mirror of
https://github.com/seaweedfs/seaweedfs.git
synced 2024-01-19 02:48:24 +00:00
simplifying volume id mechanism, removing automatic volume id generation and discovering
periodically report machine status git-svn-id: https://weed-fs.googlecode.com/svn/trunk@25 282b0af5-e82d-9cf1-ede4-77906d7719d0
This commit is contained in:
parent
ae3a53388f
commit
6913fabc64
|
@ -10,15 +10,17 @@ import (
|
|||
"mime"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
var (
|
||||
port = flag.Int("port", 8080, "http listen port")
|
||||
chunkFolder = flag.String("dir", "/tmp", "data directory to store files")
|
||||
chunkCount = flag.Int("chunks", 5, "data chunks to store files")
|
||||
volumes = flag.String("volumes", "0,1-3,4", "comma-separated list of volume ids or range of ids")
|
||||
publicServer = flag.String("pserver", "localhost:8080", "public server to serve data read")
|
||||
metaServer = flag.String("mserver", "localhost:9333", "metadata server to store mappings")
|
||||
IsDebug = flag.Bool("debug", false, "enable debug mode")
|
||||
pulse = flag.Int("pulseSeconds", 5, "number of seconds between heartbeats")
|
||||
|
||||
store *storage.Store
|
||||
)
|
||||
|
@ -38,8 +40,8 @@ func storeHandler(w http.ResponseWriter, r *http.Request) {
|
|||
}
|
||||
func GetHandler(w http.ResponseWriter, r *http.Request) {
|
||||
n := new(storage.Needle)
|
||||
vid, fid, ext := parseURLPath(r.URL.Path)
|
||||
volumeId, _ := strconv.Atoui64(vid)
|
||||
vid, fid, ext := parseURLPath(r.URL.Path)
|
||||
volumeId, _ := strconv.Atoui64(vid)
|
||||
n.ParsePath(fid)
|
||||
|
||||
if *IsDebug {
|
||||
|
@ -54,14 +56,14 @@ func GetHandler(w http.ResponseWriter, r *http.Request) {
|
|||
log.Println("request with unmaching cookie from ", r.RemoteAddr, "agent", r.UserAgent())
|
||||
return
|
||||
}
|
||||
if ext!="" {
|
||||
if ext != "" {
|
||||
w.Header().Set("Content-Type", mime.TypeByExtension(ext))
|
||||
}
|
||||
w.Write(n.Data)
|
||||
}
|
||||
func PostHandler(w http.ResponseWriter, r *http.Request) {
|
||||
vid, _, _ := parseURLPath(r.URL.Path)
|
||||
volumeId, e := strconv.Atoui64(vid)
|
||||
vid, _, _ := parseURLPath(r.URL.Path)
|
||||
volumeId, e := strconv.Atoui64(vid)
|
||||
if e != nil {
|
||||
writeJson(w, r, e)
|
||||
} else {
|
||||
|
@ -73,19 +75,19 @@ func PostHandler(w http.ResponseWriter, r *http.Request) {
|
|||
}
|
||||
func DeleteHandler(w http.ResponseWriter, r *http.Request) {
|
||||
n := new(storage.Needle)
|
||||
vid, fid, _ := parseURLPath(r.URL.Path)
|
||||
volumeId, _ := strconv.Atoui64(vid)
|
||||
n.ParsePath(fid)
|
||||
vid, fid, _ := parseURLPath(r.URL.Path)
|
||||
volumeId, _ := strconv.Atoui64(vid)
|
||||
n.ParsePath(fid)
|
||||
|
||||
cookie := n.Cookie
|
||||
count, _ := store.Read(volumeId, n)
|
||||
cookie := n.Cookie
|
||||
count, _ := store.Read(volumeId, n)
|
||||
|
||||
if n.Cookie != cookie {
|
||||
log.Println("delete with unmaching cookie from ", r.RemoteAddr, "agent", r.UserAgent())
|
||||
return
|
||||
}
|
||||
|
||||
n.Size = 0
|
||||
if n.Cookie != cookie {
|
||||
log.Println("delete with unmaching cookie from ", r.RemoteAddr, "agent", r.UserAgent())
|
||||
return
|
||||
}
|
||||
|
||||
n.Size = 0
|
||||
store.Write(volumeId, n)
|
||||
m := make(map[string]uint32)
|
||||
m["size"] = uint32(count)
|
||||
|
@ -105,33 +107,38 @@ func writeJson(w http.ResponseWriter, r *http.Request, obj interface{}) {
|
|||
}
|
||||
//log.Println("JSON Response", string(bytes))
|
||||
}
|
||||
func parseURLPath(path string)(vid, fid, ext string){
|
||||
sepIndex := strings.LastIndex(path, "/")
|
||||
commaIndex := strings.LastIndex(path[sepIndex:], ",")
|
||||
dotIndex := strings.LastIndex(path[sepIndex:], ".")
