func (m *LookupVolumeRequest) Reset()                    { *m = LookupVolumeRequest{} }
+func (m *LookupVolumeRequest) String() string            { return proto.CompactTextString(m) }
+func (*LookupVolumeRequest) ProtoMessage()               {}
+func (*LookupVolumeRequest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{19} }
+
+func (m *LookupVolumeRequest) GetVolumeIds() []string {
+	if m != nil {
+		return m.VolumeIds
+	}
+	return nil
+}
+
+type Locations struct {
+	Locations []*Location `protobuf:"bytes,1,rep,name=locations" json:"locations,omitempty"`
+}
+
+func (m *Locations) Reset()                    { *m = Locations{} }
+func (m *Locations) String() string            { return proto.CompactTextString(m) }
+func (*Locations) ProtoMessage()               {}
+func (*Locations) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{20} }
+
+func (m *Locations) GetLocations() []*Location {
+	if m != nil {
+		return m.Locations
+	}
+	return nil
+}
+
+type Location struct {
+	Url       string `protobuf:"bytes,1,opt,name=url" json:"url,omitempty"`
+	PublicUrl string `protobuf:"bytes,2,opt,name=public_url,json=publicUrl" json:"public_url,omitempty"`
+}
+
+func (m *Location) Reset()                    { *m = Location{} }
+func (m *Location) String() string            { return proto.CompactTextString(m) }
+func (*Location) ProtoMessage()               {}
+func (*Location) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{21} }
+
+func (m *Location) GetUrl() string {
+	if m != nil {
+		return m.Url
+	}
+	return ""
+}
+
+func (m *Location) GetPublicUrl() string {
+	if m != nil {
+		return m.PublicUrl
+	}
+	return ""
+}
+
+type LookupVolumeResponse struct {
+	LocationsMap map[string]*Locations `protobuf:"bytes,1,rep,name=locations_map,json=locationsMap" json:"locations_map,omitempty" protobuf_key:"bytes,1,opt,name=key" protobuf_val:"bytes,2,opt,name=value"`
+}
+
+func (m *LookupVolumeResponse) Reset()                    { *m = LookupVolumeResponse{} }
+func (m *LookupVolumeResponse) String() string            { return proto.CompactTextString(m) }
+func (*LookupVolumeResponse) ProtoMessage()               {}
+func (*LookupVolumeResponse) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{22} }
+
+func (m *LookupVolumeResponse) GetLocationsMap() map[string]*Locations {
+	if m != nil {
+		return m.LocationsMap
+	}
+	return nil
+}
 
 func init() {
 	proto.RegisterType((*LookupDirectoryEntryRequest)(nil), "filer_pb.LookupDirectoryEntryRequest")
@@ -330,13 +613,24 @@ func init() {
 	proto.RegisterType((*ListEntriesRequest)(nil), "filer_pb.ListEntriesRequest")
 	proto.RegisterType((*ListEntriesResponse)(nil), "filer_pb.ListEntriesResponse")
 	proto.RegisterType((*Entry)(nil), "filer_pb.Entry")
+	proto.RegisterType((*FileChunk)(nil), "filer_pb.FileChunk")
 	proto.RegisterType((*FuseAttributes)(nil), "filer_pb.FuseAttributes")
-	proto.RegisterType((*GetFileAttributesRequest)(nil), "filer_pb.GetFileAttributesRequest")
-	proto.RegisterType((*GetFileAttributesResponse)(nil), "filer_pb.GetFileAttributesResponse")
+	proto.RegisterType((*GetEntryAttributesRequest)(nil), "filer_pb.GetEntryAttributesRequest")
+	proto.RegisterType((*GetEntryAttributesResponse)(nil), "filer_pb.GetEntryAttributesResponse")
 	proto.RegisterType((*GetFileContentRequest)(nil), "filer_pb.GetFileContentRequest")
 	proto.RegisterType((*GetFileContentResponse)(nil), "filer_pb.GetFileContentResponse")
+	proto.RegisterType((*CreateEntryRequest)(nil), "filer_pb.CreateEntryRequest")
+	proto.RegisterType((*CreateEntryResponse)(nil), "filer_pb.CreateEntryResponse")
+	proto.RegisterType((*UpdateEntryRequest)(nil), "filer_pb.UpdateEntryRequest")
+	proto.RegisterType((*UpdateEntryResponse)(nil), "filer_pb.UpdateEntryResponse")
 	proto.RegisterType((*DeleteEntryRequest)(nil), "filer_pb.DeleteEntryRequest")
 	proto.RegisterType((*DeleteEntryResponse)(nil), "filer_pb.DeleteEntryResponse")
+	proto.RegisterType((*AssignVolumeRequest)(nil), "filer_pb.AssignVolumeRequest")
+	proto.RegisterType((*AssignVolumeResponse)(nil), "filer_pb.AssignVolumeResponse")
+	proto.RegisterType((*LookupVolumeRequest)(nil), "filer_pb.LookupVolumeRequest")
+	proto.RegisterType((*Locations)(nil), "filer_pb.Locations")
+	proto.RegisterType((*Location)(nil), "filer_pb.Location")
+	proto.RegisterType((*LookupVolumeResponse)(nil), "filer_pb.LookupVolumeResponse")
 }
 
 // Reference imports to suppress errors if they are not otherwise used.
@@ -352,9 +646,12 @@ const _ = grpc.SupportPackageIsVersion4
 type SeaweedFilerClient interface {
 	LookupDirectoryEntry(ctx context.Context, in *LookupDirectoryEntryRequest, opts ...grpc.CallOption) (*LookupDirectoryEntryResponse, error)
 	ListEntries(ctx context.Context, in *ListEntriesRequest, opts ...grpc.CallOption) (*ListEntriesResponse, error)
-	GetFileAttributes(ctx context.Context, in *GetFileAttributesRequest, opts ...grpc.CallOption) (*GetFileAttributesResponse, error)
-	GetFileContent(ctx context.Context, in *GetFileContentRequest, opts ...grpc.CallOption) (*GetFileContentResponse, error)
+	GetEntryAttributes(ctx context.Context, in *GetEntryAttributesRequest, opts ...grpc.CallOption) (*GetEntryAttributesResponse, error)
+	CreateEntry(ctx context.Context, in *CreateEntryRequest, opts ...grpc.CallOption) (*CreateEntryResponse, error)
+	UpdateEntry(ctx context.Context, in *UpdateEntryRequest, opts ...grpc.CallOption) (*UpdateEntryResponse, error)
 	DeleteEntry(ctx context.Context, in *DeleteEntryRequest, opts ...grpc.CallOption) (*DeleteEntryResponse, error)
+	AssignVolume(ctx context.Context, in *AssignVolumeRequest, opts ...grpc.CallOption) (*AssignVolumeResponse, error)
+	LookupVolume(ctx context.Context, in *LookupVolumeRequest, opts ...grpc.CallOption) (*LookupVolumeResponse, error)
 }
 
 type seaweedFilerClient struct {
@@ -383,18 +680,27 @@ func (c *seaweedFilerClient) ListEntries(ctx context.Context, in *ListEntriesReq
 	return out, nil
 }
 
-func (c *seaweedFilerClient) GetFileAttributes(ctx context.Context, in *GetFileAttributesRequest, opts ...grpc.CallOption) (*GetFileAttributesResponse, error) {
-	out := new(GetFileAttributesResponse)
-	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/GetFileAttributes", in, out, c.cc, opts...)
+func (c *seaweedFilerClient) GetEntryAttributes(ctx context.Context, in *GetEntryAttributesRequest, opts ...grpc.CallOption) (*GetEntryAttributesResponse, error) {
+	out := new(GetEntryAttributesResponse)
+	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/GetEntryAttributes", in, out, c.cc, opts...)
 	if err != nil {
 		return nil, err
 	}
 	return out, nil
 }
 
-func (c *seaweedFilerClient) GetFileContent(ctx context.Context, in *GetFileContentRequest, opts ...grpc.CallOption) (*GetFileContentResponse, error) {
-	out := new(GetFileContentResponse)
-	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/GetFileContent", in, out, c.cc, opts...)
+func (c *seaweedFilerClient) CreateEntry(ctx context.Context, in *CreateEntryRequest, opts ...grpc.CallOption) (*CreateEntryResponse, error) {
+	out := new(CreateEntryResponse)
+	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/CreateEntry", in, out, c.cc, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *seaweedFilerClient) UpdateEntry(ctx context.Context, in *UpdateEntryRequest, opts ...grpc.CallOption) (*UpdateEntryResponse, error) {
+	out := new(UpdateEntryResponse)
+	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/UpdateEntry", in, out, c.cc, opts...)
 	if err != nil {
 		return nil, err
 	}
@@ -410,14 +716,35 @@ func (c *seaweedFilerClient) DeleteEntry(ctx context.Context, in *DeleteEntryReq
 	return out, nil
 }
 
+func (c *seaweedFilerClient) AssignVolume(ctx context.Context, in *AssignVolumeRequest, opts ...grpc.CallOption) (*AssignVolumeResponse, error) {
+	out := new(AssignVolumeResponse)
+	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/AssignVolume", in, out, c.cc, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *seaweedFilerClient) LookupVolume(ctx context.Context, in *LookupVolumeRequest, opts ...grpc.CallOption) (*LookupVolumeResponse, error) {
+	out := new(LookupVolumeResponse)
+	err := grpc.Invoke(ctx, "/filer_pb.SeaweedFiler/LookupVolume", in, out, c.cc, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
 // Server API for SeaweedFiler service
 
 type SeaweedFilerServer interface {
 	LookupDirectoryEntry(context.Context, *LookupDirectoryEntryRequest) (*LookupDirectoryEntryResponse, error)
 	ListEntries(context.Context, *ListEntriesRequest) (*ListEntriesResponse, error)
-	GetFileAttributes(context.Context, *GetFileAttributesRequest) (*GetFileAttributesResponse, error)
-	GetFileContent(context.Context, *GetFileContentRequest) (*GetFileContentResponse, error)
+	GetEntryAttributes(context.Context, *GetEntryAttributesRequest) (*GetEntryAttributesResponse, error)
+	CreateEntry(context.Context, *CreateEntryRequest) (*CreateEntryResponse, error)
+	UpdateEntry(context.Context, *UpdateEntryRequest) (*UpdateEntryResponse, error)
 	DeleteEntry(context.Context, *DeleteEntryRequest) (*DeleteEntryResponse, error)
+	AssignVolume(context.Context, *AssignVolumeRequest) (*AssignVolumeResponse, error)
+	LookupVolume(context.Context, *LookupVolumeRequest) (*LookupVolumeResponse, error)
 }
 
 func RegisterSeaweedFilerServer(s *grpc.Server, srv SeaweedFilerServer) {
@@ -460,38 +787,56 @@ func _SeaweedFiler_ListEntries_Handler(srv interface{}, ctx context.Context, dec
 	return interceptor(ctx, in, info, handler)
 }
 
-func _SeaweedFiler_GetFileAttributes_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
-	in := new(GetFileAttributesRequest)
+func _SeaweedFiler_GetEntryAttributes_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(GetEntryAttributesRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(SeaweedFilerServer).GetEntryAttributes(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/filer_pb.SeaweedFiler/GetEntryAttributes",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(SeaweedFilerServer).GetEntryAttributes(ctx, req.(*GetEntryAttributesRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _SeaweedFiler_CreateEntry_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(CreateEntryRequest)
 	if err := dec(in); err != nil {
 		return nil, err
 	}
 	if interceptor == nil {
-		return srv.(SeaweedFilerServer).GetFileAttributes(ctx, in)
+		return srv.(SeaweedFilerServer).CreateEntry(ctx, in)
 	}
 	info := &grpc.UnaryServerInfo{
 		Server:     srv,
-		FullMethod: "/filer_pb.SeaweedFiler/GetFileAttributes",
+		FullMethod: "/filer_pb.SeaweedFiler/CreateEntry",
 	}
 	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
-		return srv.(SeaweedFilerServer).GetFileAttributes(ctx, req.(*GetFileAttributesRequest))
+		return srv.(SeaweedFilerServer).CreateEntry(ctx, req.(*CreateEntryRequest))
 	}
 	return interceptor(ctx, in, info, handler)
 }
 
-func _SeaweedFiler_GetFileContent_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
-	in := new(GetFileContentRequest)
+func _SeaweedFiler_UpdateEntry_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(UpdateEntryRequest)
 	if err := dec(in); err != nil {
 		return nil, err
 	}
 	if interceptor == nil {
-		return srv.(SeaweedFilerServer).GetFileContent(ctx, in)
+		return srv.(SeaweedFilerServer).UpdateEntry(ctx, in)
 	}
 	info := &grpc.UnaryServerInfo{
 		Server:     srv,
-		FullMethod: "/filer_pb.SeaweedFiler/GetFileContent",
+		FullMethod: "/filer_pb.SeaweedFiler/UpdateEntry",
 	}
 	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
-		return srv.(SeaweedFilerServer).GetFileContent(ctx, req.(*GetFileContentRequest))
+		return srv.(SeaweedFilerServer).UpdateEntry(ctx, req.(*UpdateEntryRequest))
 	}
 	return interceptor(ctx, in, info, handler)
 }
@@ -514,6 +859,42 @@ func _SeaweedFiler_DeleteEntry_Handler(srv interface{}, ctx context.Context, dec
 	return interceptor(ctx, in, info, handler)
 }
 
+func _SeaweedFiler_AssignVolume_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(AssignVolumeRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(SeaweedFilerServer).AssignVolume(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/filer_pb.SeaweedFiler/AssignVolume",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(SeaweedFilerServer).AssignVolume(ctx, req.(*AssignVolumeRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _SeaweedFiler_LookupVolume_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(LookupVolumeRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(SeaweedFilerServer).LookupVolume(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/filer_pb.SeaweedFiler/LookupVolume",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(SeaweedFilerServer).LookupVolume(ctx, req.(*LookupVolumeRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
 var _SeaweedFiler_serviceDesc = grpc.ServiceDesc{
 	ServiceName: "filer_pb.SeaweedFiler",
 	HandlerType: (*SeaweedFilerServer)(nil),
@@ -527,17 +908,29 @@ var _SeaweedFiler_serviceDesc = grpc.ServiceDesc{
 			Handler:    _SeaweedFiler_ListEntries_Handler,
 		},
 		{
-			MethodName: "GetFileAttributes",
-			Handler:    _SeaweedFiler_GetFileAttributes_Handler,
+			MethodName: "GetEntryAttributes",
+			Handler:    _SeaweedFiler_GetEntryAttributes_Handler,
+		},
+		{
+			MethodName: "CreateEntry",
+			Handler:    _SeaweedFiler_CreateEntry_Handler,
 		},
 		{
-			MethodName: "GetFileContent",
-			Handler:    _SeaweedFiler_GetFileContent_Handler,
+			MethodName: "UpdateEntry",
+			Handler:    _SeaweedFiler_UpdateEntry_Handler,
 		},
 		{
 			MethodName: "DeleteEntry",
 			Handler:    _SeaweedFiler_DeleteEntry_Handler,
 		},
+		{
+			MethodName: "AssignVolume",
+			Handler:    _SeaweedFiler_AssignVolume_Handler,
+		},
+		{
+			MethodName: "LookupVolume",
+			Handler:    _SeaweedFiler_LookupVolume_Handler,
+		},
 	},
 	Streams:  []grpc.StreamDesc{},
 	Metadata: "filer.proto",
@@ -546,39 +939,62 @@ var _SeaweedFiler_serviceDesc = grpc.ServiceDesc{
 func init() { proto.RegisterFile("filer.proto", fileDescriptor0) }
 
