// handleBackup handles a request that intends to do a backup.
func handleBackup(c *cli.Context) {
srcSnap := path.Join(c.String("data-dir"), "member", "snap")
destSnap := path.Join(c.String("backup-dir"), "member", "snap")
srcWAL := path.Join(c.String("data-dir"), "member", "wal")
destWAL := path.Join(c.String("backup-dir"), "member", "wal")
if err := os.MkdirAll(destSnap, 0700); err != nil {
log.Fatalf("failed creating backup snapshot dir %v: %v", destSnap, err)
}
ss := snap.New(srcSnap)
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
log.Fatal(err)
}
var walsnap walpb.Snapshot
if snapshot != nil {
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
newss := snap.New(destSnap)
if err = newss.SaveSnap(*snapshot); err != nil {
log.Fatal(err)
}
}
w, err := wal.OpenForRead(srcWAL, walsnap)
if err != nil {
log.Fatal(err)
}
defer w.Close()
wmetadata, state, ents, err := w.ReadAll()
switch err {
case nil:
case wal.ErrSnapshotNotFound:
fmt.Printf("Failed to find the match snapshot record %+v in wal %v.", walsnap, srcWAL)
fmt.Printf("etcdctl will add it back. Start auto fixing...")
default:
log.Fatal(err)
}
var metadata etcdserverpb.Metadata
pbutil.MustUnmarshal(&metadata, wmetadata)
idgen := idutil.NewGenerator(0, time.Now())
metadata.NodeID = idgen.Next()
metadata.ClusterID = idgen.Next()
neww, err := wal.Create(destWAL, pbutil.MustMarshal(&metadata))
if err != nil {
log.Fatal(err)
}
defer neww.Close()
if err := neww.Save(state, ents); err != nil {
log.Fatal(err)
}
if err := neww.SaveSnapshot(walsnap); err != nil {
log.Fatal(err)
}
}
func (n *Node) loadAndStart(ctx context.Context, forceNewCluster bool) error {
walDir := n.walDir()
snapDir := n.snapDir()
if err := os.MkdirAll(snapDir, 0700); err != nil {
return fmt.Errorf("create snapshot directory error: %v", err)
}
// Create a snapshotter
n.snapshotter = snap.New(snapDir)
if !wal.Exist(walDir) {
return errNoWAL
}
// Load snapshot data
snapshot, err := n.snapshotter.Load()
if err != nil && err != snap.ErrNoSnapshot {
return err
}
if snapshot != nil {
// Load the snapshot data into the store
if err := n.restoreFromSnapshot(snapshot.Data, forceNewCluster); err != nil {
return err
}
}
// Read logs to fully catch up store
if err := n.readWAL(ctx, snapshot, forceNewCluster); err != nil {
return err
}
return nil
}
// NewSnapshotter returns a new Snapshotter with the given encrypters and decrypters
func (sc snapCryptor) New(dirpath string) Snapshotter {
return &wrappedSnap{
Snapshotter: snap.New(dirpath),
encrypter: sc.encrypter,
decrypter: sc.decrypter,
}
}
// handleBackup handles a request that intends to do a backup.
func handleBackup(c *cli.Context) {
srcSnap := path.Join(c.String("data-dir"), "snap")
destSnap := path.Join(c.String("backup-dir"), "snap")
srcWAL := path.Join(c.String("data-dir"), "wal")
destWAL := path.Join(c.String("backup-dir"), "wal")
if err := os.MkdirAll(destSnap, 0700); err != nil {
log.Fatalf("failed creating backup snapshot dir %v: %v", destSnap, err)
}
ss := snap.New(srcSnap)
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
log.Fatal(err)
}
var index uint64
if snapshot != nil {
index = snapshot.Metadata.Index
newss := snap.New(destSnap)
if err := newss.SaveSnap(*snapshot); err != nil {
log.Fatal(err)
}
}
w, err := wal.OpenNotInUse(srcWAL, index)
if err != nil {
log.Fatal(err)
}
defer w.Close()
wmetadata, state, ents, err := w.ReadAll()
if err != nil {
log.Fatal(err)
}
var metadata etcdserverpb.Metadata
pbutil.MustUnmarshal(&metadata, wmetadata)
rand.Seed(time.Now().UnixNano())
metadata.NodeID = etcdserver.GenID()
metadata.ClusterID = etcdserver.GenID()
neww, err := wal.Create(destWAL, pbutil.MustMarshal(&metadata))
if err != nil {
log.Fatal(err)
}
defer neww.Close()
if err := neww.Save(state, ents); err != nil {
log.Fatal(err)
}
}
// handleBackup handles a request that intends to do a backup.