|
||||
vid = path[sepIndex+1 : commaIndex]
|
||||
fid = path[commaIndex+1:]
|
||||
ext = ""
|
||||
if dotIndex > 0 {
|
||||
fid = path[commaIndex+1 : dotIndex]
|
||||
ext = path[dotIndex+1:]
|
||||
}
|
||||
if commaIndex <= 0 {
|
||||
log.Println("unknown file id", path[sepIndex+1:commaIndex])
|
||||
return
|
||||
}
|
||||
return
|
||||
func parseURLPath(path string) (vid, fid, ext string) {
|
||||
sepIndex := strings.LastIndex(path, "/")
|
||||
commaIndex := strings.LastIndex(path[sepIndex:], ",")
|
||||
dotIndex := strings.LastIndex(path[sepIndex:], ".")
|
||||
vid = path[sepIndex+1 : commaIndex]
|
||||
fid = path[commaIndex+1:]
|
||||
ext = ""
|
||||
if dotIndex > 0 {
|
||||
fid = path[commaIndex+1 : dotIndex]
|
||||
ext = path[dotIndex+1:]
|
||||
}
|
||||
if commaIndex <= 0 {
|
||||
log.Println("unknown file id", path[sepIndex+1:commaIndex])
|
||||
return
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func main() {
|
||||
flag.Parse()
|
||||
//TODO: now default to 1G, this value should come from server?
|
||||
store = storage.NewStore(*port, *publicServer, *chunkFolder, 1024*1024*1024, *chunkCount)
|
||||
store = storage.NewStore(*port, *publicServer, *chunkFolder, *volumes)
|
||||
defer store.Close()
|
||||
http.HandleFunc("/", storeHandler)
|
||||
http.HandleFunc("/status", statusHandler)
|
||||
|
||||
store.Join(*metaServer)
|
||||
go func() {
|
||||
for {
|
||||
store.Join(*metaServer)
|
||||
time.Sleep(int64(*pulse) * 1e9)
|
||||
}
|
||||
}()
|
||||
log.Println("store joined at", *metaServer)
|
||||
|
||||
log.Println("Start storage service at http://127.0.0.1:" + strconv.Itoa(*port))
|
||||
|
|
|
@ -17,6 +17,7 @@ var (
|
|||
metaFolder = flag.String("mdir", "/tmp", "data directory to store mappings")
|
||||
capacity = flag.Int("capacity", 100, "maximum number of volumes to hold")
|
||||
mapper *directory.Mapper
|
||||
IsDebug = flag.Bool("debug", false, "verbose debug information")
|
||||
)
|
||||
|
||||
func dirLookupHandler(w http.ResponseWriter, r *http.Request) {
|
||||
|
@ -42,10 +43,8 @@ func dirJoinHandler(w http.ResponseWriter, r *http.Request) {
|
|||
publicServer := r.FormValue("publicServer")
|
||||
volumes := new([]storage.VolumeInfo)
|
||||
json.Unmarshal([]byte(r.FormValue("volumes")), volumes)
|
||||
capacity, _ := strconv.Atoi(r.FormValue("capacity"))
|
||||
log.Println("Recieved joining request from remote address", s, "capacity=", capacity, "volumes", r.FormValue("volumes"))
|
||||
vids := mapper.Add(*directory.NewMachine(s, publicServer, *volumes, capacity))
|
||||
writeJson(w, r, vids)
|
||||
log.Println("Recieved updates from", s, "volumes", r.FormValue("volumes"))
|
||||
mapper.Add(*directory.NewMachine(s, publicServer, *volumes))
|
||||
}
|