 var fileDescriptor0 = []byte{
-	// 532 bytes of a gzipped FileDescriptorProto
-	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0xa4, 0x54, 0x4d, 0x6f, 0xd3, 0x40,
-	0x10, 0xad, 0x71, 0xd2, 0x34, 0x93, 0xb4, 0xc0, 0xb4, 0x05, 0x93, 0xa6, 0x22, 0x2c, 0x2a, 0x82,
-	0x4b, 0x84, 0xc2, 0x85, 0x23, 0x88, 0xb4, 0x08, 0x29, 0x08, 0xc9, 0x55, 0xaf, 0x44, 0x49, 0x3d,
-	0x8d, 0x56, 0x24, 0x76, 0xf0, 0xae, 0x85, 0xda, 0x33, 0x7f, 0x80, 0xbf, 0xc5, 0xaf, 0x42, 0xbb,
-	0xeb, 0x8f, 0x35, 0x76, 0x0a, 0x88, 0x9b, 0xf7, 0xcd, 0xbe, 0x99, 0x37, 0x6f, 0x66, 0x0d, 0x9d,
-	0x2b, 0xbe, 0xa4, 0x78, 0xb8, 0x8e, 0x23, 0x19, 0xe1, 0x8e, 0x3e, 0x4c, 0xd7, 0x73, 0xf6, 0x09,
-	0x8e, 0x26, 0x51, 0xf4, 0x25, 0x59, 0x8f, 0x79, 0x4c, 0x97, 0x32, 0x8a, 0xaf, 0x4f, 0x43, 0x19,
-	0x5f, 0xfb, 0xf4, 0x35, 0x21, 0x21, 0xb1, 0x0f, 0xed, 0x20, 0x0b, 0x78, 0xce, 0xc0, 0x79, 0xde,
-	0xf6, 0x0b, 0x00, 0x11, 0x1a, 0xe1, 0x6c, 0x45, 0xde, 0x1d, 0x1d, 0xd0, 0xdf, 0xec, 0x14, 0xfa,
-	0xf5, 0x09, 0xc5, 0x3a, 0x0a, 0x05, 0xe1, 0x09, 0x34, 0x49, 0x01, 0x3a, 0x5b, 0x67, 0x74, 0x77,
-	0x98, 0x49, 0x19, 0x9a, 0x7b, 0x26, 0xca, 0x46, 0x80, 0x13, 0x2e, 0xa4, 0xc2, 0x38, 0x89, 0xbf,
-	0x92, 0xc3, 0xde, 0xc0, 0x7e, 0x89, 0x93, 0x56, 0x7c, 0x01, 0x2d, 0x32, 0x90, 0xe7, 0x0c, 0xdc,
-	0xba, 0x9a, 0x59, 0x9c, 0xfd, 0x70, 0xa0, 0xa9, 0xa1, 0xbc, 0x35, 0xa7, 0x68, 0x0d, 0x9f, 0x40,
-	0x97, 0x8b, 0x69, 0x21, 0x40, 0xb5, 0xbd, 0xe3, 0x77, 0xb8, 0xc8, 0x5b, 0xc5, 0x87, 0xd0, 0x52,
-	0xb9, 0xa7, 0x3c, 0xf0, 0x5c, 0xcd, 0xdc, 0x56, 0xc7, 0x0f, 0x01, 0xbe, 0x06, 0x98, 0x49, 0x19,
-	0xf3, 0x79, 0x22, 0x49, 0x78, 0x0d, 0xdd, 0xbb, 0x57, 0xe8, 0x38, 0x4b, 0x04, 0xbd, 0xcd, 0xe3,
-	0xbe, 0x75, 0x97, 0x7d, 0x77, 0x60, 0xaf, 0x1c, 0xc6, 0x23, 0x68, 0xeb, 0x2a, 0x82, 0xdf, 0x18,
-	0x85, 0x0d, 0x5f, 0x4f, 0xf4, 0x9c, 0xdf, 0x10, 0x1e, 0x40, 0x73, 0x25, 0x79, 0x3a, 0x15, 0xd7,
-	0x37, 0x87, 0x9c, 0xb2, 0x8a, 0x02, 0xd2, 0xd2, 0x76, 0x0d, 0xe5, 0x63, 0x14, 0x10, 0xde, 0x03,
-	0x37, 0xe1, 0x81, 0x56, 0xb5, 0xeb, 0xab, 0x4f, 0x85, 0x2c, 0x78, 0xe0, 0x35, 0x0d, 0xb2, 0xe0,
-	0x01, 0xbb, 0x02, 0xef, 0x3d, 0xc9, 0x33, 0xbe, 0xb4, 0x75, 0xa6, 0x63, 0xa9, 0x33, 0xeb, 0x18,
-	0x60, 0x3d, 0x8b, 0x29, 0x94, 0xca, 0xb0, 0x74, 0x43, 0xda, 0x06, 0x19, 0xf3, 0x78, 0xa3, 0x51,
-	0xec, 0x02, 0x1e, 0xd5, 0xd4, 0x49, 0x47, 0x59, 0x76, 0xd1, 0xf9, 0x07, 0x17, 0x5f, 0xc2, 0x61,
-	0x9a, 0xf6, 0x5d, 0x14, 0x4a, 0x0a, 0x65, 0xa6, 0xdd, 0x12, 0xe2, 0x94, 0x84, 0x8c, 0xe0, 0xc1,
-	0xef, 0x8c, 0x54, 0x85, 0x07, 0xad, 0x4b, 0x03, 0x69, 0x4a, 0xd7, 0xcf, 0x8e, 0x8c, 0x03, 0x8e,
-	0x69, 0x49, 0x92, 0xfe, 0xef, 0x11, 0x55, 0x36, 0xcd, 0xad, 0x6c, 0x1a, 0x3b, 0x84, 0xfd, 0x52,
-	0x29, 0xa3, 0x6d, 0xf4, 0xd3, 0x85, 0xee, 0x39, 0xcd, 0xbe, 0x11, 0x05, 0x4a, 0x7a, 0x8c, 0x0b,
-	0x38, 0xa8, 0x7b, 0x8f, 0x78, 0x52, 0xd8, 0x76, 0xcb, 0x0f, 0xa0, 0xf7, 0xec, 0x4f, 0xd7, 0x4c,
-	0x5d, 0xb6, 0x85, 0x13, 0xe8, 0x58, 0xaf, 0x0f, 0xfb, 0x16, 0xb1, 0xf2, 0x90, 0x7b, 0xc7, 0x1b,
-	0xa2, 0x79, 0xb6, 0xcf, 0x70, 0xbf, 0xb2, 0x06, 0xc8, 0x0a, 0xd6, 0xa6, 0x5d, 0xec, 0x3d, 0xbd,
-	0xf5, 0x4e, 0x9e, 0xff, 0x02, 0xf6, 0xca, 0xd3, 0xc5, 0xc7, 0x15, 0x62, 0x79, 0x53, 0x7a, 0x83,
-	0xcd, 0x17, 0x6c, 0x13, 0xac, 0xa9, 0xd8, 0x26, 0x54, 0xf7, 0xc2, 0x36, 0xa1, 0x66, 0x94, 0x6c,
-	0x6b, 0xbe, 0xad, 0xff, 0xd6, 0xaf, 0x7e, 0x05, 0x00, 0x00, 0xff, 0xff, 0x6a, 0x83, 0x8a, 0x32,
-	0xbc, 0x05, 0x00, 0x00,
+	// 906 bytes of a gzipped FileDescriptorProto
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0xb4, 0x56, 0x5b, 0x6f, 0xdc, 0x44,
+	0x14, 0x8e, 0xd7, 0xd9, 0x4d, 0x7c, 0x76, 0xc3, 0x65, 0x36, 0x2d, 0x66, 0x9b, 0x54, 0x61, 0xa0,
+	0xa8, 0x15, 0x52, 0x14, 0x05, 0x1e, 0x2a, 0x10, 0x12, 0x55, 0x53, 0xaa, 0x4a, 0xa9, 0x2a, 0xb9,
+	0x04, 0x89, 0xa7, 0x95, 0x63, 0x9f, 0x5d, 0x46, 0xf1, 0x0d, 0xcf, 0x38, 0x28, 0xbc, 0xf2, 0x5b,
+	0x78, 0xe7, 0x81, 0x7f, 0xc0, 0x1f, 0x43, 0x73, 0xb1, 0x3d, 0x8e, 0xbd, 0xbd, 0x3c, 0xf0, 0x36,
+	0x73, 0x2e, 0xdf, 0xf9, 0xce, 0xcc, 0x99, 0xcf, 0x86, 0xe9, 0x8a, 0x25, 0x58, 0x1e, 0x17, 0x65,
+	0x2e, 0x72, 0xb2, 0xab, 0x36, 0xcb, 0xe2, 0x92, 0xbe, 0x82, 0x7b, 0xe7, 0x79, 0x7e, 0x55, 0x15,
+	0x67, 0xac, 0xc4, 0x48, 0xe4, 0xe5, 0xcd, 0xb3, 0x4c, 0x94, 0x37, 0x01, 0xfe, 0x56, 0x21, 0x17,
+	0xe4, 0x00, 0xbc, 0xb8, 0x76, 0xf8, 0xce, 0x91, 0xf3, 0xd0, 0x0b, 0x5a, 0x03, 0x21, 0xb0, 0x9d,
+	0x85, 0x29, 0xfa, 0x23, 0xe5, 0x50, 0x6b, 0xfa, 0x0c, 0x0e, 0x86, 0x01, 0x79, 0x91, 0x67, 0x1c,
+	0xc9, 0x03, 0x18, 0xa3, 0x34, 0x28, 0xb4, 0xe9, 0xe9, 0x87, 0xc7, 0x35, 0x95, 0x63, 0x1d, 0xa7,
+	0xbd, 0xf4, 0x14, 0xc8, 0x39, 0xe3, 0x42, 0xda, 0x18, 0xf2, 0x77, 0xa2, 0x43, 0x7f, 0x80, 0x79,
+	0x27, 0xc7, 0x54, 0x7c, 0x04, 0x3b, 0xa8, 0x4d, 0xbe, 0x73, 0xe4, 0x0e, 0xd5, 0xac, 0xfd, 0xf4,
+	0x2f, 0x07, 0xc6, 0xca, 0xd4, 0xb4, 0xe6, 0xb4, 0xad, 0x91, 0xcf, 0x60, 0xc6, 0xf8, 0xb2, 0x25,
+	0x20, 0xdb, 0xde, 0x0d, 0xa6, 0x8c, 0x37, 0xad, 0x92, 0xaf, 0x60, 0x12, 0xfd, 0x5a, 0x65, 0x57,
+	0xdc, 0x77, 0x55, 0xa9, 0x79, 0x5b, 0xea, 0x47, 0x96, 0xe0, 0x53, 0xe9, 0x0b, 0x4c, 0x08, 0x79,
+	0x0c, 0x10, 0x0a, 0x51, 0xb2, 0xcb, 0x4a, 0x20, 0xf7, 0xb7, 0xd5, 0x79, 0xf8, 0x56, 0x42, 0xc5,
+	0xf1, 0x49, 0xe3, 0x0f, 0xac, 0x58, 0xba, 0x02, 0xaf, 0x81, 0x23, 0x9f, 0xc0, 0x8e, 0xcc, 0x59,
+	0xb2, 0xd8, 0xb0, 0x9d, 0xc8, 0xed, 0x8b, 0x98, 0xdc, 0x85, 0x49, 0xbe, 0x5a, 0x71, 0x14, 0x8a,
+	0xa9, 0x1b, 0x98, 0x9d, 0xec, 0x8d, 0xb3, 0x3f, 0xd0, 0x77, 0x8f, 0x9c, 0x87, 0xdb, 0x81, 0x5a,
+	0x93, 0x7d, 0x18, 0xa7, 0x82, 0xa5, 0xa8, 0x68, 0xb8, 0x81, 0xde, 0xd0, 0xbf, 0x1d, 0xf8, 0xa0,
+	0x4b, 0x83, 0xdc, 0x03, 0x4f, 0x55, 0x53, 0x08, 0x8e, 0x42, 0x50, 0xd3, 0xf4, 0xba, 0x83, 0x32,
+	0xb2, 0x50, 0x9a, 0x94, 0x34, 0x8f, 0x75, 0xd1, 0x3d, 0x9d, 0xf2, 0x32, 0x8f, 0x91, 0x7c, 0x04,
+	0x6e, 0xc5, 0x62, 0x55, 0x76, 0x2f, 0x90, 0x4b, 0x69, 0x59, 0xb3, 0xd8, 0x1f, 0x6b, 0xcb, 0x9a,
+	0xa9, 0x46, 0xa2, 0x52, 0xe1, 0x4e, 0x74, 0x23, 0x7a, 0x27, 0x1b, 0x49, 0xa5, 0x75, 0x47, 0x5f,
+	0x92, 0x5c, 0xd3, 0x35, 0x7c, 0xfa, 0x1c, 0xd5, 0x0c, 0xdc, 0x58, 0x87, 0x67, 0xe6, 0x67, 0xe8,
+	0x56, 0x0f, 0x01, 0x8a, 0xb0, 0xc4, 0x4c, 0xc8, 0x9b, 0x35, 0xa3, 0xec, 0x69, 0xcb, 0x19, 0x2b,
+	0xed, 0xd3, 0x75, 0xed, 0xd3, 0xa5, 0x7f, 0x3a, 0xb0, 0x18, 0xaa, 0x64, 0xa6, 0xae, 0x7b, 0xb9,
+	0xce, 0xbb, 0x5f, 0xae, 0x35, 0x43, 0xa3, 0xb7, 0xce, 0x10, 0x3d, 0x81, 0x3b, 0xcf, 0x51, 0x28,
+	0x7b, 0x9e, 0x09, 0xcc, 0x44, 0xdd, 0xea, 0xa6, 0xa9, 0xa0, 0xa7, 0x70, 0xf7, 0x76, 0x86, 0xa1,
+	0xec, 0xc3, 0x4e, 0xa4, 0x4d, 0x2a, 0x65, 0x16, 0xd4, 0x5b, 0xfa, 0x0b, 0x90, 0xa7, 0x25, 0x86,
+	0x02, 0xdf, 0x43, 0x1c, 0x9a, 0x87, 0x3e, 0x7a, 0xe3, 0x43, 0xbf, 0x03, 0xf3, 0x0e, 0xb4, 0xe6,
+	0x22, 0x2b, 0x5e, 0x14, 0xf1, 0xff, 0x55, 0xb1, 0x03, 0x6d, 0x2a, 0x32, 0x20, 0x67, 0x98, 0xe0,
+	0x7b, 0x55, 0x1c, 0x10, 0xc0, 0x9e, 0x4a, 0xb8, 0x3d, 0x95, 0x90, 0x0c, 0x3a, 0xa5, 0x0c, 0x83,
+	0x14, 0xe6, 0x4f, 0x38, 0x67, 0xeb, 0xec, 0xe7, 0x3c, 0xa9, 0x52, 0xac, 0x29, 0xec, 0xc3, 0x38,
+	0xca, 0x2b, 0x73, 0x29, 0xe3, 0x40, 0x6f, 0xc8, 0x7d, 0x80, 0x28, 0x4f, 0x12, 0x8c, 0x04, 0xcb,
+	0x33, 0x43, 0xc0, 0xb2, 0x90, 0x23, 0x98, 0x96, 0x58, 0x24, 0x2c, 0x0a, 0x55, 0x80, 0x9e, 0x5d,
+	0xdb, 0x44, 0xaf, 0x61, 0xbf, 0x5b, 0xce, 0x8c, 0xc1, 0x46, 0x3d, 0x91, 0x4f, 0xb5, 0x4c, 0x4c,
+	0x2d, 0xb9, 0x54, 0x6f, 0xa7, 0xba, 0x4c, 0x58, 0xb4, 0x94, 0x0e, 0xd7, 0xbc, 0x1d, 0x65, 0xb9,
+	0x28, 0x93, 0x96, 0xf9, 0xb6, 0xc5, 0x9c, 0x7e, 0x03, 0x73, 0xfd, 0x85, 0xe8, 0xb6, 0x79, 0x08,
+	0x70, 0xad, 0x0c, 0x4b, 0x16, 0x6b, 0xa5, 0xf6, 0x02, 0x4f, 0x5b, 0x5e, 0xc4, 0x9c, 0x7e, 0x0f,
+	0xde, 0x79, 0xae, 0x99, 0x73, 0x72, 0x02, 0x5e, 0x52, 0x6f, 0x8c, 0xa8, 0x93, 0xf6, 0xb6, 0xeb,
+	0xb8, 0xa0, 0x0d, 0xa2, 0xdf, 0xc1, 0x6e, 0x6d, 0xae, 0xfb, 0x70, 0x36, 0xf5, 0x31, 0xba, 0xd5,
+	0x07, 0xfd, 0xd7, 0x81, 0xfd, 0x2e, 0x65, 0x73, 0x54, 0x17, 0xb0, 0xd7, 0x94, 0x58, 0xa6, 0x61,
+	0x61, 0xb8, 0x9c, 0xd8, 0x5c, 0xfa, 0x69, 0x0d, 0x41, 0xfe, 0x32, 0x2c, 0xf4, 0x08, 0xcc, 0x12,
+	0xcb, 0xb4, 0xf8, 0x09, 0x3e, 0xee, 0x85, 0x48, 0xd6, 0x57, 0x58, 0xcf, 0xa0, 0x5c, 0x92, 0x47,
+	0x30, 0xbe, 0x0e, 0x93, 0x0a, 0xcd, 0xbc, 0xcf, 0xfb, 0x27, 0xc0, 0x03, 0x1d, 0xf1, 0xed, 0xe8,
+	0xb1, 0x73, 0xfa, 0xcf, 0x18, 0x66, 0xaf, 0x31, 0xfc, 0x1d, 0x31, 0x96, 0xaf, 0xbf, 0x24, 0xeb,
+	0xba, 0xab, 0xee, 0xa7, 0x9a, 0x3c, 0xb8, 0x4d, 0x7f, 0xf0, 0xdf, 0x60, 0xf1, 0xe5, 0xdb, 0xc2,
+	0xcc, 0x58, 0x6f, 0x91, 0x73, 0x98, 0x5a, 0x1f, 0x66, 0x72, 0x60, 0x25, 0xf6, 0xbe, 0xf1, 0x8b,
+	0xc3, 0x0d, 0xde, 0x06, 0x2d, 0x04, 0xd2, 0xd7, 0x5d, 0xf2, 0x79, 0x9b, 0xb6, 0x51, 0xff, 0x17,
+	0x5f, 0xbc, 0x39, 0xc8, 0x26, 0x6c, 0x89, 0x92, 0x4d, 0xb8, 0x2f, 0x83, 0x36, 0xe1, 0x21, 0x25,
+	0x53, 0x68, 0x96, 0xe0, 0xd8, 0x68, 0x7d, 0x89, 0xb3, 0xd1, 0x86, 0x54, 0x4a, 0xa1, 0x59, 0xe2,
+	0x61, 0xa3, 0xf5, 0xe5, 0xcb, 0x46, 0x1b, 0x52, 0x9c, 0x2d, 0xf2, 0x0a, 0x66, 0xb6, 0x08, 0x10,
+	0x2b, 0x61, 0x40, 0x8b, 0x16, 0xf7, 0x37, 0xb9, 0x6d, 0x40, 0x7b, 0xe6, 0x6d, 0xc0, 0x81, 0x57,
+	0x6f, 0x03, 0x0e, 0x3d, 0x15, 0xba, 0x75, 0x39, 0x51, 0xbf, 0xac, 0x5f, 0xff, 0x17, 0x00, 0x00,
+	0xff, 0xff, 0xf1, 0x42, 0x51, 0xbb, 0xc1, 0x0a, 0x00, 0x00,
 }

weed/pb/master_pb/seaweed.pb.go

@@ -12,6 +12,7 @@ It has these top-level messages:
 	Heartbeat
 	HeartbeatResponse
 	VolumeInformationMessage
+	Empty
 */
 package master_pb
 
@@ -235,10 +236,19 @@ func (m *VolumeInformationMessage) GetTtl() uint32 {
 	return 0
 }
 
+type Empty struct {
+}
+
+func (m *Empty) Reset()                    { *m = Empty{} }
+func (m *Empty) String() string            { return proto.CompactTextString(m) }
+func (*Empty) ProtoMessage()               {}
+func (*Empty) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{3} }
+
 func init() {
 	proto.RegisterType((*Heartbeat)(nil), "master_pb.Heartbeat")
 	proto.RegisterType((*HeartbeatResponse)(nil), "master_pb.HeartbeatResponse")
 	proto.RegisterType((*VolumeInformationMessage)(nil), "master_pb.VolumeInformationMessage")
+	proto.RegisterType((*Empty)(nil), "master_pb.Empty")
 }
 
 // Reference imports to suppress errors if they are not otherwise used.
@@ -253,6 +263,7 @@ const _ = grpc.SupportPackageIsVersion4
 
 type SeaweedClient interface {
 	SendHeartbeat(ctx context.Context, opts ...grpc.CallOption) (Seaweed_SendHeartbeatClient, error)
+	KeepConnected(ctx context.Context, opts ...grpc.CallOption) (Seaweed_KeepConnectedClient, error)
 }
 
 type seaweedClient struct {
@@ -294,10 +305,42 @@ func (x *seaweedSendHeartbeatClient) Recv() (*HeartbeatResponse, error) {
 	return m, nil
 }
 