func handleBackup(c *cli.Context) {
srcSnap := path.Join(c.String("data-dir"), "snap")
destSnap := path.Join(c.String("backup-dir"), "snap")
srcWAL := path.Join(c.String("data-dir"), "wal")
destWAL := path.Join(c.String("backup-dir"), "wal")
if err := os.MkdirAll(destSnap, 0700); err != nil {
log.Fatalf("failed creating backup snapshot dir %v: %v", destSnap, err)
}
ss := snap.New(srcSnap)
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
log.Fatal(err)
}
var walsnap walpb.Snapshot
if snapshot != nil {
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
newss := snap.New(destSnap)
if err := newss.SaveSnap(*snapshot); err != nil {
log.Fatal(err)
}
}
w, err := wal.OpenNotInUse(srcWAL, walsnap)
if err != nil {
log.Fatal(err)
}
defer w.Close()
wmetadata, state, ents, err := w.ReadAll()
if err != nil {
log.Fatal(err)
}
var metadata etcdserverpb.Metadata
pbutil.MustUnmarshal(&metadata, wmetadata)
idgen := idutil.NewGenerator(0, time.Now())
metadata.NodeID = idgen.Next()
metadata.ClusterID = idgen.Next()
neww, err := wal.Create(destWAL, pbutil.MustMarshal(&metadata))
if err != nil {
log.Fatal(err)
}
defer neww.Close()
if err := neww.Save(state, ents); err != nil {
log.Fatal(err)
}
}
func rebuildStoreV2() store.Store {
waldir := migrateWALdir
if len(waldir) == 0 {
waldir = path.Join(migrateDatadir, "member", "wal")
}
snapdir := path.Join(migrateDatadir, "member", "snap")
ss := snap.New(snapdir)
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
ExitWithError(ExitError, err)
}
var walsnap walpb.Snapshot
if snapshot != nil {
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
}
w, err := wal.OpenForRead(waldir, walsnap)
if err != nil {
ExitWithError(ExitError, err)
}
defer w.Close()
_, _, ents, err := w.ReadAll()
if err != nil {
ExitWithError(ExitError, err)
}
st := store.New()
if snapshot != nil {
err := st.Recovery(snapshot.Data)
if err != nil {
ExitWithError(ExitError, err)
}
}
applier := etcdserver.NewApplierV2(st, nil)
for _, ent := range ents {
if ent.Type != raftpb.EntryNormal {
continue
}
var raftReq pb.InternalRaftRequest
if !pbutil.MaybeUnmarshal(&raftReq, ent.Data) { // backward compatible
var r pb.Request
pbutil.MustUnmarshal(&r, ent.Data)
applyRequest(&r, applier)
} else {
if raftReq.V2 != nil {
req := raftReq.V2
applyRequest(req, applier)
}
}
}
return st
}
func (rc *raftNode) startRaft() {
if !fileutil.Exist(rc.snapdir) {
if err := os.Mkdir(rc.snapdir, 0750); err != nil {
log.Fatalf("raftexample: cannot create dir for snapshot (%v)", err)
}
}
rc.snapshotter = snap.New(rc.snapdir)
rc.snapshotterReady <- rc.snapshotter
oldwal := wal.Exist(rc.waldir)
rc.wal = rc.replayWAL()
rpeers := make([]raft.Peer, len(rc.peers))
for i := range rpeers {
rpeers[i] = raft.Peer{ID: uint64(i + 1)}
}
c := &raft.Config{
ID: uint64(rc.id),
ElectionTick: 10,
HeartbeatTick: 1,
Storage: rc.raftStorage,
MaxSizePerMsg: 1024 * 1024,
MaxInflightMsgs: 256,
}
if oldwal {
rc.node = raft.RestartNode(c)
} else {
startPeers := rpeers
if rc.join {
startPeers = nil