||||
func dirStatusHandler(w http.ResponseWriter, r *http.Request) {
|
||||
writeJson(w, r, mapper)
|
||||
|
@ -67,8 +66,7 @@ func writeJson(w http.ResponseWriter, r *http.Request, obj interface{}) {
|
|||
|
||||
func main() {
|
||||
flag.Parse()
|
||||
mapper = directory.NewMapper(*metaFolder, "directory", *capacity)
|
||||
defer mapper.Save()
|
||||
mapper = directory.NewMapper(*metaFolder, "directory")
|
||||
http.HandleFunc("/dir/assign", dirAssignHandler)
|
||||
http.HandleFunc("/dir/lookup", dirLookupHandler)
|
||||
http.HandleFunc("/dir/write", dirWriteHandler)
|
||||
|
|
|
@ -22,65 +22,40 @@ type MachineInfo struct {
|
|||
type Machine struct {
|
||||
Server MachineInfo
|
||||
Volumes []storage.VolumeInfo
|
||||
Capacity int
|
||||
}
|
||||
|
||||
type Mapper struct {
|
||||
dir string
|
||||
fileName string
|
||||
capacity int
|
||||
|
||||
lock sync.Mutex
|
||||
Machines []*Machine
|
||||
vid2machineId map[uint32]int
|
||||
Writers []int // transient array of Writers volume id
|
||||
|
||||
GlobalVolumeSequence uint32
|
||||
|
||||
FileIdSequence uint64
|
||||
fileIdCounter uint64
|
||||
}
|
||||
|
||||
func NewMachine(server, publicServer string, volumes []storage.VolumeInfo, capacity int) *Machine {
|
||||
return &Machine{Server: MachineInfo{Url: server, PublicUrl: publicServer}, Volumes: volumes, Capacity: capacity}
|
||||
func NewMachine(server, publicServer string, volumes []storage.VolumeInfo) *Machine {
|
||||
return &Machine{Server: MachineInfo{Url: server, PublicUrl: publicServer}, Volumes: volumes}
|
||||
}
|
||||
|
||||
func NewMapper(dirname string, filename string, capacity int) (m *Mapper) {
|
||||
m = &Mapper{dir: dirname, fileName: filename, capacity: capacity}
|
||||
log.Println("Loading volume id to maching mapping:", path.Join(m.dir, m.fileName+".map"))
|
||||
dataFile, e := os.OpenFile(path.Join(m.dir, m.fileName+".map"), os.O_RDONLY, 0644)
|
||||
func NewMapper(dirname string, filename string) (m *Mapper) {
|
||||
m = &Mapper{dir: dirname, fileName: filename}
|
||||
m.vid2machineId = make(map[uint32]int)
|
||||
m.Writers = *new([]int)
|
||||
if e != nil {
|
||||
log.Println("Mapping File Read", e)
|
||||
m.Machines = *new([]*Machine)
|
||||
} else {
|
||||
decoder := gob.NewDecoder(dataFile)
|
||||
defer dataFile.Close()
|
||||
decoder.Decode(&m.Machines)
|
||||
decoder.Decode(&m.GlobalVolumeSequence)
|
||||
|
||||
//add to vid2machineId map, and Writers array
|
||||
for machine_index, machine := range m.Machines {
|
||||
for _, v := range machine.Volumes {
|
||||
m.vid2machineId[v.Id] = machine_index
|
||||
if v.Size < ChunkSizeLimit {
|
||||
m.Writers = append(m.Writers, machine_index)
|
||||
}
|
||||
}
|
||||
}
|
||||
log.Println("Loaded mapping size", len(m.Machines))