+func (c *seaweedClient) KeepConnected(ctx context.Context, opts ...grpc.CallOption) (Seaweed_KeepConnectedClient, error) {
+	stream, err := grpc.NewClientStream(ctx, &_Seaweed_serviceDesc.Streams[1], c.cc, "/master_pb.Seaweed/KeepConnected", opts...)
+	if err != nil {
+		return nil, err
+	}
+	x := &seaweedKeepConnectedClient{stream}
+	return x, nil
+}
+
+type Seaweed_KeepConnectedClient interface {
+	Send(*Empty) error
+	Recv() (*Empty, error)
+	grpc.ClientStream
+}
+
+type seaweedKeepConnectedClient struct {
+	grpc.ClientStream
+}
+
+func (x *seaweedKeepConnectedClient) Send(m *Empty) error {
+	return x.ClientStream.SendMsg(m)
+}
+
+func (x *seaweedKeepConnectedClient) Recv() (*Empty, error) {
+	m := new(Empty)
+	if err := x.ClientStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
 // Server API for Seaweed service
 
 type SeaweedServer interface {
 	SendHeartbeat(Seaweed_SendHeartbeatServer) error
+	KeepConnected(Seaweed_KeepConnectedServer) error
 }
 
 func RegisterSeaweedServer(s *grpc.Server, srv SeaweedServer) {
@@ -330,6 +373,32 @@ func (x *seaweedSendHeartbeatServer) Recv() (*Heartbeat, error) {
 	return m, nil
 }
 
+func _Seaweed_KeepConnected_Handler(srv interface{}, stream grpc.ServerStream) error {
+	return srv.(SeaweedServer).KeepConnected(&seaweedKeepConnectedServer{stream})
+}
+
+type Seaweed_KeepConnectedServer interface {
+	Send(*Empty) error
+	Recv() (*Empty, error)
+	grpc.ServerStream
+}
+
+type seaweedKeepConnectedServer struct {
+	grpc.ServerStream
+}
+
+func (x *seaweedKeepConnectedServer) Send(m *Empty) error {
+	return x.ServerStream.SendMsg(m)
+}
+
+func (x *seaweedKeepConnectedServer) Recv() (*Empty, error) {
+	m := new(Empty)
+	if err := x.ServerStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
 var _Seaweed_serviceDesc = grpc.ServiceDesc{
 	ServiceName: "master_pb.Seaweed",
 	HandlerType: (*SeaweedServer)(nil),
@@ -341,6 +410,12 @@ var _Seaweed_serviceDesc = grpc.ServiceDesc{
 			ServerStreams: true,
 			ClientStreams: true,
 		},
+		{
+			StreamName:    "KeepConnected",
+			Handler:       _Seaweed_KeepConnected_Handler,
+			ServerStreams: true,
+			ClientStreams: true,
+		},
 	},
 	Metadata: "seaweed.proto",
 }
@@ -348,37 +423,39 @@ var _Seaweed_serviceDesc = grpc.ServiceDesc{
 func init() { proto.RegisterFile("seaweed.proto", fileDescriptor0) }
 
 var fileDescriptor0 = []byte{
-	// 504 bytes of a gzipped FileDescriptorProto
-	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0x74, 0x93, 0xdf, 0x6e, 0xd3, 0x30,
-	0x14, 0xc6, 0x49, 0x16, 0xda, 0xfa, 0x74, 0x1d, 0x9d, 0x85, 0x90, 0x05, 0x03, 0x4a, 0xb9, 0x89,
-	0x04, 0xaa, 0xd0, 0xb8, 0xe6, 0x86, 0x49, 0x88, 0x69, 0x20, 0x26, 0x17, 0xb8, 0x8d, 0xdc, 0xe4,
-	0x0c, 0x59, 0x73, 0xfe, 0xc8, 0x76, 0x47, 0xb3, 0x77, 0xe2, 0x2d, 0x78, 0x30, 0xe4, 0x93, 0xa6,
-	0x9d, 0x10, 0xdc, 0x1d, 0xff, 0xce, 0xe7, 0xf8, 0xe4, 0xfb, 0x6c, 0x98, 0x38, 0x54, 0x3f, 0x11,
-	0x8b, 0x45, 0x63, 0x6b, 0x5f, 0x73, 0x56, 0x2a, 0xe7, 0xd1, 0x66, 0xcd, 0x6a, 0xfe, 0x2b, 0x06,
-	0xf6, 0x11, 0x95, 0xf5, 0x2b, 0x54, 0x9e, 0x1f, 0x41, 0xac, 0x1b, 0x11, 0xcd, 0xa2, 0x94, 0xc9,
-	0x58, 0x37, 0x9c, 0x43, 0xd2, 0xd4, 0xd6, 0x8b, 0x78, 0x16, 0xa5, 0x13, 0x49, 0x35, 0x7f, 0x0a,
-	0xd0, 0xac, 0x57, 0x46, 0xe7, 0xd9, 0xda, 0x1a, 0x71, 0x40, 0x5a, 0xd6, 0x91, 0x6f, 0xd6, 0xf0,
-	0x14, 0xa6, 0xa5, 0xda, 0x64, 0x37, 0xb5, 0x59, 0x97, 0x98, 0xe5, 0xf5, 0xba, 0xf2, 0x22, 0xa1,
-	0xed, 0x47, 0xa5, 0xda, 0x7c, 0x27, 0x7c, 0x16, 0x28, 0x9f, 0xc1, 0x61, 0x50, 0x5e, 0x69, 0x83,
-	0xd9, 0x35, 0xb6, 0xe2, 0xfe, 0x2c, 0x4a, 0x13, 0x09, 0xa5, 0xda, 0x7c, 0xd0, 0x06, 0x2f, 0xb0,
-	0xe5, 0xcf, 0x61, 0x5c, 0x28, 0xaf, 0xb2, 0x1c, 0x2b, 0x8f, 0x56, 0x0c, 0xe8, 0x2c, 0x08, 0xe8,
-	0x8c, 0x48, 0x98, 0xcf, 0xaa, 0xfc, 0x5a, 0x0c, 0xa9, 0x43, 0x75, 0x98, 0x4f, 0x15, 0xa5, 0xae,
-	0x32, 0x9a, 0x7c, 0x44, 0x47, 0x33, 0x22, 0x97, 0x61, 0xfc, 0x77, 0x30, 0xec, 0x66, 0x73, 0x82,
-	0xcd, 0x0e, 0xd2, 0xf1, 0xe9, 0xcb, 0xc5, 0xce, 0x8d, 0x45, 0x37, 0xde, 0x79, 0x75, 0x55, 0xdb,
-	0x52, 0x79, 0x5d, 0x57, 0x9f, 0xd1, 0x39, 0xf5, 0x03, 0x65, 0xbf, 0x67, 0xee, 0xe0, 0x78, 0x67,
-	0x97, 0x44, 0xd7, 0xd4, 0x95, 0x43, 0x9e, 0xc2, 0x83, 0xae, 0xbf, 0xd4, 0xb7, 0xf8, 0x49, 0x97,
-	0xda, 0x93, 0x87, 0x89, 0xfc, 0x1b, 0xf3, 0x13, 0x60, 0x0e, 0x73, 0x8b, 0xfe, 0x02, 0x5b, 0x72,
-	0x95, 0xc9, 0x3d, 0xe0, 0x8f, 0x60, 0x60, 0x50, 0x15, 0x68, 0xb7, 0xb6, 0x6e, 0x57, 0xf3, 0xdf,
-	0x31, 0x88, 0xff, 0x8d, 0x46, 0x99, 0x15, 0x74, 0xde, 0x44, 0xc6, 0xba, 0x08, 0x9e, 0x38, 0x7d,
-	0x8b, 0xf4, 0xf5, 0x44, 0x52, 0xcd, 0x9f, 0x01, 0xe4, 0xb5, 0x31, 0x98, 0x87, 0x8d, 0xdb, 0x8f,
-	0xdf, 0x21, 0xc1, 0x33, 0x8a, 0x61, 0x1f, 0x57, 0x22, 0x59, 0x20, 0x5d, 0x52, 0x2f, 0xe0, 0xb0,
-	0x40, 0x83, 0xbe, 0x17, 0x74, 0x49, 0x8d, 0x3b, 0xd6, 0x49, 0x5e, 0x03, 0xef, 0x96, 0x45, 0xb6,
-	0x6a, 0x77, 0xc2, 0x01, 0x09, 0xa7, 0xdb, 0xce, 0xfb, 0xb6, 0x57, 0x3f, 0x01, 0x66, 0x51, 0x15,
-	0x59, 0x5d, 0x99, 0x96, 0xc2, 0x1b, 0xc9, 0x51, 0x00, 0x5f, 0x2a, 0xd3, 0xf2, 0x57, 0x70, 0x6c,
-	0xb1, 0x31, 0x3a, 0x57, 0x59, 0x63, 0x54, 0x8e, 0x25, 0x56, 0x7d, 0x8e, 0xd3, 0x6d, 0xe3, 0xb2,
-	0xe7, 0x5c, 0xc0, 0xf0, 0x06, 0xad, 0x0b, 0xbf, 0xc5, 0x48, 0xd2, 0x2f, 0xf9, 0x14, 0x0e, 0xbc,
-	0x37, 0x02, 0x88, 0x86, 0xf2, 0xf4, 0x2b, 0x0c, 0x97, 0xdd, 0x3b, 0xe0, 0xe7, 0x30, 0x59, 0x62,
-	0x55, 0xec, 0x6f, 0xfe, 0xc3, 0x3b, 0xb7, 0x60, 0x47, 0x1f, 0x9f, 0xfc, 0x8b, 0xf6, 0xb1, 0xcf,
-	0xef, 0xa5, 0xd1, 0x9b, 0x68, 0x35, 0xa0, 0x37, 0xf5, 0xf6, 0x4f, 0x00, 0x00, 0x00, 0xff, 0xff,
-	0x01, 0x14, 0xbb, 0x3a, 0x64, 0x03, 0x00, 0x00,
+	// 540 bytes of a gzipped FileDescriptorProto
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0x74, 0x93, 0xcf, 0x6e, 0xd3, 0x4e,
+	0x10, 0xc7, 0x7f, 0x76, 0xdd, 0xa4, 0x9e, 0x34, 0xfd, 0xa5, 0x2b, 0x84, 0xac, 0x52, 0x20, 0x84,
+	0x8b, 0x25, 0x50, 0x84, 0xca, 0x89, 0x03, 0x17, 0x22, 0x10, 0x55, 0x40, 0x54, 0x8e, 0xe0, 0x6a,
+	0x6d, 0xec, 0x29, 0x5a, 0x75, 0xbd, 0xb6, 0x76, 0x37, 0x25, 0xee, 0x4b, 0xf0, 0x24, 0xbc, 0x05,
+	0x0f, 0x86, 0x76, 0x36, 0x4e, 0x22, 0xfe, 0xdc, 0x66, 0x3f, 0xf3, 0x1d, 0xcf, 0x78, 0xbe, 0xbb,
+	0x30, 0x34, 0xc8, 0xbf, 0x21, 0x96, 0xd3, 0x46, 0xd7, 0xb6, 0x66, 0x71, 0xc5, 0x8d, 0x45, 0x9d,
+	0x37, 0xcb, 0xc9, 0x8f, 0x10, 0xe2, 0xf7, 0xc8, 0xb5, 0x5d, 0x22, 0xb7, 0xec, 0x04, 0x42, 0xd1,
+	0x24, 0xc1, 0x38, 0x48, 0xe3, 0x2c, 0x14, 0x0d, 0x63, 0x10, 0x35, 0xb5, 0xb6, 0x49, 0x38, 0x0e,
+	0xd2, 0x61, 0x46, 0x31, 0x7b, 0x08, 0xd0, 0xac, 0x96, 0x52, 0x14, 0xf9, 0x4a, 0xcb, 0xe4, 0x80,
+	0xb4, 0xb1, 0x27, 0x9f, 0xb5, 0x64, 0x29, 0x8c, 0x2a, 0xbe, 0xce, 0x6f, 0x6b, 0xb9, 0xaa, 0x30,
+	0x2f, 0xea, 0x95, 0xb2, 0x49, 0x44, 0xe5, 0x27, 0x15, 0x5f, 0x7f, 0x21, 0x3c, 0x73, 0x94, 0x8d,
+	0xe1, 0xd8, 0x29, 0xaf, 0x85, 0xc4, 0xfc, 0x06, 0xdb, 0xe4, 0x70, 0x1c, 0xa4, 0x51, 0x06, 0x15,
+	0x5f, 0xbf, 0x13, 0x12, 0xe7, 0xd8, 0xb2, 0xc7, 0x30, 0x28, 0xb9, 0xe5, 0x79, 0x81, 0xca, 0xa2,
+	0x4e, 0x7a, 0xd4, 0x0b, 0x1c, 0x9a, 0x11, 0x71, 0xf3, 0x69, 0x5e, 0xdc, 0x24, 0x7d, 0xca, 0x50,
+	0xec, 0xe6, 0xe3, 0x65, 0x25, 0x54, 0x4e, 0x93, 0x1f, 0x51, 0xeb, 0x98, 0xc8, 0x95, 0x1b, 0xff,
+	0x35, 0xf4, 0xfd, 0x6c, 0x26, 0x89, 0xc7, 0x07, 0xe9, 0xe0, 0xe2, 0xe9, 0x74, 0xbb, 0x8d, 0xa9,
+	0x1f, 0xef, 0x52, 0x5d, 0xd7, 0xba, 0xe2, 0x56, 0xd4, 0xea, 0x23, 0x1a, 0xc3, 0xbf, 0x62, 0xd6,
+	0xd5, 0x4c, 0x0c, 0x9c, 0x6e, 0xd7, 0x95, 0xa1, 0x69, 0x6a, 0x65, 0x90, 0xa5, 0xf0, 0xbf, 0xcf,
+	0x2f, 0xc4, 0x1d, 0x7e, 0x10, 0x95, 0xb0, 0xb4, 0xc3, 0x28, 0xfb, 0x1d, 0xb3, 0x73, 0x88, 0x0d,
+	0x16, 0x1a, 0xed, 0x1c, 0x5b, 0xda, 0x6a, 0x9c, 0xed, 0x00, 0xbb, 0x0f, 0x3d, 0x89, 0xbc, 0x44,
+	0xbd, 0x59, 0xeb, 0xe6, 0x34, 0xf9, 0x19, 0x42, 0xf2, 0xaf, 0xd1, 0xc8, 0xb3, 0x92, 0xfa, 0x0d,
+	0xb3, 0x50, 0x94, 0x6e, 0x27, 0x46, 0xdc, 0x21, 0x7d, 0x3d, 0xca, 0x28, 0x66, 0x8f, 0x00, 0x8a,
+	0x5a, 0x4a, 0x2c, 0x5c, 0xe1, 0xe6, 0xe3, 0x7b, 0xc4, 0xed, 0x8c, 0x6c, 0xd8, 0xd9, 0x15, 0x65,
+	0xb1, 0x23, 0xde, 0xa9, 0x27, 0x70, 0x5c, 0xa2, 0x44, 0xdb, 0x09, 0xbc, 0x53, 0x03, 0xcf, 0xbc,
+	0xe4, 0x39, 0x30, 0x7f, 0x2c, 0xf3, 0x65, 0xbb, 0x15, 0xf6, 0x48, 0x38, 0xda, 0x64, 0xde, 0xb4,
+	0x9d, 0xfa, 0x01, 0xc4, 0x1a, 0x79, 0x99, 0xd7, 0x4a, 0xb6, 0x64, 0xde, 0x51, 0x76, 0xe4, 0xc0,
+	0x27, 0x25, 0x5b, 0xf6, 0x0c, 0x4e, 0x35, 0x36, 0x52, 0x14, 0x3c, 0x6f, 0x24, 0x2f, 0xb0, 0x42,
+	0xd5, 0xf9, 0x38, 0xda, 0x24, 0xae, 0x3a, 0xce, 0x12, 0xe8, 0xdf, 0xa2, 0x36, 0xee, 0xb7, 0x62,
+	0x92, 0x74, 0x47, 0x36, 0x82, 0x03, 0x6b, 0x65, 0x02, 0x44, 0x5d, 0x38, 0xe9, 0xc3, 0xe1, 0xdb,
+	0xaa, 0xb1, 0xed, 0xc5, 0xf7, 0x00, 0xfa, 0x0b, 0xff, 0x22, 0xd8, 0x25, 0x0c, 0x17, 0xa8, 0xca,
+	0xdd, 0x1b, 0xb8, 0xb7, 0x77, 0x1f, 0xb6, 0xf4, 0xec, 0xfc, 0x6f, 0xb4, 0xbb, 0x00, 0x93, 0xff,
+	0xd2, 0xe0, 0x45, 0xc0, 0x5e, 0xc1, 0x70, 0x8e, 0xd8, 0xcc, 0x6a, 0xa5, 0xb0, 0xb0, 0x58, 0xb2,
+	0xd1, 0x5e, 0x11, 0x75, 0x3e, 0xfb, 0x83, 0xf8, 0xd2, 0x65, 0x8f, 0x1e, 0xe6, 0xcb, 0x5f, 0x01,
+	0x00, 0x00, 0xff, 0xff, 0x41, 0x64, 0x8a, 0xd9, 0xa9, 0x03, 0x00, 0x00,
 }

weed/pb/seaweed.proto

@@ -5,7 +5,10 @@ package master_pb;
 //////////////////////////////////////////////////
 
 service Seaweed {
-  rpc SendHeartbeat(stream Heartbeat) returns (stream HeartbeatResponse) {}
+    rpc SendHeartbeat (stream Heartbeat) returns (stream HeartbeatResponse) {
+    }
+    rpc KeepConnected (stream Empty) returns (stream Empty) {
+    }
 }
 
 //////////////////////////////////////////////////
@@ -21,6 +24,7 @@ message Heartbeat {
     uint32 admin_port = 8;
     repeated VolumeInformationMessage volumes = 9;
 }
+
 message HeartbeatResponse {
     uint64 volumeSizeLimit = 1;
     string secretKey = 2;
@@ -39,3 +43,6 @@ message VolumeInformationMessage {
     uint32 version = 9;
     uint32 ttl = 10;
 }
+
+message Empty {
+}

weed/server/common.go

@@ -12,6 +12,7 @@ import (
 	"time"
 
 	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/chrislusf/seaweedfs/weed/images"
 	"github.com/chrislusf/seaweedfs/weed/operation"
 	"github.com/chrislusf/seaweedfs/weed/security"
 	"github.com/chrislusf/seaweedfs/weed/stats"
@@ -188,3 +189,15 @@ func statsMemoryHandler(w http.ResponseWriter, r *http.Request) {
 	m["Memory"] = stats.MemStat()
 	writeJsonQuiet(w, r, http.StatusOK, m)
 }
+
+func faviconHandler(w http.ResponseWriter, r *http.Request) {
+	data, err := images.Asset("favicon/favicon.ico")
+	if err != nil {
+		glog.V(2).Infoln("favicon read error:", err)
+		return
+	}
+
+	if e := writeResponseContent("favicon.ico", "image/x-icon", bytes.NewReader(data), w, r); e != nil {
+		glog.V(2).Infoln("response write error:", e)
+	}
+}

weed/server/filer_grpc_server.go

@@ -2,108 +2,210 @@ package weed_server
 
 import (
 	"context"
-	"strconv"
+	"fmt"
+	"os"
+	"path/filepath"
+	"time"
 