}
rc.node = raft.StartNode(c, startPeers)
}
ss := &stats.ServerStats{}
ss.Initialize()
rc.transport = &rafthttp.Transport{
ID: types.ID(rc.id),
ClusterID: 0x1000,
Raft: rc,
ServerStats: ss,
LeaderStats: stats.NewLeaderStats(strconv.Itoa(rc.id)),
ErrorC: make(chan error),
}
rc.transport.Start()
for i := range rc.peers {
if i+1 != rc.id {
rc.transport.AddPeer(types.ID(i+1), []string{rc.peers[i]})
}
}
go rc.serveRaft()
go rc.serveChannels()
}
func (n *Node) loadAndStart(ctx context.Context, forceNewCluster bool) error {
walDir := n.walDir()
snapDir := n.snapDir()
if !fileutil.Exist(snapDir) {
// If snapshots created by the etcd-v2 code exist, hard link
// them at the new path. This prevents etc-v2 creating
// snapshots that are visible to us, but out of sync with our
// WALs, after a downgrade.
legacySnapDir := n.legacySnapDir()
if fileutil.Exist(legacySnapDir) {
if err := migrateSnapshots(legacySnapDir, snapDir); err != nil {
return err
}
} else if err := os.MkdirAll(snapDir, 0700); err != nil {
return errors.Wrap(err, "failed to create snapshot directory")
}
}
// Create a snapshotter
n.snapshotter = snap.New(snapDir)
if !wal.Exist(walDir) {
// If wals created by the etcd-v2 wal code exist, copy them to
// the new path to avoid adding backwards-incompatible entries
// to those files.
legacyWALDir := n.legacyWALDir()
if !wal.Exist(legacyWALDir) {
return errNoWAL
}
if err := migrateWALs(legacyWALDir, walDir); err != nil {
return err
}
}
// Load snapshot data
snapshot, err := n.snapshotter.Load()
if err != nil && err != snap.ErrNoSnapshot {
return err
}
if snapshot != nil {
// Load the snapshot data into the store
if err := n.restoreFromSnapshot(snapshot.Data, forceNewCluster); err != nil {
return err
}
}
// Read logs to fully catch up store
if err := n.readWAL(ctx, snapshot, forceNewCluster); err != nil {
return err
}
return nil
}
func (n *Node) loadAndStart(ctx context.Context, forceNewCluster bool) error {
walDir := n.walDir()
snapDir := n.snapDir()
if err := os.MkdirAll(snapDir, 0700); err != nil {
return fmt.Errorf("create snapshot directory error: %v", err)
}
// Create a snapshotter
n.snapshotter = snap.New(snapDir)
if !wal.Exist(walDir) {
raftNode := &api.RaftMember{
RaftID: n.Config.ID,
Addr: n.Address,
}
metadata, err := raftNode.Marshal()
if err != nil {
return fmt.Errorf("error marshalling raft node: %v", err)
}
n.wal, err = wal.Create(walDir, metadata)
if err != nil {
return fmt.Errorf("create wal error: %v", err)
}
n.cluster.AddMember(&membership.Member{RaftMember: raftNode})
n.startNodePeers = []raft.Peer{{ID: n.Config.ID, Context: metadata}}
return nil
}
// Load snapshot data
snapshot, err := n.snapshotter.Load()
if err != nil && err != snap.ErrNoSnapshot {
return err
}
if snapshot != nil {
// Load the snapshot data into the store
if err := n.restoreFromSnapshot(snapshot.Data, forceNewCluster); err != nil {
return err
}
}
// Read logs to fully catch up store
if err := n.readWAL(ctx, snapshot, forceNewCluster); err != nil {
return err