|
||||
}
|
||||
m.Machines = *new([]*Machine)
|
||||
|
||||
seqFile, se := os.OpenFile(path.Join(m.dir, m.fileName+".seq"), os.O_RDONLY, 0644)
|
||||
if se != nil {
|
||||
m.FileIdSequence = FileIdSaveInterval
|
||||
log.Println("Setting file id sequence", m.FileIdSequence)
|
||||
log.Println("Setting file id sequence", m.FileIdSequence)
|
||||
} else {
|
||||
decoder := gob.NewDecoder(seqFile)
|
||||
defer seqFile.Close()
|
||||
decoder.Decode(&m.FileIdSequence)
|
||||
log.Println("Loading file id sequence", m.FileIdSequence, "=>", m.FileIdSequence + FileIdSaveInterval)
|
||||
log.Println("Loading file id sequence", m.FileIdSequence, "=>", m.FileIdSequence+FileIdSaveInterval)
|
||||
//in case the server stops between intervals
|
||||
m.FileIdSequence += FileIdSaveInterval
|
||||
}
|
||||
|
@ -89,22 +64,20 @@ func NewMapper(dirname string, filename string, capacity int) (m *Mapper) {
|
|||
func (m *Mapper) PickForWrite() (string, MachineInfo) {
|
||||
machine := m.Machines[m.Writers[rand.Intn(len(m.Writers))]]
|
||||
vid := machine.Volumes[rand.Intn(len(machine.Volumes))].Id
|
||||
return NewFileId(vid,m.NextFileId(),rand.Uint32()).String(), machine.Server
|
||||
return NewFileId(vid, m.NextFileId(), rand.Uint32()).String(), machine.Server
|
||||
}
|
||||
func (m *Mapper) NextFileId() uint64 {
|
||||
if m.fileIdCounter <= 0 {
|
||||
m.fileIdCounter = FileIdSaveInterval
|
||||
m.saveSequence()
|
||||
}
|
||||
if m.fileIdCounter <= 0 {
|
||||
m.fileIdCounter = FileIdSaveInterval
|
||||
m.saveSequence()
|
||||
}
|
||||
m.fileIdCounter--
|
||||
return m.FileIdSequence - m.fileIdCounter
|
||||
}
|
||||
func (m *Mapper) Get(vid uint32) *Machine {
|
||||
return m.Machines[m.vid2machineId[vid]]
|
||||
}
|
||||
func (m *Mapper) Add(machine Machine) []uint32 {
|
||||
log.Println("Adding existing", machine.Server.Url, len(machine.Volumes), "volumes to dir", len(m.Machines))
|
||||
log.Println("Adding new ", machine.Server.Url, machine.Capacity-len(machine.Volumes), "volumes to dir", len(m.Machines))
|
||||
func (m *Mapper) Add(machine Machine){
|
||||
//check existing machine, linearly
|
||||
m.lock.Lock()
|
||||
foundExistingMachineId := -1
|
||||
|
@ -119,24 +92,11 @@ func (m *Mapper) Add(machine Machine) []uint32 {
|
|||
machineId = len(m.Machines)
|
||||
m.Machines = append(m.Machines, &machine)
|
||||
}
|
||||
|
||||
//generate new volumes
|
||||
vids := new([]uint32)
|
||||
for vid, i := m.GlobalVolumeSequence, len(machine.Volumes); i < machine.Capacity; i, vid = i+1, vid+1 {
|
||||
newVolume := storage.VolumeInfo{Id: vid, Size: 0}
|
||||
machine.Volumes = append(machine.Volumes, newVolume)
|
||||
m.vid2machineId[vid] = machineId
|
||||