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/operation"
-	"github.com/chrislusf/seaweedfs/weed/util"
 	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
-	"fmt"
 )
 
 func (fs *FilerServer) LookupDirectoryEntry(ctx context.Context, req *filer_pb.LookupDirectoryEntryRequest) (*filer_pb.LookupDirectoryEntryResponse, error) {
 
-	found, fileId, err := fs.filer.LookupDirectoryEntry(req.Directory, req.Name)
+	entry, err := fs.filer.FindEntry(filer2.FullPath(filepath.Join(req.Directory, req.Name)))
 	if err != nil {
-		return nil, err
-	}
-	if !found {
-		return nil, fmt.Errorf("%s not found under %s", req.Name, req.Directory)
+		return nil, fmt.Errorf("%s not found under %s: %v", req.Name, req.Directory, err)
 	}
 
 	return &filer_pb.LookupDirectoryEntryResponse{
 		Entry: &filer_pb.Entry{
 			Name:        req.Name,
-			IsDirectory: fileId == "",
-			FileId:      fileId,
+			IsDirectory: entry.IsDirectory(),
+			Chunks:      entry.Chunks,
 		},
 	}, nil
 }
 
 func (fs *FilerServer) ListEntries(ctx context.Context, req *filer_pb.ListEntriesRequest) (*filer_pb.ListEntriesResponse, error) {
 
-	directoryNames, err := fs.filer.ListDirectories(req.Directory)
-	if err != nil {
-		return nil, err
-	}
-	files, err := fs.filer.ListFiles(req.Directory, "", 1000)
+	entries, err := fs.filer.ListDirectoryEntries(filer2.FullPath(req.Directory), "", false, 1000)
 	if err != nil {
 		return nil, err
 	}
 
 	resp := &filer_pb.ListEntriesResponse{}
-	for _, dir := range directoryNames {
-		resp.Entries = append(resp.Entries, &filer_pb.Entry{
-			Name:        string(dir),
-			IsDirectory: true,
-		})
-	}
-	for _, fileEntry := range files {
+	for _, entry := range entries {
+
 		resp.Entries = append(resp.Entries, &filer_pb.Entry{
-			Name:        fileEntry.Name,
-			IsDirectory: false,
-			FileId:      string(fileEntry.Id),
+			Name:        entry.Name(),
+			IsDirectory: entry.IsDirectory(),
+			Chunks:      entry.Chunks,
+			Attributes: &filer_pb.FuseAttributes{
+				FileSize: entry.Size(),
+				Mtime:    entry.Mtime.Unix(),
+				Crtime:   entry.Crtime.Unix(),
+				Gid:      entry.Gid,
+				Uid:      entry.Uid,
+				FileMode: uint32(entry.Mode),
+				Mime:     entry.Mime,
+			},
 		})
 	}
 
 	return resp, nil
 }
 
-func (fs *FilerServer) GetFileAttributes(ctx context.Context, req *filer_pb.GetFileAttributesRequest) (*filer_pb.GetFileAttributesResponse, error) {
+func (fs *FilerServer) GetEntryAttributes(ctx context.Context, req *filer_pb.GetEntryAttributesRequest) (*filer_pb.GetEntryAttributesResponse, error) {
 
 	attributes := &filer_pb.FuseAttributes{}
 
-	server, err := operation.LookupFileId(fs.getMasterNode(), req.FileId)
-	if err != nil {
-		return nil, err
-	}
-	head, err := util.Head(server)
-	if err != nil {
-		return nil, err
-	}
-	attributes.FileSize, err = strconv.ParseUint(head.Get("Content-Length"), 10, 0)
+	fullpath := filer2.NewFullPath(req.ParentDir, req.Name)
+
+	entry, err := fs.filer.FindEntry(fullpath)
 	if err != nil {
-		return nil, err
+		attributes.FileSize = 0
+		return nil, fmt.Errorf("FindEntry %s: %v", fullpath, err)
 	}
 
-	return &filer_pb.GetFileAttributesResponse{
+	attributes.FileSize = entry.Size()
+	attributes.FileMode = uint32(entry.Mode)
+	attributes.Uid = entry.Uid
+	attributes.Gid = entry.Gid
+	attributes.Mtime = entry.Mtime.Unix()
+	attributes.Crtime = entry.Crtime.Unix()
+	attributes.Mime = entry.Mime
+
+	glog.V(3).Infof("GetEntryAttributes %v size %d chunks %d: %+v", fullpath, attributes.FileSize, len(entry.Chunks), attributes)
+
+	return &filer_pb.GetEntryAttributesResponse{
 		Attributes: attributes,
+		Chunks:     entry.Chunks,
 	}, nil
 }
 
-func (fs *FilerServer) GetFileContent(ctx context.Context, req *filer_pb.GetFileContentRequest) (*filer_pb.GetFileContentResponse, error) {
+func (fs *FilerServer) LookupVolume(ctx context.Context, req *filer_pb.LookupVolumeRequest) (*filer_pb.LookupVolumeResponse, error) {
 
-	server, err := operation.LookupFileId(fs.getMasterNode(), req.FileId)
+	lookupResult, err := operation.LookupVolumeIds(fs.filer.GetMaster(), req.VolumeIds)
 	if err != nil {
 		return nil, err
 	}
-	content, err := util.Get(server)
+
+	resp := &filer_pb.LookupVolumeResponse{
+		LocationsMap: make(map[string]*filer_pb.Locations),
+	}
+
+	for vid, locations := range lookupResult {
+		var locs []*filer_pb.Location
+		for _, loc := range locations.Locations {
+			locs = append(locs, &filer_pb.Location{
+				Url:       loc.Url,
+				PublicUrl: loc.PublicUrl,
+			})
+		}
+		resp.LocationsMap[vid] = &filer_pb.Locations{
+			Locations: locs,
+		}
+	}
+
+	return resp, nil
+}
+
+func (fs *FilerServer) CreateEntry(ctx context.Context, req *filer_pb.CreateEntryRequest) (resp *filer_pb.CreateEntryResponse, err error) {
+	err = fs.filer.CreateEntry(&filer2.Entry{
+		FullPath: filer2.FullPath(filepath.Join(req.Directory, req.Entry.Name)),
+		Attr: filer2.Attr{
+			Mtime:  time.Unix(req.Entry.Attributes.Mtime, 0),
+			Crtime: time.Unix(req.Entry.Attributes.Mtime, 0),
+			Mode:   os.FileMode(req.Entry.Attributes.FileMode),
+			Uid:    req.Entry.Attributes.Uid,
+			Gid:    req.Entry.Attributes.Gid,
+			Mime:   req.Entry.Attributes.Mime,
+		},
+		Chunks: req.Entry.Chunks,
+	})
+
+	if err == nil {
+	}
+
+	return &filer_pb.CreateEntryResponse{}, err
+}
+
+func (fs *FilerServer) UpdateEntry(ctx context.Context, req *filer_pb.UpdateEntryRequest) (*filer_pb.UpdateEntryResponse, error) {
+
+	fullpath := filepath.Join(req.Directory, req.Entry.Name)
+	entry, err := fs.filer.FindEntry(filer2.FullPath(fullpath))
 	if err != nil {
-		return nil, err
+		return &filer_pb.UpdateEntryResponse{}, fmt.Errorf("not found %s: %v", fullpath, err)
 	}
 
-	return &filer_pb.GetFileContentResponse{
-		Content: content,
-	}, nil
+	// remove old chunks if not included in the new ones
+	unusedChunks := filer2.FindUnusedFileChunks(entry.Chunks, req.Entry.Chunks)
+
+	chunks, garbages := filer2.CompactFileChunks(req.Entry.Chunks)
+
+	newEntry := &filer2.Entry{
+		FullPath: filer2.FullPath(filepath.Join(req.Directory, req.Entry.Name)),
+		Attr:     entry.Attr,
+		Chunks:   chunks,
+	}
+
+	glog.V(3).Infof("updating %s: %+v, chunks %d: %v => %+v, chunks %d: %v",
+		fullpath, entry.Attr, len(entry.Chunks), entry.Chunks,
+		req.Entry.Attributes, len(req.Entry.Chunks), req.Entry.Chunks)
+
+	if req.Entry.Attributes != nil {
+		if req.Entry.Attributes.Mtime != 0 {
+			newEntry.Attr.Mtime = time.Unix(req.Entry.Attributes.Mtime, 0)
+		}
+		if req.Entry.Attributes.FileMode != 0 {
+			newEntry.Attr.Mode = os.FileMode(req.Entry.Attributes.FileMode)
+		}
+		newEntry.Attr.Uid = req.Entry.Attributes.Uid
+		newEntry.Attr.Gid = req.Entry.Attributes.Gid
+		newEntry.Attr.Mime = req.Entry.Attributes.Mime
+
+	}
+
+	if err = fs.filer.UpdateEntry(newEntry); err == nil {
+		for _, garbage := range unusedChunks {
+			glog.V(0).Infof("deleting %s old chunk: %v, [%d, %d)", fullpath, garbage.FileId, garbage.Offset, garbage.Offset+int64(garbage.Size))
+			operation.DeleteFile(fs.filer.GetMaster(), garbage.FileId, fs.jwt(garbage.FileId))
+		}
+		for _, garbage := range garbages {
+			glog.V(0).Infof("deleting %s garbage chunk: %v, [%d, %d)", fullpath, garbage.FileId, garbage.Offset, garbage.Offset+int64(garbage.Size))
+			operation.DeleteFile(fs.filer.GetMaster(), garbage.FileId, fs.jwt(garbage.FileId))
+		}
+	}
+
+	return &filer_pb.UpdateEntryResponse{}, err
 }
 
 func (fs *FilerServer) DeleteEntry(ctx context.Context, req *filer_pb.DeleteEntryRequest) (resp *filer_pb.DeleteEntryResponse, err error) {
-	if req.IsDirectory {
-		err = fs.filer.DeleteDirectory(req.Directory+req.Name, false)
-	} else {
-		fid, err := fs.filer.DeleteFile(req.Directory + req.Name)
-		if err == nil && fid != "" {
-			err = operation.DeleteFile(fs.getMasterNode(), fid, fs.jwt(fid))
+	err = fs.filer.DeleteEntryMetaAndData(filer2.FullPath(filepath.Join(req.Directory, req.Name)))
+	return &filer_pb.DeleteEntryResponse{}, err
+}
+
+func (fs *FilerServer) AssignVolume(ctx context.Context, req *filer_pb.AssignVolumeRequest) (resp *filer_pb.AssignVolumeResponse, err error) {
+
+	assignResult, err := operation.Assign(fs.filer.GetMaster(), &operation.VolumeAssignRequest{
+		Count:       uint64(req.Count),
+		Replication: req.Replication,
+		Collection:  req.Collection,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("assign volume: %v", err)
 	}
+	if assignResult.Error != "" {
+		return nil, fmt.Errorf("assign volume result: %v", assignResult.Error)
 	}
-	return nil, err
+
+	return &filer_pb.AssignVolumeResponse{
+		FileId:    assignResult.Fid,
+		Count:     int32(assignResult.Count),
+		Url:       assignResult.Url,
+		PublicUrl: assignResult.PublicUrl,
+	}, err
 }

weed/server/filer_server.go

@@ -1,72 +1,38 @@
 package weed_server
 
 import (
-	"encoding/json"
-	"math/rand"
 	"net/http"
-	"os"
 	"strconv"
-	"sync"
-	"time"
-
-	"github.com/chrislusf/seaweedfs/weed/filer"
-	"github.com/chrislusf/seaweedfs/weed/filer/cassandra_store"
-	"github.com/chrislusf/seaweedfs/weed/filer/embedded_filer"
-	"github.com/chrislusf/seaweedfs/weed/filer/flat_namespace"
-	"github.com/chrislusf/seaweedfs/weed/filer/mysql_store"
-	"github.com/chrislusf/seaweedfs/weed/filer/postgres_store"
-	"github.com/chrislusf/seaweedfs/weed/filer/redis_store"
-	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	_ "github.com/chrislusf/seaweedfs/weed/filer2/cassandra"
+	_ "github.com/chrislusf/seaweedfs/weed/filer2/leveldb"
+	_ "github.com/chrislusf/seaweedfs/weed/filer2/memdb"
+	_ "github.com/chrislusf/seaweedfs/weed/filer2/mysql"
+	_ "github.com/chrislusf/seaweedfs/weed/filer2/postgres"
+	_ "github.com/chrislusf/seaweedfs/weed/filer2/redis"
 	"github.com/chrislusf/seaweedfs/weed/security"
-	"github.com/chrislusf/seaweedfs/weed/storage"
-	"github.com/chrislusf/seaweedfs/weed/util"
+	"github.com/chrislusf/seaweedfs/weed/glog"
 )
 
-type filerConf struct {
-	MysqlConf []mysql_store.MySqlConf `json:"mysql"`
-	mysql_store.ShardingConf
-	PostgresConf *postgres_store.PostgresConf `json:"postgres"`
-}
-
-func parseConfFile(confPath string) (*filerConf, error) {
-	var setting filerConf
-	configFile, err := os.Open(confPath)
-	defer configFile.Close()
-	if err != nil {
-		return nil, err
-	}
-
-	jsonParser := json.NewDecoder(configFile)
-	if err = jsonParser.Decode(&setting); err != nil {
-		return nil, err
-	}
-	return &setting, nil
-}
-
 type FilerServer struct {
 	port               string
-	master             string
-	mnLock             sync.RWMutex
+	masters            []string
 	collection         string
 	defaultReplication string
 	redirectOnRead     bool
 	disableDirListing  bool
 	secret             security.Secret
-	filer              filer.Filer
+	filer              *filer2.Filer
 	maxMB              int
-	masterNodes        *storage.MasterNodes
 }
 
-func NewFilerServer(defaultMux, readonlyMux *http.ServeMux, ip string, port int, master string, dir string, collection string,
+func NewFilerServer(defaultMux, readonlyMux *http.ServeMux, ip string, port int, masters []string, collection string,
 	replication string, redirectOnRead bool, disableDirListing bool,
-	confFile string,
 	maxMB int,
 	secret string,
-	cassandra_server string, cassandra_keyspace string,
-	redis_server string, redis_password string, redis_database int,
 ) (fs *FilerServer, err error) {
 	fs = &FilerServer{
-		master:             master,
+		masters:            masters,
 		collection:         collection,
 		defaultReplication: replication,
 		redirectOnRead:     redirectOnRead,
@@ -75,117 +41,25 @@ func NewFilerServer(defaultMux, readonlyMux *http.ServeMux, ip string, port int,
 		port:               ip + ":" + strconv.Itoa(port),
 	}
 
-	var setting *filerConf
-	if confFile != "" {
-		setting, err = parseConfFile(confFile)
-		if err != nil {
-			return nil, err
-		}
-	} else {
-		setting = new(filerConf)
+	if len(masters) == 0 {
+		glog.Fatal("master list is required!")
 	}
 
-	if setting.MysqlConf != nil && len(setting.MysqlConf) != 0 {
-		mysql_store := mysql_store.NewMysqlStore(setting.MysqlConf, setting.IsSharding, setting.ShardCount)
-		fs.filer = flat_namespace.NewFlatNamespaceFiler(master, mysql_store)
-	} else if setting.PostgresConf != nil {
-		fs.filer = postgres_store.NewPostgresStore(master, *setting.PostgresConf)
-	} else if cassandra_server != "" {
-		cassandra_store, err := cassandra_store.NewCassandraStore(cassandra_keyspace, cassandra_server)
-		if err != nil {
-			glog.Fatalf("Can not connect to cassandra server %s with keyspace %s: %v", cassandra_server, cassandra_keyspace, err)
-		}
-		fs.filer = flat_namespace.NewFlatNamespaceFiler(master, cassandra_store)
-	} else if redis_server != "" {
-		redis_store := redis_store.NewRedisStore(redis_server, redis_password, redis_database)
-		fs.filer = flat_namespace.NewFlatNamespaceFiler(master, redis_store)
-	} else {
-		if fs.filer, err = embedded_filer.NewFilerEmbedded(master, dir); err != nil {
-			glog.Fatalf("Can not start filer in dir %s : %v", dir, err)
-			return
-		}
+	fs.filer = filer2.NewFiler(masters)
 
-		defaultMux.HandleFunc("/admin/mv", fs.moveHandler)
-	}
+	go fs.filer.KeepConnectedToMaster()
 
-	defaultMux.HandleFunc("/admin/register", fs.registerHandler)
+	fs.filer.LoadConfiguration()
+
+	defaultMux.HandleFunc("/favicon.ico", faviconHandler)
 	defaultMux.HandleFunc("/", fs.filerHandler)
 	if defaultMux != readonlyMux {
 		readonlyMux.HandleFunc("/", fs.readonlyFilerHandler)
 	}
 
-	go func() {
-		connected := true
-
-		fs.masterNodes = storage.NewMasterNodes(fs.master)
-		glog.V(0).Infof("Filer server bootstraps with master %s", fs.getMasterNode())
-
-		for {
-			glog.V(4).Infof("Filer server sending to master %s", fs.getMasterNode())
-			master, err := fs.detectHealthyMaster(fs.getMasterNode())
-			if err == nil {
-				if !connected {
-					connected = true
-					if fs.getMasterNode() != master {
-						fs.setMasterNode(master)
-					}
-					glog.V(0).Infoln("Filer Server Connected with master at", master)
-				}
-			} else {
-				glog.V(1).Infof("Filer Server Failed to talk with master %s: %v", fs.getMasterNode(), err)
-				if connected {
-					connected = false
-				}
-			}
-			if connected {
-				time.Sleep(time.Duration(float32(10*1e3)*(1+rand.Float32())) * time.Millisecond)
-			} else {
-				time.Sleep(time.Duration(float32(10*1e3)*0.25) * time.Millisecond)
-			}
-		}
-	}()
-
 	return fs, nil
 }
 
 func (fs *FilerServer) jwt(fileId string) security.EncodedJwt {
 	return security.GenJwt(fs.secret, fileId)
 }
-
-func (fs *FilerServer) getMasterNode() string {
-	fs.mnLock.RLock()
-	defer fs.mnLock.RUnlock()
-	return fs.master
-}
-
-func (fs *FilerServer) setMasterNode(masterNode string) {
-	fs.mnLock.Lock()
-	defer fs.mnLock.Unlock()
-	fs.master = masterNode
-}
-
-func (fs *FilerServer) detectHealthyMaster(masterNode string) (master string, e error) {
-	if e = checkMaster(masterNode); e != nil {
-		fs.masterNodes.Reset()
-		for i := 0; i <= 3; i++ {
-			master, e = fs.masterNodes.FindMaster()
-			if e != nil {
-				continue
-			} else {
-				if e = checkMaster(master); e == nil {
-					break
-				}
-			}
-		}
-	} else {
-		master = masterNode
-	}
-	return
-}
-
-func checkMaster(masterNode string) error {
-	statUrl := "http://" + masterNode + "/stats/health"
-	glog.V(4).Infof("Connecting to %s ...", statUrl)
-	_, e := util.Get(statUrl)
-	return e
-}

weed/server/filer_server_handlers_admin.go

@@ -1,41 +0,0 @@
-package weed_server
-
-import (
-	"net/http"
-
-	"github.com/chrislusf/seaweedfs/weed/glog"
-)
-
-/*
-Move a folder or a file, with 4 Use cases:
-	mv fromDir toNewDir
-	mv fromDir toOldDir
-	mv fromFile toDir
-	mv fromFile toFile
-
-Wildcard is not supported.
-
-*/
-func (fs *FilerServer) moveHandler(w http.ResponseWriter, r *http.Request) {
-	from := r.FormValue("from")
-	to := r.FormValue("to")
-	err := fs.filer.Move(from, to)
-	if err != nil {
-		glog.V(4).Infoln("moving", from, "->", to, err.Error())
-		writeJsonError(w, r, http.StatusInternalServerError, err)
-	} else {
-		w.WriteHeader(http.StatusOK)
-	}
-}
-
-func (fs *FilerServer) registerHandler(w http.ResponseWriter, r *http.Request) {
-	path := r.FormValue("path")
-	fileId := r.FormValue("fileId")
-	err := fs.filer.CreateFile(path, fileId)
-	if err != nil {
-		glog.V(4).Infof("register %s to %s error: %v", fileId, path, err)
-		writeJsonError(w, r, http.StatusInternalServerError, err)
-	} else {
-		w.WriteHeader(http.StatusOK)
-	}
-}

weed/server/filer_server_handlers_read.go

@@ -4,112 +4,77 @@ import (
 	"io"
 	"net/http"
 	"net/url"
-	"strconv"
 	"strings"
 
-	"github.com/chrislusf/seaweedfs/weed/filer"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/operation"
-	ui "github.com/chrislusf/seaweedfs/weed/server/filer_ui"
 	"github.com/chrislusf/seaweedfs/weed/util"
-	"github.com/syndtr/goleveldb/leveldb"
+	"strconv"
+	"mime/multipart"
+	"mime"
+	"path"
 )
 
-// listDirectoryHandler lists directories and folers under a directory
-// files are sorted by name and paginated via "lastFileName" and "limit".
-// sub directories are listed on the first page, when "lastFileName"
-// is empty.
-func (fs *FilerServer) listDirectoryHandler(w http.ResponseWriter, r *http.Request) {
-	if !strings.HasSuffix(r.URL.Path, "/") {
-		return
-	}
-	limit, limit_err := strconv.Atoi(r.FormValue("limit"))
-	if limit_err != nil {
-		limit = 100
+func (fs *FilerServer) GetOrHeadHandler(w http.ResponseWriter, r *http.Request, isGetMethod bool) {
+	path := r.URL.Path
+	if strings.HasSuffix(path, "/") && len(path) > 1 {
+		path = path[:len(path)-1]
 	}
 
-	lastFileName := r.FormValue("lastFileName")
-	files, err := fs.filer.ListFiles(r.URL.Path, lastFileName, limit)
-
-	if err == leveldb.ErrNotFound {
-		glog.V(0).Infof("Error %s", err)
+	entry, err := fs.filer.FindEntry(filer2.FullPath(path))
+	if err != nil {
+		glog.V(1).Infof("Not found %s: %v", path, err)
 		w.WriteHeader(http.StatusNotFound)
 		return
 	}
 
-	directories, err2 := fs.filer.ListDirectories(r.URL.Path)
-	if err2 == leveldb.ErrNotFound {
-		glog.V(0).Infof("Error %s", err)
-		w.WriteHeader(http.StatusNotFound)
+	if entry.IsDirectory() {
+		if fs.disableDirListing {
+			w.WriteHeader(http.StatusMethodNotAllowed)
 			return
 		}
-
-	shouldDisplayLoadMore := len(files) > 0
-
-	lastFileName = ""
-	if len(files) > 0 {
-		lastFileName = files[len(files)-1].Name
-
-		files2, err3 := fs.filer.ListFiles(r.URL.Path, lastFileName, limit)
-		if err3 == leveldb.ErrNotFound {
-			glog.V(0).Infof("Error %s", err)
-			w.WriteHeader(http.StatusNotFound)
+		fs.listDirectoryHandler(w, r)
 		return
 	}
-		shouldDisplayLoadMore = len(files2) > 0
-	}
-
-	args := struct {
-		Path                  string
-		Files                 interface{}
-		Directories           interface{}
-		Limit                 int
-		LastFileName          string
-		ShouldDisplayLoadMore bool
-	}{
-		r.URL.Path,
-		files,
-		directories,
-		limit,
-		lastFileName,
-		shouldDisplayLoadMore,
-	}
 
-	if r.Header.Get("Accept") == "application/json" {
-		writeJsonQuiet(w, r, http.StatusOK, args)
-	} else {
-		ui.StatusTpl.Execute(w, args)
-	}
-}
-
-func (fs *FilerServer) GetOrHeadHandler(w http.ResponseWriter, r *http.Request, isGetMethod bool) {
-	if strings.HasSuffix(r.URL.Path, "/") {
-		if fs.disableDirListing {
-			w.WriteHeader(http.StatusMethodNotAllowed)
+	if len(entry.Chunks) == 0 {
+		glog.V(1).Infof("no file chunks for %s, attr=%+v", path, entry.Attr)
+		w.WriteHeader(http.StatusNoContent)
 		return
 	}
-		fs.listDirectoryHandler(w, r)
+
+	w.Header().Set("Accept-Ranges", "bytes")
+	if r.Method == "HEAD" {
+		w.Header().Set("Content-Length", strconv.FormatInt(int64(filer2.TotalSize(entry.Chunks)), 10))
 		return
 	}
 
-	fileId, err := fs.filer.FindFile(r.URL.Path)
-	if err == filer.ErrNotFound {
-		glog.V(3).Infoln("Not found in db", r.URL.Path)
-		w.WriteHeader(http.StatusNotFound)
+	if len(entry.Chunks) == 1 {
+		fs.handleSingleChunk(w, r, entry)
 		return
 	}
 
-	urlLocation, err := operation.LookupFileId(fs.getMasterNode(), fileId)
+	fs.handleMultipleChunks(w, r, entry)
+
+}
+
+func (fs *FilerServer) handleSingleChunk(w http.ResponseWriter, r *http.Request, entry *filer2.Entry) {
+
+	fileId := entry.Chunks[0].FileId
+
+	urlString, err := operation.LookupFileId(fs.filer.GetMaster(), fileId)
 	if err != nil {
-		glog.V(1).Infoln("operation LookupFileId %s failed, err is %s", fileId, err.Error())
+		glog.V(1).Infof("operation LookupFileId %s failed, err: %v", fileId, err)
 		w.WriteHeader(http.StatusNotFound)
 		return
 	}
-	urlString := urlLocation
+
 	if fs.redirectOnRead {
 		http.Redirect(w, r, urlString, http.StatusFound)
 		return
 	}
+
 	u, _ := url.Parse(urlString)
 	q := u.Query()
 	for key, values := range r.URL.Query() {
@@ -142,5 +107,143 @@ func (fs *FilerServer) GetOrHeadHandler(w http.ResponseWriter, r *http.Request,
 	}
 	w.WriteHeader(resp.StatusCode)
 	io.Copy(w, resp.Body)
+}
+
+func (fs *FilerServer) handleMultipleChunks(w http.ResponseWriter, r *http.Request, entry *filer2.Entry) {
+
+	mimeType := entry.Mime
+	if mimeType == "" {
+		if ext := path.Ext(entry.Name()); ext != "" {
+			mimeType = mime.TypeByExtension(ext)
+		}
+	}
+	if mimeType != "" {
+		w.Header().Set("Content-Type", mimeType)
+	}
+
+	println("mime type:", mimeType)
+
+	totalSize := int64(filer2.TotalSize(entry.Chunks))
+
+	rangeReq := r.Header.Get("Range")
+
+	if rangeReq == "" {
+		w.Header().Set("Content-Length", strconv.FormatInt(totalSize, 10))
+		if err := fs.writeContent(w, entry, 0, int(totalSize)); err != nil {
+			http.Error(w, err.Error(), http.StatusInternalServerError)
+			return
+		}
+		return
+	}
+
+	//the rest is dealing with partial content request
+	//mostly copy from src/pkg/net/http/fs.go
+	ranges, err := parseRange(rangeReq, totalSize)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusRequestedRangeNotSatisfiable)
+		return
+	}
+	if sumRangesSize(ranges) > totalSize {
+		// The total number of bytes in all the ranges
+		// is larger than the size of the file by
+		// itself, so this is probably an attack, or a
+		// dumb client.  Ignore the range request.
+		return
+	}
+	if len(ranges) == 0 {
+		return
+	}
+	if len(ranges) == 1 {
+		// RFC 2616, Section 14.16:
+		// "When an HTTP message includes the content of a single
+		// range (for example, a response to a request for a
+		// single range, or to a request for a set of ranges
+		// that overlap without any holes), this content is
+		// transmitted with a Content-Range header, and a
+		// Content-Length header showing the number of bytes
+		// actually transferred.
+		// ...
+		// A response to a request for a single range MUST NOT
+		// be sent using the multipart/byteranges media type."
+		ra := ranges[0]
+		w.Header().Set("Content-Length", strconv.FormatInt(ra.length, 10))
+		w.Header().Set("Content-Range", ra.contentRange(totalSize))
+		w.WriteHeader(http.StatusPartialContent)
+
+		err = fs.writeContent(w, entry, ra.start, int(ra.length))
+		if err != nil {
+			http.Error(w, err.Error(), http.StatusInternalServerError)
+			return
+		}
+		return
+	}
+
+	// process multiple ranges
+	for _, ra := range ranges {
+		if ra.start > totalSize {
+			http.Error(w, "Out of Range", http.StatusRequestedRangeNotSatisfiable)
+			return
+		}
+	}
+	sendSize := rangesMIMESize(ranges, mimeType, totalSize)
+	pr, pw := io.Pipe()
+	mw := multipart.NewWriter(pw)
+	w.Header().Set("Content-Type", "multipart/byteranges; boundary="+mw.Boundary())
+	sendContent := pr
+	defer pr.Close() // cause writing goroutine to fail and exit if CopyN doesn't finish.
+	go func() {
+		for _, ra := range ranges {
+			part, e := mw.CreatePart(ra.mimeHeader(mimeType, totalSize))
+			if e != nil {
+				pw.CloseWithError(e)
+				return
+			}
+			if e = fs.writeContent(part, entry, ra.start, int(ra.length)); e != nil {
+				pw.CloseWithError(e)
+				return
+			}
+		}
+		mw.Close()
+		pw.Close()
+	}()
+	if w.Header().Get("Content-Encoding") == "" {
+		w.Header().Set("Content-Length", strconv.FormatInt(sendSize, 10))
+	}
+	w.WriteHeader(http.StatusPartialContent)
+	if _, err := io.CopyN(w, sendContent, sendSize); err != nil {
+		http.Error(w, "Internal Error", http.StatusInternalServerError)
+		return
+	}
+
+}
+
+func (fs *FilerServer) writeContent(w io.Writer, entry *filer2.Entry, offset int64, size int) error {
+
+	chunkViews := filer2.ViewFromChunks(entry.Chunks, offset, size)
+
+	fileId2Url := make(map[string]string)
+
+	for _, chunkView := range chunkViews {
+
+		urlString, err := operation.LookupFileId(fs.filer.GetMaster(), chunkView.FileId)
+		if err != nil {
+			glog.V(1).Infof("operation LookupFileId %s failed, err: %v", chunkView.FileId, err)
+			return err
+		}
+		fileId2Url[chunkView.FileId] = urlString
+	}
+
+	for _, chunkView := range chunkViews {
+		urlString := fileId2Url[chunkView.FileId]
+		_, err := util.ReadUrlAsStream(urlString, chunkView.Offset, int(chunkView.Size), func(data []byte) {
+			w.Write(data)
+		})
+		if err != nil {
+			glog.V(1).Infof("read %s failed, err: %v", chunkView.FileId, err)
+			return err
+		}
+	}
+
+	return nil
 
 }

weed/server/filer_server_handlers_read_dir.go

@@ -0,0 +1,70 @@
+package weed_server
+
+import (
+	"net/http"
+	"strconv"
+	"strings"
+
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	ui "github.com/chrislusf/seaweedfs/weed/server/filer_ui"
+)
+
+// listDirectoryHandler lists directories and folers under a directory
+// files are sorted by name and paginated via "lastFileName" and "limit".
+// sub directories are listed on the first page, when "lastFileName"
+// is empty.
+func (fs *FilerServer) listDirectoryHandler(w http.ResponseWriter, r *http.Request) {
+	path := r.URL.Path
+	if strings.HasSuffix(path, "/") && len(path) > 1 {
+		path = path[:len(path)-1]
+	}
+
+	limit, limit_err := strconv.Atoi(r.FormValue("limit"))
+	if limit_err != nil {
+		limit = 100
+	}
+
+	lastFileName := r.FormValue("lastFileName")
+
+	entries, err := fs.filer.ListDirectoryEntries(filer2.FullPath(path), lastFileName, false, limit)
+
+	if err != nil {
+		glog.V(0).Infof("listDirectory %s %s $d: %s", path, lastFileName, limit, err)
+		w.WriteHeader(http.StatusNotFound)
+		return
+	}
+
+	shouldDisplayLoadMore := len(entries) == limit
+	if path == "/" {
+		path = ""
+	}
+
+	if len(entries) > 0 {
+		lastFileName = entries[len(entries)-1].Name()
+	}
+
+	glog.V(4).Infof("listDirectory %s, last file %s, limit %d: %d items", path, lastFileName, limit, len(entries))
+
+	args := struct {
+		Path                  string
+		Breadcrumbs           []ui.Breadcrumb
+		Entries               interface{}
+		Limit                 int
+		LastFileName          string
+		ShouldDisplayLoadMore bool
+	}{
+		path,
+		ui.ToBreadcrumb(path),
+		entries,
+		limit,
+		lastFileName,
+		shouldDisplayLoadMore,
+	}
+
+	if r.Header.Get("Accept") == "application/json" {
+		writeJsonQuiet(w, r, http.StatusOK, args)
+	} else {
+		ui.StatusTpl.Execute(w, args)
+	}
+}

weed/server/filer_server_handlers_write.go

@@ -1,26 +1,19 @@
 package weed_server
 
 import (
-	"bytes"
-	"crypto/md5"
-	"encoding/base64"
 	"encoding/json"
 	"errors"
-	"fmt"
-	"io"
 	"io/ioutil"
-	"mime/multipart"
 	"net/http"
-	"net/textproto"
 	"net/url"
-	"path"
 	"strconv"
 	"strings"
+	"time"
 
-	"github.com/chrislusf/seaweedfs/weed/filer"
+	"github.com/chrislusf/seaweedfs/weed/filer2"
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/operation"
-	"github.com/chrislusf/seaweedfs/weed/storage"
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
 	"github.com/chrislusf/seaweedfs/weed/util"
 )
 
@@ -32,58 +25,18 @@ type FilerPostResult struct {
 	Url   string `json:"url,omitempty"`
 }
 
-var quoteEscaper = strings.NewReplacer("\\", "\\\\", `"`, "\\\"")
-
-func escapeQuotes(s string) string {
-	return quoteEscaper.Replace(s)
-}
-
-func createFormFile(writer *multipart.Writer, fieldname, filename, mime string) (io.Writer, error) {
-	h := make(textproto.MIMEHeader)
-	h.Set("Content-Disposition",
-		fmt.Sprintf(`form-data; name="%s"; filename="%s"`,
-			escapeQuotes(fieldname), escapeQuotes(filename)))
-	if len(mime) == 0 {
-		mime = "application/octet-stream"
-	}
-	h.Set("Content-Type", mime)
-	return writer.CreatePart(h)
-}
-
-func makeFormData(filename, mimeType string, content io.Reader) (formData io.Reader, contentType string, err error) {
-	buf := new(bytes.Buffer)
-	writer := multipart.NewWriter(buf)
-	defer writer.Close()
-
-	part, err := createFormFile(writer, "file", filename, mimeType)
-	if err != nil {
-		glog.V(0).Infoln(err)
-		return
-	}
-	_, err = io.Copy(part, content)
-	if err != nil {
-		glog.V(0).Infoln(err)
-		return
-	}
-
-	formData = buf
-	contentType = writer.FormDataContentType()
-
-	return
-}
-
 func (fs *FilerServer) queryFileInfoByPath(w http.ResponseWriter, r *http.Request, path string) (fileId, urlLocation string, err error) {
-	if fileId, err = fs.filer.FindFile(path); err != nil && err != filer.ErrNotFound {
+	var entry *filer2.Entry
+	if entry, err = fs.filer.FindEntry(filer2.FullPath(path)); err != nil {
 		glog.V(0).Infoln("failing to find path in filer store", path, err.Error())
 		writeJsonError(w, r, http.StatusInternalServerError, err)
-	} else if fileId != "" && err == nil {
-		urlLocation, err = operation.LookupFileId(fs.getMasterNode(), fileId)
+	} else {
+		fileId = entry.Chunks[0].FileId
+		urlLocation, err = operation.LookupFileId(fs.filer.GetMaster(), fileId)
 		if err != nil {
-			glog.V(1).Infoln("operation LookupFileId %s failed, err is %s", fileId, err.Error())
+			glog.V(1).Infof("operation LookupFileId %s failed, err is %s", fileId, err.Error())
 			w.WriteHeader(http.StatusNotFound)
 		}
-	} else if fileId == "" && err == filer.ErrNotFound {
-		w.WriteHeader(http.StatusNotFound)
 	}
 	return
 }
@@ -95,7 +48,7 @@ func (fs *FilerServer) assignNewFileInfo(w http.ResponseWriter, r *http.Request,
 		Collection:  collection,
 		Ttl:         r.URL.Query().Get("ttl"),
 	}
-	assignResult, ae := operation.Assign(fs.getMasterNode(), ar)
+	assignResult, ae := operation.Assign(fs.filer.GetMaster(), ar)
 	if ae != nil {
 		glog.V(0).Infoln("failing to assign a file id", ae.Error())
 		writeJsonError(w, r, http.StatusInternalServerError, ae)
@@ -107,117 +60,6 @@ func (fs *FilerServer) assignNewFileInfo(w http.ResponseWriter, r *http.Request,
 	return
 }
 