log.Println("Adding volume", vid, "from", machine.Server.Url)
|
||||
*vids = append(*vids, vid)
|
||||
m.GlobalVolumeSequence = vid + 1
|
||||
}
|
||||
|
||||
m.Save()
|
||||
m.lock.Unlock()
|
||||
|
||||
//add to vid2machineId map, and Writers array
|
||||
for _, v := range machine.Volumes {
|
||||
log.Println("Setting volume", v.Id, "to", machine.Server.Url)
|
||||
//log.Println("Setting volume", v.Id, "to", machine.Server.Url)
|
||||
m.vid2machineId[v.Id] = machineId
|
||||
if v.Size < ChunkSizeLimit {
|
||||
m.Writers = append(m.Writers, machineId)
|
||||
|
@ -152,20 +112,6 @@ func (m *Mapper) Add(machine Machine) []uint32 {
|
|||
}
|
||||
}
|
||||
m.Writers = Writers
|
||||
|
||||
log.Println("Machines:", len(m.Machines), "Volumes:", len(m.vid2machineId), "Writable:", len(m.Writers))
|
||||
return *vids
|
||||
}
|
||||
func (m *Mapper) Save() {
|
||||
log.Println("Saving virtual to physical:", path.Join(m.dir, m.fileName+".map"))
|
||||
dataFile, e := os.OpenFile(path.Join(m.dir, m.fileName+".map"), os.O_CREATE|os.O_WRONLY, 0644)
|
||||
if e != nil {
|
||||
log.Fatalf("Mapping File Save [ERROR] %s\n", e)
|
||||
}
|
||||
defer dataFile.Close()
|
||||
encoder := gob.NewEncoder(dataFile)
|
||||
encoder.Encode(m.Machines)
|
||||
encoder.Encode(m.GlobalVolumeSequence)
|
||||
}
|
||||
func (m *Mapper) saveSequence() {
|
||||
log.Println("Saving file id sequence", m.FileIdSequence, "to", path.Join(m.dir, m.fileName+".seq"))
|
||||
|
|
|
@ -2,7 +2,6 @@ package storage
|
|||
|
||||
import (
|
||||
"log"
|
||||
"io/ioutil"
|
||||
"json"
|
||||
"os"
|
||||
"strings"
|
||||
|
@ -13,7 +12,6 @@ import (
|
|||
|
||||
type Store struct {
|
||||
volumes map[uint64]*Volume
|
||||
capacity int
|
||||
dir string
|
||||
Port int
|
||||
PublicServer string
|
||||
|
@ -23,35 +21,48 @@ type VolumeInfo struct {
|
|||
Size int64
|
||||
}
|
||||
|
||||
func NewStore(port int, publicServer, dirname string, chunkSize, capacity int) (s *Store) {
|
||||
s = &Store{Port: port, PublicServer: publicServer, dir: dirname, capacity: capacity}
|
||||
func NewStore(port int, publicServer, dirname string, volumeListString string) (s *Store) {
|
||||
s = &Store{Port: port, PublicServer: publicServer, dir: dirname}
|
||||
s.volumes = make(map[uint64]*Volume)
|
||||
|
||||
files, _ := ioutil.ReadDir(dirname)
|
||||
for _, f := range files {
|
||||
if f.IsDirectory() || !strings.HasSuffix(f.Name, ".dat") {
|
||||
continue
|
||||
|
||||
for _, range_string := range strings.Split(volumeListString, ",") {
|
||||
if strings.Index(range_string, "-") < 0 {
|
||||
id_string := range_string
|
||||
id, err := strconv.Atoui64(id_string)
|
||||
if err != nil {
|
||||
log.Println("Volume Id", id_string, "is not a valid unsigned integer! Skipping it...")