-func (fs *FilerServer) multipartUploadAnalyzer(w http.ResponseWriter, r *http.Request, replication, collection string) (fileId, urlLocation string, err error) {
-	//Default handle way for http multipart
-	if r.Method == "PUT" {
-		buf, _ := ioutil.ReadAll(r.Body)
-		r.Body = ioutil.NopCloser(bytes.NewBuffer(buf))
-		fileName, _, _, _, _, _, _, _, pe := storage.ParseUpload(r)
-		if pe != nil {
-			glog.V(0).Infoln("failing to parse post body", pe.Error())
-			writeJsonError(w, r, http.StatusInternalServerError, pe)
-			err = pe
-			return
-		}
-		//reconstruct http request body for following new request to volume server
-		r.Body = ioutil.NopCloser(bytes.NewBuffer(buf))
-
-		path := r.URL.Path
-		if strings.HasSuffix(path, "/") {
-			if fileName != "" {
-				path += fileName
-			}
-		}
-		fileId, urlLocation, err = fs.queryFileInfoByPath(w, r, path)
-	} else {
-		fileId, urlLocation, err = fs.assignNewFileInfo(w, r, replication, collection)
-	}
-	return
-}
-
-func multipartHttpBodyBuilder(w http.ResponseWriter, r *http.Request, fileName string) (err error) {
-	body, contentType, te := makeFormData(fileName, r.Header.Get("Content-Type"), r.Body)
-	if te != nil {
-		glog.V(0).Infoln("S3 protocol to raw seaweed protocol failed", te.Error())
-		writeJsonError(w, r, http.StatusInternalServerError, te)
-		err = te
-		return
-	}
-
-	if body != nil {
-		switch v := body.(type) {
-		case *bytes.Buffer:
-			r.ContentLength = int64(v.Len())
-		case *bytes.Reader:
-			r.ContentLength = int64(v.Len())
-		case *strings.Reader:
-			r.ContentLength = int64(v.Len())
-		}
-	}
-
-	r.Header.Set("Content-Type", contentType)
-	rc, ok := body.(io.ReadCloser)
-	if !ok && body != nil {
-		rc = ioutil.NopCloser(body)
-	}
-	r.Body = rc
-	return
-}
-
-func checkContentMD5(w http.ResponseWriter, r *http.Request) (err error) {
-	if contentMD5 := r.Header.Get("Content-MD5"); contentMD5 != "" {
-		buf, _ := ioutil.ReadAll(r.Body)
-		//checkMD5
-		sum := md5.Sum(buf)
-		fileDataMD5 := base64.StdEncoding.EncodeToString(sum[0:len(sum)])
-		if strings.ToLower(fileDataMD5) != strings.ToLower(contentMD5) {
-			glog.V(0).Infof("fileDataMD5 [%s] is not equal to Content-MD5 [%s]", fileDataMD5, contentMD5)
-			err = fmt.Errorf("MD5 check failed")
-			writeJsonError(w, r, http.StatusNotAcceptable, err)
-			return
-		}
-		//reconstruct http request body for following new request to volume server
-		r.Body = ioutil.NopCloser(bytes.NewBuffer(buf))
-	}
-	return
-}
-
-func (fs *FilerServer) monolithicUploadAnalyzer(w http.ResponseWriter, r *http.Request, replication, collection string) (fileId, urlLocation string, err error) {
-	/*
-		Amazon S3 ref link:[http://docs.aws.amazon.com/AmazonS3/latest/API/Welcome.html]
-		There is a long way to provide a completely compatibility against all Amazon S3 API, I just made
-		a simple data stream adapter between S3 PUT API and seaweedfs's volume storage Write API
-		1. The request url format should be http://$host:$port/$bucketName/$objectName
-		2. bucketName will be mapped to seaweedfs's collection name
-		3. You could customize and make your enhancement.
-	*/
-	lastPos := strings.LastIndex(r.URL.Path, "/")
-	if lastPos == -1 || lastPos == 0 || lastPos == len(r.URL.Path)-1 {
-		glog.V(0).Infoln("URL Path [%s] is invalid, could not retrieve file name", r.URL.Path)
-		err = fmt.Errorf("URL Path is invalid")
-		writeJsonError(w, r, http.StatusInternalServerError, err)
-		return
-	}
-
-	if err = checkContentMD5(w, r); err != nil {
-		return
-	}
-
-	fileName := r.URL.Path[lastPos+1:]
-	if err = multipartHttpBodyBuilder(w, r, fileName); err != nil {
-		return
-	}
-
-	secondPos := strings.Index(r.URL.Path[1:], "/") + 1
-	collection = r.URL.Path[1:secondPos]
-	path := r.URL.Path
-
-	if fileId, urlLocation, err = fs.queryFileInfoByPath(w, r, path); err == nil && fileId == "" {
-		fileId, urlLocation, err = fs.assignNewFileInfo(w, r, replication, collection)
-	}
-	return
-}
-
 func (fs *FilerServer) PostHandler(w http.ResponseWriter, r *http.Request) {
 
 	query := r.URL.Query()
@@ -303,7 +145,7 @@ func (fs *FilerServer) PostHandler(w http.ResponseWriter, r *http.Request) {
 		if ret.Name != "" {
 			path += ret.Name
 		} else {
-			operation.DeleteFile(fs.getMasterNode(), fileId, fs.jwt(fileId)) //clean up
+			operation.DeleteFile(fs.filer.GetMaster(), fileId, fs.jwt(fileId)) //clean up
 			glog.V(0).Infoln("Can not to write to folder", path, "without a file name!")
 			writeJsonError(w, r, http.StatusInternalServerError,
 				errors.New("Can not to write to folder "+path+" without a file name"))
@@ -313,16 +155,28 @@ func (fs *FilerServer) PostHandler(w http.ResponseWriter, r *http.Request) {
 
 	// also delete the old fid unless PUT operation
 	if r.Method != "PUT" {
-		if oldFid, err := fs.filer.FindFile(path); err == nil {
-			operation.DeleteFile(fs.getMasterNode(), oldFid, fs.jwt(oldFid))
-		} else if err != nil && err != filer.ErrNotFound {
+		if entry, err := fs.filer.FindEntry(filer2.FullPath(path)); err == nil {
+			oldFid := entry.Chunks[0].FileId
+			operation.DeleteFile(fs.filer.GetMaster(), oldFid, fs.jwt(oldFid))
+		} else if err != nil && err != filer2.ErrNotFound {
 			glog.V(0).Infof("error %v occur when finding %s in filer store", err, path)
 		}
 	}
 
 	glog.V(4).Infoln("saving", path, "=>", fileId)
-	if db_err := fs.filer.CreateFile(path, fileId); db_err != nil {
-		operation.DeleteFile(fs.getMasterNode(), fileId, fs.jwt(fileId)) //clean up
+	entry := &filer2.Entry{
+		FullPath: filer2.FullPath(path),
+		Attr: filer2.Attr{
+			Mode: 0660,
+		},
+		Chunks: []*filer_pb.FileChunk{{
+			FileId: fileId,
+			Size:   uint64(r.ContentLength),
+			Mtime:  time.Now().UnixNano(),
+		}},
+	}
+	if db_err := fs.filer.CreateEntry(entry); db_err != nil {
+		operation.DeleteFile(fs.filer.GetMaster(), fileId, fs.jwt(fileId)) //clean up
 		glog.V(0).Infof("failing to write %s to filer server : %v", path, db_err)
 		writeJsonError(w, r, http.StatusInternalServerError, db_err)
 		return
@@ -338,217 +192,15 @@ func (fs *FilerServer) PostHandler(w http.ResponseWriter, r *http.Request) {
 	writeJsonQuiet(w, r, http.StatusCreated, reply)
 }
 
-func (fs *FilerServer) autoChunk(w http.ResponseWriter, r *http.Request, replication string, collection string) bool {
-	if r.Method != "POST" {
-		glog.V(4).Infoln("AutoChunking not supported for method", r.Method)
-		return false
-	}
-
-	// autoChunking can be set at the command-line level or as a query param. Query param overrides command-line
-	query := r.URL.Query()
-
-	parsedMaxMB, _ := strconv.ParseInt(query.Get("maxMB"), 10, 32)
-	maxMB := int32(parsedMaxMB)
-	if maxMB <= 0 && fs.maxMB > 0 {
-		maxMB = int32(fs.maxMB)
-	}
-	if maxMB <= 0 {
-		glog.V(4).Infoln("AutoChunking not enabled")
-		return false
-	}
-	glog.V(4).Infoln("AutoChunking level set to", maxMB, "(MB)")
-
-	chunkSize := 1024 * 1024 * maxMB
-
-	contentLength := int64(0)
-	if contentLengthHeader := r.Header["Content-Length"]; len(contentLengthHeader) == 1 {
-		contentLength, _ = strconv.ParseInt(contentLengthHeader[0], 10, 64)
-		if contentLength <= int64(chunkSize) {
-			glog.V(4).Infoln("Content-Length of", contentLength, "is less than the chunk size of", chunkSize, "so autoChunking will be skipped.")
-			return false
-		}
-	}
-
-	if contentLength <= 0 {
-		glog.V(4).Infoln("Content-Length value is missing or unexpected so autoChunking will be skipped.")
-		return false
-	}
+// curl -X DELETE http://localhost:8888/path/to
+func (fs *FilerServer) DeleteHandler(w http.ResponseWriter, r *http.Request) {
 
-	reply, err := fs.doAutoChunk(w, r, contentLength, chunkSize, replication, collection)
+	err := fs.filer.DeleteEntryMetaAndData(filer2.FullPath(r.URL.Path))
 	if err != nil {
+		glog.V(4).Infoln("deleting", r.URL.Path, ":", err.Error())
 		writeJsonError(w, r, http.StatusInternalServerError, err)
-	} else if reply != nil {
-		writeJsonQuiet(w, r, http.StatusCreated, reply)
-	}
-	return true
-}
-
-func (fs *FilerServer) doAutoChunk(w http.ResponseWriter, r *http.Request, contentLength int64, chunkSize int32, replication string, collection string) (filerResult *FilerPostResult, replyerr error) {
-
-	multipartReader, multipartReaderErr := r.MultipartReader()
-	if multipartReaderErr != nil {
-		return nil, multipartReaderErr
-	}
-
-	part1, part1Err := multipartReader.NextPart()
-	if part1Err != nil {
-		return nil, part1Err
-	}
-
-	fileName := part1.FileName()
-	if fileName != "" {
-		fileName = path.Base(fileName)
-	}
-
-	chunks := (int64(contentLength) / int64(chunkSize)) + 1
-	cm := operation.ChunkManifest{
-		Name:   fileName,
-		Size:   0, // don't know yet
-		Mime:   "application/octet-stream",
-		Chunks: make([]*operation.ChunkInfo, 0, chunks),
-	}
-
-	totalBytesRead := int64(0)
-	tmpBufferSize := int32(1024 * 1024)
-	tmpBuffer := bytes.NewBuffer(make([]byte, 0, tmpBufferSize))
-	chunkBuf := make([]byte, chunkSize+tmpBufferSize, chunkSize+tmpBufferSize) // chunk size plus a little overflow
-	chunkBufOffset := int32(0)
-	chunkOffset := int64(0)
-	writtenChunks := 0
-
-	filerResult = &FilerPostResult{
-		Name: fileName,
-	}
-
-	for totalBytesRead < contentLength {
-		tmpBuffer.Reset()
-		bytesRead, readErr := io.CopyN(tmpBuffer, part1, int64(tmpBufferSize))
-		readFully := readErr != nil && readErr == io.EOF
-		tmpBuf := tmpBuffer.Bytes()
-		bytesToCopy := tmpBuf[0:int(bytesRead)]
-
-		copy(chunkBuf[chunkBufOffset:chunkBufOffset+int32(bytesRead)], bytesToCopy)
-		chunkBufOffset = chunkBufOffset + int32(bytesRead)
-
-		if chunkBufOffset >= chunkSize || readFully || (chunkBufOffset > 0 && bytesRead == 0) {
-			writtenChunks = writtenChunks + 1
-			fileId, urlLocation, assignErr := fs.assignNewFileInfo(w, r, replication, collection)
-			if assignErr != nil {
-				return nil, assignErr
-			}
-
-			// upload the chunk to the volume server
-			chunkName := fileName + "_chunk_" + strconv.FormatInt(int64(cm.Chunks.Len()+1), 10)
-			uploadErr := fs.doUpload(urlLocation, w, r, chunkBuf[0:chunkBufOffset], chunkName, "application/octet-stream", fileId)
-			if uploadErr != nil {
-				return nil, uploadErr
-			}
-
-			// Save to chunk manifest structure
-			cm.Chunks = append(cm.Chunks,
-				&operation.ChunkInfo{
-					Offset: chunkOffset,
-					Size:   int64(chunkBufOffset),
-					Fid:    fileId,
-				},
-			)
-
-			// reset variables for the next chunk
-			chunkBufOffset = 0
-			chunkOffset = totalBytesRead + int64(bytesRead)
-		}
-
-		totalBytesRead = totalBytesRead + int64(bytesRead)
-
-		if bytesRead == 0 || readFully {
-			break
-		}
-
-		if readErr != nil {
-			return nil, readErr
-		}
-	}
-
-	cm.Size = totalBytesRead
-	manifestBuf, marshalErr := cm.Marshal()
-	if marshalErr != nil {
-		return nil, marshalErr
-	}
-
-	manifestStr := string(manifestBuf)
-	glog.V(4).Infoln("Generated chunk manifest: ", manifestStr)
-
-	manifestFileId, manifestUrlLocation, manifestAssignmentErr := fs.assignNewFileInfo(w, r, replication, collection)
-	if manifestAssignmentErr != nil {
-		return nil, manifestAssignmentErr
-	}
-	glog.V(4).Infoln("Manifest uploaded to:", manifestUrlLocation, "Fid:", manifestFileId)
-	filerResult.Fid = manifestFileId
-
-	u, _ := url.Parse(manifestUrlLocation)
-	q := u.Query()
-	q.Set("cm", "true")
-	u.RawQuery = q.Encode()
-
-	manifestUploadErr := fs.doUpload(u.String(), w, r, manifestBuf, fileName+"_manifest", "application/json", manifestFileId)
-	if manifestUploadErr != nil {
-		return nil, manifestUploadErr
-	}
-
-	path := r.URL.Path
-	// also delete the old fid unless PUT operation
-	if r.Method != "PUT" {
-		if oldFid, err := fs.filer.FindFile(path); err == nil {
-			operation.DeleteFile(fs.getMasterNode(), oldFid, fs.jwt(oldFid))
-		} else if err != nil && err != filer.ErrNotFound {
-			glog.V(0).Infof("error %v occur when finding %s in filer store", err, path)
-		}
-	}
-
-	glog.V(4).Infoln("saving", path, "=>", manifestFileId)
-	if db_err := fs.filer.CreateFile(path, manifestFileId); db_err != nil {
-		replyerr = db_err
-		filerResult.Error = db_err.Error()
-		operation.DeleteFile(fs.getMasterNode(), manifestFileId, fs.jwt(manifestFileId)) //clean up
-		glog.V(0).Infof("failing to write %s to filer server : %v", path, db_err)
 		return
 	}
 
-	return
-}
-
-func (fs *FilerServer) doUpload(urlLocation string, w http.ResponseWriter, r *http.Request, chunkBuf []byte, fileName string, contentType string, fileId string) (err error) {
-	err = nil
-
-	ioReader := ioutil.NopCloser(bytes.NewBuffer(chunkBuf))
-	uploadResult, uploadError := operation.Upload(urlLocation, fileName, ioReader, false, contentType, nil, fs.jwt(fileId))
-	if uploadResult != nil {
-		glog.V(0).Infoln("Chunk upload result. Name:", uploadResult.Name, "Fid:", fileId, "Size:", uploadResult.Size)
-	}
-	if uploadError != nil {
-		err = uploadError
-	}
-	return
-}
-
-// curl -X DELETE http://localhost:8888/path/to
-// curl -X DELETE http://localhost:8888/path/to/?recursive=true
-func (fs *FilerServer) DeleteHandler(w http.ResponseWriter, r *http.Request) {
-	var err error
-	var fid string
-	if strings.HasSuffix(r.URL.Path, "/") {
-		isRecursive := r.FormValue("recursive") == "true"
-		err = fs.filer.DeleteDirectory(r.URL.Path, isRecursive)
-	} else {
-		fid, err = fs.filer.DeleteFile(r.URL.Path)
-		if err == nil && fid != "" {
-			err = operation.DeleteFile(fs.getMasterNode(), fid, fs.jwt(fid))
-		}
-	}
-	if err == nil {
 	writeJsonQuiet(w, r, http.StatusAccepted, map[string]string{"error": ""})
-	} else {
-		glog.V(4).Infoln("deleting", r.URL.Path, ":", err.Error())
-		writeJsonError(w, r, http.StatusInternalServerError, err)
-	}
 }

weed/server/filer_server_handlers_write_autochunk.go

@@ -0,0 +1,189 @@
+package weed_server
+
+import (
+	"bytes"
+	"io"
+	"io/ioutil"
+	"net/http"
+	"path"
+	"strconv"
+	"time"
+
+	"github.com/chrislusf/seaweedfs/weed/filer2"
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/chrislusf/seaweedfs/weed/operation"
+	"github.com/chrislusf/seaweedfs/weed/pb/filer_pb"
+)
+
+func (fs *FilerServer) autoChunk(w http.ResponseWriter, r *http.Request, replication string, collection string) bool {
+	if r.Method != "POST" {
+		glog.V(4).Infoln("AutoChunking not supported for method", r.Method)
+		return false
+	}
+
+	// autoChunking can be set at the command-line level or as a query param. Query param overrides command-line
+	query := r.URL.Query()
+
+	parsedMaxMB, _ := strconv.ParseInt(query.Get("maxMB"), 10, 32)
+	maxMB := int32(parsedMaxMB)
+	if maxMB <= 0 && fs.maxMB > 0 {
+		maxMB = int32(fs.maxMB)
+	}
+	if maxMB <= 0 {
+		glog.V(4).Infoln("AutoChunking not enabled")
+		return false
+	}
+	glog.V(4).Infoln("AutoChunking level set to", maxMB, "(MB)")
+
+	chunkSize := 1024 * 1024 * maxMB
+
+	contentLength := int64(0)
+	if contentLengthHeader := r.Header["Content-Length"]; len(contentLengthHeader) == 1 {
+		contentLength, _ = strconv.ParseInt(contentLengthHeader[0], 10, 64)
+		if contentLength <= int64(chunkSize) {
+			glog.V(4).Infoln("Content-Length of", contentLength, "is less than the chunk size of", chunkSize, "so autoChunking will be skipped.")
+			return false
+		}
+	}
+
+	if contentLength <= 0 {
+		glog.V(4).Infoln("Content-Length value is missing or unexpected so autoChunking will be skipped.")
+		return false
+	}
+
+	reply, err := fs.doAutoChunk(w, r, contentLength, chunkSize, replication, collection)
+	if err != nil {
+		writeJsonError(w, r, http.StatusInternalServerError, err)
+	} else if reply != nil {
+		writeJsonQuiet(w, r, http.StatusCreated, reply)
+	}
+	return true
+}
+
+func (fs *FilerServer) doAutoChunk(w http.ResponseWriter, r *http.Request, contentLength int64, chunkSize int32, replication string, collection string) (filerResult *FilerPostResult, replyerr error) {
+
+	multipartReader, multipartReaderErr := r.MultipartReader()
+	if multipartReaderErr != nil {
+		return nil, multipartReaderErr
+	}
+
+	part1, part1Err := multipartReader.NextPart()
+	if part1Err != nil {
+		return nil, part1Err
+	}
+
+	fileName := part1.FileName()
+	if fileName != "" {
+		fileName = path.Base(fileName)
+	}
+
+	var fileChunks []*filer_pb.FileChunk
+
+	totalBytesRead := int64(0)
+	tmpBufferSize := int32(1024 * 1024)
+	tmpBuffer := bytes.NewBuffer(make([]byte, 0, tmpBufferSize))
+	chunkBuf := make([]byte, chunkSize+tmpBufferSize, chunkSize+tmpBufferSize) // chunk size plus a little overflow
+	chunkBufOffset := int32(0)
+	chunkOffset := int64(0)
+	writtenChunks := 0
+
+	filerResult = &FilerPostResult{
+		Name: fileName,
+	}
+
+	for totalBytesRead < contentLength {
+		tmpBuffer.Reset()
+		bytesRead, readErr := io.CopyN(tmpBuffer, part1, int64(tmpBufferSize))
+		readFully := readErr != nil && readErr == io.EOF
+		tmpBuf := tmpBuffer.Bytes()
+		bytesToCopy := tmpBuf[0:int(bytesRead)]
+
+		copy(chunkBuf[chunkBufOffset:chunkBufOffset+int32(bytesRead)], bytesToCopy)
+		chunkBufOffset = chunkBufOffset + int32(bytesRead)
+
+		if chunkBufOffset >= chunkSize || readFully || (chunkBufOffset > 0 && bytesRead == 0) {
+			writtenChunks = writtenChunks + 1
+			fileId, urlLocation, assignErr := fs.assignNewFileInfo(w, r, replication, collection)
+			if assignErr != nil {
+				return nil, assignErr
+			}
+
+			// upload the chunk to the volume server
+			chunkName := fileName + "_chunk_" + strconv.FormatInt(int64(len(fileChunks)+1), 10)
+			uploadErr := fs.doUpload(urlLocation, w, r, chunkBuf[0:chunkBufOffset], chunkName, "application/octet-stream", fileId)
+			if uploadErr != nil {
+				return nil, uploadErr
+			}
+
+			// Save to chunk manifest structure
+			fileChunks = append(fileChunks,
+				&filer_pb.FileChunk{
+					FileId: fileId,
+					Offset: chunkOffset,
+					Size:   uint64(chunkBufOffset),
+					Mtime:  time.Now().UnixNano(),
+				},
+			)
+
+			// reset variables for the next chunk
+			chunkBufOffset = 0
+			chunkOffset = totalBytesRead + int64(bytesRead)
+		}
+
+		totalBytesRead = totalBytesRead + int64(bytesRead)
+
+		if bytesRead == 0 || readFully {
+			break
+		}
+
+		if readErr != nil {
+			return nil, readErr
+		}
+	}
+
+	path := r.URL.Path
+	// also delete the old fid unless PUT operation
+	if r.Method != "PUT" {
+		if entry, err := fs.filer.FindEntry(filer2.FullPath(path)); err == nil {
+			for _, chunk := range entry.Chunks {
+				oldFid := chunk.FileId
+				operation.DeleteFile(fs.filer.GetMaster(), oldFid, fs.jwt(oldFid))
+			}
+		} else if err != nil {
+			glog.V(0).Infof("error %v occur when finding %s in filer store", err, path)
+		}
+	}
+
+	glog.V(4).Infoln("saving", path)
+	entry := &filer2.Entry{
+		FullPath: filer2.FullPath(path),
+		Attr: filer2.Attr{
+			Mtime:  time.Now(),
+			Crtime: time.Now(),
+			Mode:   0660,
+		},
+		Chunks: fileChunks,
+	}
+	if db_err := fs.filer.CreateEntry(entry); db_err != nil {
+		replyerr = db_err
+		filerResult.Error = db_err.Error()
+		glog.V(0).Infof("failing to write %s to filer server : %v", path, db_err)
+		return
+	}
+
+	return
+}
+
+func (fs *FilerServer) doUpload(urlLocation string, w http.ResponseWriter, r *http.Request, chunkBuf []byte, fileName string, contentType string, fileId string) (err error) {
+	err = nil
+
+	ioReader := ioutil.NopCloser(bytes.NewBuffer(chunkBuf))
+	uploadResult, uploadError := operation.Upload(urlLocation, fileName, ioReader, false, contentType, nil, fs.jwt(fileId))
+	if uploadResult != nil {
+		glog.V(0).Infoln("Chunk upload result. Name:", uploadResult.Name, "Fid:", fileId, "Size:", uploadResult.Size)
+	}
+	if uploadError != nil {
+		err = uploadError
+	}
+	return
+}

weed/server/filer_server_handlers_write_monopart.go

@@ -0,0 +1,139 @@
+package weed_server
+
+import (
+	"bytes"
+	"crypto/md5"
+	"encoding/base64"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"mime/multipart"
+	"net/http"
+	"net/textproto"
+	"strings"
+
+	"github.com/chrislusf/seaweedfs/weed/glog"
+)
+
+var quoteEscaper = strings.NewReplacer("\\", "\\\\", `"`, "\\\"")
+
+func escapeQuotes(s string) string {
+	return quoteEscaper.Replace(s)
+}
+
+func createFormFile(writer *multipart.Writer, fieldname, filename, mime string) (io.Writer, error) {
+	h := make(textproto.MIMEHeader)
+	h.Set("Content-Disposition",
+		fmt.Sprintf(`form-data; name="%s"; filename="%s"`,
+			escapeQuotes(fieldname), escapeQuotes(filename)))
+	if len(mime) == 0 {
+		mime = "application/octet-stream"
+	}
+	h.Set("Content-Type", mime)
+	return writer.CreatePart(h)
+}
+
+func makeFormData(filename, mimeType string, content io.Reader) (formData io.Reader, contentType string, err error) {
+	buf := new(bytes.Buffer)
+	writer := multipart.NewWriter(buf)
+	defer writer.Close()
+
+	part, err := createFormFile(writer, "file", filename, mimeType)
+	if err != nil {
+		glog.V(0).Infoln(err)
+		return
+	}
+	_, err = io.Copy(part, content)
+	if err != nil {
+		glog.V(0).Infoln(err)
+		return
+	}
+
+	formData = buf
+	contentType = writer.FormDataContentType()
+
+	return
+}
+
+func checkContentMD5(w http.ResponseWriter, r *http.Request) (err error) {
+	if contentMD5 := r.Header.Get("Content-MD5"); contentMD5 != "" {
+		buf, _ := ioutil.ReadAll(r.Body)
+		//checkMD5
+		sum := md5.Sum(buf)
+		fileDataMD5 := base64.StdEncoding.EncodeToString(sum[0:len(sum)])
+		if strings.ToLower(fileDataMD5) != strings.ToLower(contentMD5) {
+			glog.V(0).Infof("fileDataMD5 [%s] is not equal to Content-MD5 [%s]", fileDataMD5, contentMD5)
+			err = fmt.Errorf("MD5 check failed")
+			writeJsonError(w, r, http.StatusNotAcceptable, err)
+			return
+		}
+		//reconstruct http request body for following new request to volume server
+		r.Body = ioutil.NopCloser(bytes.NewBuffer(buf))
+	}
+	return
+}
+
+func (fs *FilerServer) monolithicUploadAnalyzer(w http.ResponseWriter, r *http.Request, replication, collection string) (fileId, urlLocation string, err error) {
+	/*
+		Amazon S3 ref link:[http://docs.aws.amazon.com/AmazonS3/latest/API/Welcome.html]
+		There is a long way to provide a completely compatibility against all Amazon S3 API, I just made
+		a simple data stream adapter between S3 PUT API and seaweedfs's volume storage Write API
+		1. The request url format should be http://$host:$port/$bucketName/$objectName
+		2. bucketName will be mapped to seaweedfs's collection name
+		3. You could customize and make your enhancement.
+	*/
+	lastPos := strings.LastIndex(r.URL.Path, "/")
+	if lastPos == -1 || lastPos == 0 || lastPos == len(r.URL.Path)-1 {
+		glog.V(0).Infoln("URL Path [%s] is invalid, could not retrieve file name", r.URL.Path)
+		err = fmt.Errorf("URL Path is invalid")
+		writeJsonError(w, r, http.StatusInternalServerError, err)
+		return
+	}
+
+	if err = checkContentMD5(w, r); err != nil {
+		return
+	}
+
+	fileName := r.URL.Path[lastPos+1:]
+	if err = multipartHttpBodyBuilder(w, r, fileName); err != nil {
+		return
+	}
+
+	secondPos := strings.Index(r.URL.Path[1:], "/") + 1
+	collection = r.URL.Path[1:secondPos]
+	path := r.URL.Path
+
+	if fileId, urlLocation, err = fs.queryFileInfoByPath(w, r, path); err == nil && fileId == "" {
+		fileId, urlLocation, err = fs.assignNewFileInfo(w, r, replication, collection)
+	}
+	return
+}
+
+func multipartHttpBodyBuilder(w http.ResponseWriter, r *http.Request, fileName string) (err error) {
+	body, contentType, te := makeFormData(fileName, r.Header.Get("Content-Type"), r.Body)
+	if te != nil {
+		glog.V(0).Infoln("S3 protocol to raw seaweed protocol failed", te.Error())
+		writeJsonError(w, r, http.StatusInternalServerError, te)
+		err = te
+		return
+	}
+
+	if body != nil {
+		switch v := body.(type) {
+		case *bytes.Buffer:
+			r.ContentLength = int64(v.Len())
+		case *bytes.Reader:
+			r.ContentLength = int64(v.Len())
+		case *strings.Reader:
+			r.ContentLength = int64(v.Len())
+		}
+	}
+
+	r.Header.Set("Content-Type", contentType)
+	rc, ok := body.(io.ReadCloser)
+	if !ok && body != nil {
+		rc = ioutil.NopCloser(body)
+	}
+	r.Body = rc
+	return
+}

weed/server/filer_server_handlers_write_multipart.go

@@ -0,0 +1,39 @@
+package weed_server
+
+import (
+	"bytes"
+	"io/ioutil"
+	"net/http"
+	"strings"
+
+	"github.com/chrislusf/seaweedfs/weed/glog"
+	"github.com/chrislusf/seaweedfs/weed/storage"
+)
+
+func (fs *FilerServer) multipartUploadAnalyzer(w http.ResponseWriter, r *http.Request, replication, collection string) (fileId, urlLocation string, err error) {
+	//Default handle way for http multipart
+	if r.Method == "PUT" {
+		buf, _ := ioutil.ReadAll(r.Body)
+		r.Body = ioutil.NopCloser(bytes.NewBuffer(buf))
+		fileName, _, _, _, _, _, _, _, pe := storage.ParseUpload(r)
+		if pe != nil {
+			glog.V(0).Infoln("failing to parse post body", pe.Error())
+			writeJsonError(w, r, http.StatusInternalServerError, pe)
+			err = pe
+			return
+		}
+		//reconstruct http request body for following new request to volume server
+		r.Body = ioutil.NopCloser(bytes.NewBuffer(buf))
+
+		path := r.URL.Path
+		if strings.HasSuffix(path, "/") {
+			if fileName != "" {
+				path += fileName
+			}
+		}
+		fileId, urlLocation, err = fs.queryFileInfoByPath(w, r, path)
+	} else {
+		fileId, urlLocation, err = fs.assignNewFileInfo(w, r, replication, collection)
+	}
+	return
+}

weed/server/filer_ui/breadcrumb.go

@@ -0,0 +1,24 @@
+package master_ui
+
+import (
+	"path/filepath"
+	"strings"
+)
+
+type Breadcrumb struct {
+	Name string
+	Link string
+}
+
+func ToBreadcrumb(fullpath string) (crumbs []Breadcrumb) {
+	parts := strings.Split(fullpath, "/")
+
+	for i := 0; i < len(parts); i++ {
+		crumbs = append(crumbs, Breadcrumb{
+			Name: parts[i] + "/",
+			Link: "/" + filepath.Join(parts[0:i+1]...),
+		})
+	}
+
+	return
+}

weed/server/filer_ui/templates.go

@@ -20,29 +20,51 @@ var StatusTpl = template.Must(template.New("status").Parse(`<!DOCTYPE html>
 			</h1>
 		</div>
 		<div class="row">
-			{{.Path}}
+			{{ range $entry := .Breadcrumbs }}
+				<a href={{ $entry.Link }} >
+					{{ $entry.Name }}
+				</a>
+			{{ end }}
+
 		</div>
 
 		<div class="row">
-			<ul>
+			<table width="90%">
 				{{$path := .Path }}
-				{{ range $dirs_index, $dir := .Directories }}
-				<li>
+				{{ range $entry_index, $entry := .Entries }}
+				<tr>
+					<td>
+					{{if $entry.IsDirectory}}
 						<img src="https://www.w3.org/TR/WWWicn/folder.gif" width="20" height="23">
-					<a href={{ print $path  $dir  "/"}} >
-						{{ $dir }}
+						<a href={{ print $path  "/" $entry.Name  "/"}} >
+							{{ $entry.Name }}
 						</a>
-				</li>
-				{{ end }}
-
-				{{ range $file_index, $file := .Files }}
-				<li>
-					<a href={{ print $path  $file.Name}} >
-					{{ $file.Name }}
+					{{else}}
+						<a href={{ print $path  "/" $entry.Name }} >
+							{{ $entry.Name }}
 						</a>
-				</li>
+					{{end}}
+					</td>
+					<td align="right">
+					{{if $entry.IsDirectory}}
+					{{else}}
+						{{ $entry.Mime }}
+					{{end}}
+					</td>
+					<td align="right">
+					{{if $entry.IsDirectory}}
+					{{else}}
+						{{ $entry.Size }} bytes
+						&nbsp;&nbsp;&nbsp;
+					{{end}}
+					</td>
+					<td>
+						{{ $entry.Timestamp.Format "2006-01-02 15:04" }}
+					</td>
+				</tr>
 				{{ end }}
-			</ul>
+
+			</table>
 		</div>
 
 		{{if .ShouldDisplayLoadMore}}

weed/server/master_grpc_server.go

@@ -11,7 +11,7 @@ import (
 	"google.golang.org/grpc/peer"
 )
 
-func (ms MasterServer) SendHeartbeat(stream master_pb.Seaweed_SendHeartbeatServer) error {
+func (ms *MasterServer) SendHeartbeat(stream master_pb.Seaweed_SendHeartbeatServer) error {
 	var dn *topology.DataNode
 	t := ms.Topo
 	for {
@@ -77,3 +77,15 @@ func (ms MasterServer) SendHeartbeat(stream master_pb.Seaweed_SendHeartbeatServe
 		}
 	}
 }
+
+func (ms *MasterServer) KeepConnected(stream master_pb.Seaweed_KeepConnectedServer) error {
+	for {
+		_, err := stream.Recv()
+		if err != nil {
+			return err
+		}
+		if err := stream.Send(&master_pb.Empty{}); err != nil {
+			return err
+		}
+	}
+}

weed/server/raft_server.go

@@ -198,7 +198,7 @@ func postFollowingOneRedirect(target string, contentType string, b bytes.Buffer)
 	reply := string(data)
 
 	if strings.HasPrefix(reply, "\"http") {
-		urlStr := reply[1: len(reply)-1]
+		urlStr := reply[1 : len(reply)-1]
 
 		glog.V(0).Infoln("Post redirected to ", urlStr)
 		resp2, err2 := http.Post(urlStr, contentType, backupReader)

weed/server/volume_grpc_client.go

@@ -7,49 +7,51 @@ import (
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/pb/master_pb"
 	"github.com/chrislusf/seaweedfs/weed/security"
-	"github.com/chrislusf/seaweedfs/weed/storage"
 	"golang.org/x/net/context"
 	"google.golang.org/grpc"
 )
 
+func (vs *VolumeServer) GetMaster() string {
+	return vs.currentMaster
+}
 func (vs *VolumeServer) heartbeat() {
 
-	glog.V(0).Infof("Volume server bootstraps with master %s", vs.GetMasterNode())
-	vs.masterNodes = storage.NewMasterNodes(vs.masterNode)
+	glog.V(0).Infof("Volume server start with masters: %v", vs.MasterNodes)
 	vs.store.SetDataCenter(vs.dataCenter)
 	vs.store.SetRack(vs.rack)
 
+	var err error
+	var newLeader string
 	for {
-		err := vs.doHeartbeat(time.Duration(vs.pulseSeconds) * time.Second)
+		for _, master := range vs.MasterNodes {
+			if newLeader != "" {
+				master = newLeader
+			}
+			newLeader, err = vs.doHeartbeat(master, time.Duration(vs.pulseSeconds)*time.Second)
 			if err != nil {
 				glog.V(0).Infof("heartbeat error: %v", err)
 				time.Sleep(time.Duration(vs.pulseSeconds) * time.Second)
 			}
 		}
+	}
 }
 