|
||||
continue
|
||||
}
|
||||
s.volumes[id] = NewVolume(s.dir, uint32(id))
|
||||
} else {
|
||||
pair := strings.Split(range_string, "-")
|
||||
start, start_err := strconv.Atoui64(pair[0])
|
||||
if start_err != nil {
|
||||
log.Println("Volume Id", pair[0], "is not a valid unsigned integer! Skipping it...")
|
||||
continue
|
||||
}
|
||||
end, end_err := strconv.Atoui64(pair[1])
|
||||
if end_err != nil {
|
||||
log.Println("Volume Id", pair[1], "is not a valid unsigned integer! Skipping it...")
|
||||
continue
|
||||
}
|
||||
for id := start; id<=end; id++ {
|
||||
s.volumes[id] = NewVolume(s.dir, uint32(id))
|
||||
}
|
||||
}
|
||||
id, err := strconv.Atoui64(f.Name[0:(strings.LastIndex(f.Name, ".dat"))])
|
||||
log.Println("Loading data file name:", f.Name)
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
s.volumes[id] = NewVolume(s.dir, uint32(id))
|
||||
}
|
||||
log.Println("Store started on dir:", dirname, "with", len(s.volumes), "existing volumes")
|
||||
log.Println("Expected capacity=", s.capacity, "volumes")
|
||||
return
|
||||
}
|
||||
|
||||
func (s *Store) Status()(*[]*VolumeInfo){
|
||||
stats := new([]*VolumeInfo)
|
||||
for k, v := range s.volumes {
|
||||
s := new(VolumeInfo)
|
||||
s.Id, s.Size = uint32(k), v.Size()
|
||||
*stats = append(*stats, s)
|
||||
}
|
||||
return stats
|
||||
func (s *Store) Status() *[]*VolumeInfo {
|
||||
stats := new([]*VolumeInfo)
|
||||
for k, v := range s.volumes {
|
||||
s := new(VolumeInfo)
|
||||
s.Id, s.Size = uint32(k), v.Size()
|
||||
*stats = append(*stats, s)
|
||||
}
|
||||
return stats
|
||||
}
|
||||
func (s *Store) Join(mserver string) {
|
||||
stats := new([]*VolumeInfo)
|
||||
|
@ -65,29 +76,17 @@ func (s *Store) Join(mserver string) {
|
|||
values.Add("port", strconv.Itoa(s.Port))
|
||||
values.Add("publicServer", s.PublicServer)
|
||||
values.Add("volumes", string(bytes))
|
||||
log.Println("Registering exiting volumes", string(bytes))
|
||||
values.Add("capacity", strconv.Itoa(s.capacity))
|
||||
retString := util.Post("http://"+mserver+"/dir/join", values)
|
||||
if retString != nil {
|
||||
newVids := new([]int)
|
||||
log.Println("Instructed to create volume", string(retString))
|
||||
e := json.Unmarshal(retString, newVids)
|
||||
if e == nil {
|
||||
for _, vid := range *newVids {
|
||||
s.volumes[uint64(vid)] = NewVolume(s.dir, uint32(vid))
|
||||
log.Println("Adding volume", vid)
|
||||
}
|
||||
}
|
||||
}
|
||||
log.Println("Exiting volumes", string(bytes))
|
||||
util.Post("http://"+mserver+"/dir/join", values)
|
||||
}
|
||||
func (s *Store) Close() {
|
||||
for _, v := range s.volumes {
|
||||
v.Close()
|
||||
}
|
||||
}
|
||||
func (s *Store) Write(i uint64, n *Needle) (uint32){
|
||||
func (s *Store) Write(i uint64, n *Needle) uint32 {
|
||||
return s.volumes[i].write(n)
|
||||
}
|
||||
func (s *Store) Read(i uint64, n *Needle) (int, os.Error){
|
||||
func (s *Store) Read(i uint64, n *Needle) (int, os.Error) {
|
||||
return s.volumes[i].read(n)
|
||||
}
|
||||
|
|
Loading…
Reference in a new issue