-func (vs *VolumeServer) doHeartbeat(sleepInterval time.Duration) error {
-
-	vs.masterNodes.Reset()
-	masterNode, err := vs.masterNodes.FindMaster()
-	if err != nil {
-		return fmt.Errorf("No master found: %v", err)
-	}
+func (vs *VolumeServer) doHeartbeat(masterNode string, sleepInterval time.Duration) (newLeader string, err error) {
 
 	grpcConection, err := grpc.Dial(masterNode, grpc.WithInsecure())
 	if err != nil {
-		return fmt.Errorf("fail to dial: %v", err)
+		return "", fmt.Errorf("fail to dial: %v", err)
 	}
 	defer grpcConection.Close()
 
 	client := master_pb.NewSeaweedClient(grpcConection)
 	stream, err := client.SendHeartbeat(context.Background())
 	if err != nil {
-		glog.V(0).Infof("%v.SendHeartbeat(_) = _, %v", client, err)
-		return err
+		glog.V(0).Infof("SendHeartbeat to %s: %v", masterNode, err)
+		return "", err
 	}
-	vs.SetMasterNode(masterNode)
-	glog.V(0).Infof("Heartbeat to %s", masterNode)
+	glog.V(0).Infof("Heartbeat to: %v", masterNode)
+	vs.currentMaster = masterNode
 
 	vs.store.Client = stream
 	defer func() { vs.store.Client = nil }()
@@ -70,7 +72,8 @@ func (vs *VolumeServer) doHeartbeat(sleepInterval time.Duration) error {
 				vs.guard.SecretKey = security.Secret(in.GetSecretKey())
 			}
 			if in.GetLeader() != "" && masterNode != in.GetLeader() {
-				vs.masterNodes.SetPossibleLeader(in.GetLeader())
+				glog.V(0).Infof("Volume Server found a new master newLeader: %v instead of %v", in.GetLeader(), masterNode)
+				newLeader = in.GetLeader()
 				doneChan <- nil
 				return
 			}
@@ -79,7 +82,7 @@ func (vs *VolumeServer) doHeartbeat(sleepInterval time.Duration) error {
 
 	if err = stream.Send(vs.store.CollectHeartbeat()); err != nil {
 		glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", masterNode, err)
-		return err
+		return "", err
 	}
 
 	tickChan := time.Tick(sleepInterval)
@@ -89,11 +92,10 @@ func (vs *VolumeServer) doHeartbeat(sleepInterval time.Duration) error {
 		case <-tickChan:
 			if err = stream.Send(vs.store.CollectHeartbeat()); err != nil {
 				glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", masterNode, err)
-				return err
+				return "", err
 			}
-		case err := <-doneChan:
-			glog.V(0).Infof("Volume Server heart beat stops with %v", err)
-			return err
+		case <-doneChan:
+			return
 		}
 	}
 }

weed/server/volume_server.go

@@ -2,22 +2,19 @@ package weed_server
 
 import (
 	"net/http"
-	"sync"
-
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/security"
 	"github.com/chrislusf/seaweedfs/weed/storage"
 )
 
 type VolumeServer struct {
-	masterNode   string
-	mnLock       sync.RWMutex
+	MasterNodes   []string
+	currentMaster string
 	pulseSeconds  int
 	dataCenter    string
 	rack          string
 	store         *storage.Store
 	guard         *security.Guard
-	masterNodes  *storage.MasterNodes
 
 	needleMapKind     storage.NeedleMapType
 	FixJpgOrientation bool
@@ -28,7 +25,7 @@ func NewVolumeServer(adminMux, publicMux *http.ServeMux, ip string,
 	port int, publicUrl string,
 	folders []string, maxCounts []int,
 	needleMapKind storage.NeedleMapType,
-	masterNode string, pulseSeconds int,
+	masterNodes []string, pulseSeconds int,
 	dataCenter string, rack string,
 	whiteList []string,
 	fixJpgOrientation bool,
@@ -41,7 +38,7 @@ func NewVolumeServer(adminMux, publicMux *http.ServeMux, ip string,
 		FixJpgOrientation: fixJpgOrientation,
 		ReadRedirect:      readRedirect,
 	}
-	vs.SetMasterNode(masterNode)
+	vs.MasterNodes = masterNodes
 	vs.store = storage.NewStore(port, ip, publicUrl, folders, maxCounts, vs.needleMapKind)
 
 	vs.guard = security.NewGuard(whiteList, "")
@@ -67,7 +64,7 @@ func NewVolumeServer(adminMux, publicMux *http.ServeMux, ip string,
 	adminMux.HandleFunc("/", vs.privateStoreHandler)
 	if publicMux != adminMux {
 		// separated admin and public port
-		publicMux.HandleFunc("/favicon.ico", vs.faviconHandler)
+		publicMux.HandleFunc("/favicon.ico", faviconHandler)
 		publicMux.HandleFunc("/", vs.publicReadOnlyHandler)
 	}
 
@@ -76,18 +73,6 @@ func NewVolumeServer(adminMux, publicMux *http.ServeMux, ip string,
 	return vs
 }
 
-func (vs *VolumeServer) GetMasterNode() string {
-	vs.mnLock.RLock()
-	defer vs.mnLock.RUnlock()
-	return vs.masterNode
-}
-
-func (vs *VolumeServer) SetMasterNode(masterNode string) {
-	vs.mnLock.Lock()
-	defer vs.mnLock.Unlock()
-	vs.masterNode = masterNode
-}
-
 func (vs *VolumeServer) Shutdown() {
 	glog.V(0).Infoln("Shutting down volume server...")
 	vs.store.Close()

weed/server/volume_server_handlers.go

@@ -51,7 +51,3 @@ func (vs *VolumeServer) publicReadOnlyHandler(w http.ResponseWriter, r *http.Req
 		vs.GetOrHeadHandler(w, r)
 	}
 }
-
-func (vs *VolumeServer) faviconHandler(w http.ResponseWriter, r *http.Request) {
-	vs.FaviconHandler(w, r)
-}

weed/server/volume_server_handlers_admin.go

@@ -8,8 +8,8 @@ import (
 
 	"github.com/chrislusf/seaweedfs/weed/glog"
 	"github.com/chrislusf/seaweedfs/weed/stats"
-	"github.com/chrislusf/seaweedfs/weed/util"
 	"github.com/chrislusf/seaweedfs/weed/storage"
+	"github.com/chrislusf/seaweedfs/weed/util"
 )
 
 func (vs *VolumeServer) statusHandler(w http.ResponseWriter, r *http.Request) {
@@ -25,7 +25,7 @@ func (vs *VolumeServer) assignVolumeHandler(w http.ResponseWriter, r *http.Reque
 	if r.FormValue("preallocate") != "" {
 		preallocate, err = strconv.ParseInt(r.FormValue("preallocate"), 10, 64)
 		if err != nil {
-			glog.V(0).Infoln("ignoring invalid int64 value for preallocate = %v", r.FormValue("preallocate"))
+			glog.V(0).Infof("ignoring invalid int64 value for preallocate = %v", r.FormValue("preallocate"))
 		}
 	}
 	err = vs.store.AddVolume(
@@ -41,7 +41,7 @@ func (vs *VolumeServer) assignVolumeHandler(w http.ResponseWriter, r *http.Reque
 	} else {
 		writeJsonError(w, r, http.StatusNotAcceptable, err)
 	}
-	glog.V(2).Infoln("assign volume = %s, collection = %s , replication = %s, error = %v",
+	glog.V(2).Infof("assign volume = %s, collection = %s , replication = %s, error = %v",
 		r.FormValue("volume"), r.FormValue("collection"), r.FormValue("replication"), err)
 }
 

weed/server/volume_server_handlers_read.go

@@ -46,7 +46,7 @@ func (vs *VolumeServer) GetOrHeadHandler(w http.ResponseWriter, r *http.Request)
 			w.WriteHeader(http.StatusNotFound)
 			return
 		}
-		lookupResult, err := operation.Lookup(vs.GetMasterNode(), volumeId.String())
+		lookupResult, err := operation.Lookup(vs.GetMaster(), volumeId.String())
 		glog.V(2).Infoln("volume", volumeId, "found on", lookupResult, "error", err)
 		if err == nil && len(lookupResult.Locations) > 0 {
 			u, _ := url.Parse(util.NormalizeUrl(lookupResult.Locations[0].PublicUrl))
@@ -151,18 +151,6 @@ func (vs *VolumeServer) GetOrHeadHandler(w http.ResponseWriter, r *http.Request)
 	}
 }
 
-func (vs *VolumeServer) FaviconHandler(w http.ResponseWriter, r *http.Request) {
-	data, err := images.Asset("favicon/favicon.ico")
-	if err != nil {
-		glog.V(2).Infoln("favicon read error:", err)
-		return
-	}
-
-	if e := writeResponseContent("favicon.ico", "image/x-icon", bytes.NewReader(data), w, r); e != nil {
-		glog.V(2).Infoln("response write error:", e)
-	}
-}
-
 func (vs *VolumeServer) tryHandleChunkedFile(n *storage.Needle, fileName string, w http.ResponseWriter, r *http.Request) (processed bool) {
 	if !n.IsChunkedManifest() || r.URL.Query().Get("cm") == "false" {
 		return false
@@ -188,7 +176,7 @@ func (vs *VolumeServer) tryHandleChunkedFile(n *storage.Needle, fileName string,
 
 	chunkedFileReader := &operation.ChunkedFileReader{
 		Manifest: chunkManifest,
-		Master:   vs.GetMasterNode(),
+		Master:   vs.GetMaster(),
 	}
 	defer chunkedFileReader.Close()
 	if e := writeResponseContent(fileName, mType, chunkedFileReader, w, r); e != nil {

weed/server/volume_server_handlers_ui.go

@@ -5,9 +5,9 @@ import (
 	"path/filepath"
 	"time"
 
+	ui "github.com/chrislusf/seaweedfs/weed/server/volume_server_ui"
 	"github.com/chrislusf/seaweedfs/weed/stats"
 	"github.com/chrislusf/seaweedfs/weed/util"
-	ui "github.com/chrislusf/seaweedfs/weed/server/volume_server_ui"
 )
 
 func (vs *VolumeServer) uiStatusHandler(w http.ResponseWriter, r *http.Request) {
@@ -21,14 +21,14 @@ func (vs *VolumeServer) uiStatusHandler(w http.ResponseWriter, r *http.Request)
 	}
 	args := struct {
 		Version      string
-		Master       string
+		Masters      []string
 		Volumes      interface{}
 		DiskStatuses interface{}
 		Stats        interface{}
 		Counters     *stats.ServerStats
 	}{
 		util.VERSION,
-		vs.masterNode,
+		vs.MasterNodes,
 		vs.store.Status(),
 		ds,
 		infos,

weed/server/volume_server_handlers_vacuum.go

@@ -22,7 +22,7 @@ func (vs *VolumeServer) vacuumVolumeCompactHandler(w http.ResponseWriter, r *htt
 	if r.FormValue("preallocate") != "" {
 		preallocate, err = strconv.ParseInt(r.FormValue("preallocate"), 10, 64)
 		if err != nil {
-			glog.V(0).Infoln("Failed to parse int64 preallocate = %s: %v", r.FormValue("preallocate"), err)
+			glog.V(0).Infof("Failed to parse int64 preallocate = %s: %v", r.FormValue("preallocate"), err)
 		}
 	}
 	err = vs.store.CompactVolume(r.FormValue("volume"), preallocate)

weed/server/volume_server_handlers_write.go

@@ -31,7 +31,7 @@ func (vs *VolumeServer) PostHandler(w http.ResponseWriter, r *http.Request) {
 	}
 
 	ret := operation.UploadResult{}
-	size, errorStatus := topology.ReplicatedWrite(vs.GetMasterNode(),
+	size, errorStatus := topology.ReplicatedWrite(vs.GetMaster(),
 		vs.store, volumeId, needle, r)
 	httpStatus := http.StatusCreated
 	if errorStatus != "" {
@@ -80,14 +80,14 @@ func (vs *VolumeServer) DeleteHandler(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 		// make sure all chunks had deleted before delete manifest
-		if e := chunkManifest.DeleteChunks(vs.GetMasterNode()); e != nil {
+		if e := chunkManifest.DeleteChunks(vs.GetMaster()); e != nil {
 			writeJsonError(w, r, http.StatusInternalServerError, fmt.Errorf("Delete chunks error: %v", e))
 			return
 		}
 		count = chunkManifest.Size
 	}
 
-	_, err := topology.ReplicatedDelete(vs.GetMasterNode(), vs.store, volumeId, n, r)
+	_, err := topology.ReplicatedDelete(vs.GetMaster(), vs.store, volumeId, n, r)
 
 	if err == nil {
 		m := make(map[string]int64)

weed/server/volume_server_ui/templates.go

@@ -72,8 +72,8 @@ var StatusTpl = template.Must(template.New("status").Funcs(funcMap).Parse(`<!DOC
           <h2>System Stats</h2>
           <table class="table table-condensed table-striped">
             <tr>
-              <th>Master</th>
-              <td><a href="http://{{.Master}}/ui/index.html">{{.Master}}</a></td>
+              <th>Masters</th>
+              <td>{{.Masters}}</td>
             </tr>
             <tr>
               <th>Weekly # ReadRequests</th>

weed/storage/needle_map_memory.go

@@ -53,14 +53,14 @@ func doLoading(file *os.File, nm *NeedleMap) (*NeedleMap, error) {
 			nm.FileCounter++
 			nm.FileByteCounter = nm.FileByteCounter + uint64(size)
 			oldOffset, oldSize := nm.m.Set(needle.Key(key), offset, size)
-			glog.V(3).Infoln("reading key", key, "offset", offset*NeedlePaddingSize, "size", size, "oldSize", oldSize)
+			// glog.V(3).Infoln("reading key", key, "offset", offset*NeedlePaddingSize, "size", size, "oldSize", oldSize)
 			if oldOffset > 0 && oldSize != TombstoneFileSize {
 				nm.DeletionCounter++
 				nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
 			}
 		} else {
 			oldSize := nm.m.Delete(needle.Key(key))
-			glog.V(3).Infoln("removing key", key, "offset", offset*NeedlePaddingSize, "size", size, "oldSize", oldSize)
+			// glog.V(3).Infoln("removing key", key, "offset", offset*NeedlePaddingSize, "size", size, "oldSize", oldSize)
 			nm.DeletionCounter++
 			nm.DeletionByteCounter = nm.DeletionByteCounter + uint64(oldSize)
 		}

weed/storage/store.go

@@ -1,13 +1,11 @@
 package storage
 
 import (
-	"errors"
 	"fmt"
 	"strconv"
 	"strings"
 
 	"github.com/chrislusf/seaweedfs/weed/glog"
-	"github.com/chrislusf/seaweedfs/weed/operation"
 	"github.com/chrislusf/seaweedfs/weed/pb/master_pb"
 )
 
@@ -15,55 +13,6 @@ const (
 	MAX_TTL_VOLUME_REMOVAL_DELAY = 10 // 10 minutes
 )
 
-type MasterNodes struct {
-	nodes          []string
-	leader         string
-	possibleLeader string
-}
-
-func (mn *MasterNodes) String() string {
-	return fmt.Sprintf("nodes:%v, leader:%s", mn.nodes, mn.leader)
-}
-
-func NewMasterNodes(bootstrapNode string) (mn *MasterNodes) {
-	mn = &MasterNodes{nodes: []string{bootstrapNode}, leader: ""}
-	return
-}
-func (mn *MasterNodes) Reset() {
-	if mn.leader != "" {
-		mn.leader = ""
-		glog.V(0).Infof("Resetting master nodes: %v", mn)
-	}
-}
-func (mn *MasterNodes) SetPossibleLeader(possibleLeader string) {
-	// TODO try to check this leader first
-	mn.possibleLeader = possibleLeader
-}
-func (mn *MasterNodes) FindMaster() (leader string, err error) {
-	if len(mn.nodes) == 0 {
-		return "", errors.New("No master node found!")
-	}
-	if mn.leader == "" {
-		for _, m := range mn.nodes {
-			glog.V(4).Infof("Listing masters on %s", m)
-			if leader, masters, e := operation.ListMasters(m); e == nil {
-				if leader != "" {
-					mn.nodes = append(masters, m)
-					mn.leader = leader
-					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.leader == "" {
-		return "", errors.New("No master node available!")
-	}
-	return mn.leader, nil
-}
-
 /*
  * A VolumeServer contains one Store
  */

weed/storage/volume_checking.go

@@ -12,7 +12,7 @@ func getActualSize(size uint32) int64 {
 	return NeedleHeaderSize + int64(size) + NeedleChecksumSize + int64(padding)
 }
 
-func CheckVolumeDataIntegrity(v *Volume, indexFile *os.File) (error) {
+func CheckVolumeDataIntegrity(v *Volume, indexFile *os.File) error {
 	var indexSize int64
 	var e error
 	if indexSize, e = verifyIndexFileIntegrity(indexFile); e != nil {

weed/topology/store_replicate.go

@@ -159,5 +159,4 @@ func distributedOperation(masterNode string, store *storage.Store, volumeId stor
 		glog.V(0).Infoln()
 		return fmt.Errorf("Failed to lookup for %d: %v", volumeId, lookupErr)
 	}
-	return nil
 }

weed/util/constants.go

@@ -1,5 +1,5 @@
 package util
 
 const (
-	VERSION = "0.77"
+	VERSION = "0.90 beta"
 )

weed/util/http_util.go

@@ -141,7 +141,6 @@ func GetBufferStream(url string, values url.Values, allocatedBytes []byte, eachB
 			return err
 		}
 	}
-	return nil
 }
 
 func GetUrlStream(url string, values url.Values, readFn func(io.Reader) error) error {
@@ -183,3 +182,70 @@ func NormalizeUrl(url string) string {
 	}
 	return "http://" + url
 }
+
+func ReadUrl(fileUrl string, offset int64, size int, buf []byte) (n int64, e error) {
+
+	req, _ := http.NewRequest("GET", fileUrl, nil)
+	req.Header.Add("Range", fmt.Sprintf("bytes=%d-%d", offset, offset+int64(size)))
+
+	r, err := client.Do(req)
+	if err != nil {
+		return 0, err
+	}
+	defer r.Body.Close()
+	if r.StatusCode >= 400 {
+		return 0, fmt.Errorf("%s: %s", fileUrl, r.Status)
+	}
+
+	var i, m int
+
+	for {
+		m, err = r.Body.Read(buf[i:])
+		if m == 0 {
+			return
+		}
+		i += m
+		n += int64(m)
+		if err == io.EOF {
+			return n, nil
+		}
+		if e != nil {
+			return n, e
+		}
+	}
+
+}
+
+func ReadUrlAsStream(fileUrl string, offset int64, size int, fn func(data []byte)) (n int64, e error) {
+
+	req, _ := http.NewRequest("GET", fileUrl, nil)
+	req.Header.Add("Range", fmt.Sprintf("bytes=%d-%d", offset, offset+int64(size)))
+
+	r, err := client.Do(req)
+	if err != nil {
+		return 0, err
+	}
+	defer r.Body.Close()
+	if r.StatusCode >= 400 {
+		return 0, fmt.Errorf("%s: %s", fileUrl, r.Status)
+	}
+
+	var m int
+	buf := make([]byte, 64*1024)
+
+	for {
+		m, err = r.Body.Read(buf)
+		if m == 0 {
+			return
+		}
+		fn(buf[:m])
+		n += int64(m)
+		if err == io.EOF {
+			return n, nil
+		}
+		if e != nil {
+			return n, e
+		}
+	}
+
+}