}
n.Node = raft.RestartNode(n.Config)
return nil
}
func testSnapshotSend(t *testing.T, sm *snap.Message) (bool, []os.FileInfo) {
d, err := ioutil.TempDir(os.TempDir(), "snapdir")
if err != nil {
t.Fatal(err)
}
defer os.RemoveAll(d)
r := &fakeRaft{}
tr := &Transport{pipelineRt: &http.Transport{}, ClusterID: types.ID(1), Raft: r}
ch := make(chan struct{}, 1)
h := &syncHandler{newSnapshotHandler(tr, r, snap.New(d), types.ID(1)), ch}
srv := httptest.NewServer(h)
defer srv.Close()
picker := mustNewURLPicker(t, []string{srv.URL})
snapsend := newSnapshotSender(tr, picker, types.ID(1), newPeerStatus(types.ID(1)))
defer snapsend.stop()
snapsend.send(*sm)
sent := false
select {
case <-time.After(time.Second):
t.Fatalf("timed out sending snapshot")
case sent = <-sm.CloseNotify():
}
// wait for handler to finish accepting snapshot
<-ch
files, rerr := ioutil.ReadDir(d)
if rerr != nil {
t.Fatal(rerr)
}
return sent, files
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (*EtcdServer, error) {
st := store.New()
var w *wal.WAL
var n raft.Node
var s *raft.MemoryStorage
var id types.ID
walVersion, err := wal.DetectVersion(cfg.DataDir)
if err != nil {
return nil, err
}
if walVersion == wal.WALUnknown {
return nil, fmt.Errorf("unknown wal version in data dir %s", cfg.DataDir)
}
haveWAL := walVersion != wal.WALNotExist
if haveWAL && walVersion != wal.WALv0_5 {
err := UpgradeWAL(cfg, walVersion)
if err != nil {
return nil, err
}
}
ss := snap.New(cfg.SnapDir())
switch {
case !haveWAL && !cfg.NewCluster:
us := getOtherPeerURLs(cfg.Cluster, cfg.Name)
existingCluster, err := GetClusterFromPeers(us)
if err != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", err)
}
if err := ValidateClusterAndAssignIDs(cfg.Cluster, existingCluster); err != nil {
return nil, fmt.Errorf("error validating peerURLs %s: %v", existingCluster, err)
}
cfg.Cluster.SetID(existingCluster.id)
cfg.Cluster.SetStore(st)
cfg.Print()
id, n, s, w = startNode(cfg, nil)
case !haveWAL && cfg.NewCluster:
if err := cfg.VerifyBootstrapConfig(); err != nil {
return nil, err
}
if err := checkClientURLsEmptyFromPeers(cfg.Cluster, cfg.Name); err != nil {
return nil, err
}
m := cfg.Cluster.MemberByName(cfg.Name)
if cfg.ShouldDiscover() {
s, err := discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.Cluster.String())
if err != nil {
return nil, err
}
if cfg.Cluster, err = NewClusterFromString(cfg.Cluster.token, s); err != nil {
return nil, err
}
}
cfg.Cluster.SetStore(st)
cfg.PrintWithInitial()
id, n, s, w = startNode(cfg, cfg.Cluster.MemberIDs())
case haveWAL:
if cfg.ShouldDiscover() {
log.Printf("etcdserver: discovery token ignored since a cluster has already been initialized. Valid log found at %q", cfg.WALDir())
}
var index uint64
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
return nil, err
}
if snapshot != nil {
if err := st.Recovery(snapshot.Data); err != nil {
log.Panicf("etcdserver: recovered store from snapshot error: %v", err)
}
log.Printf("etcdserver: recovered store from snapshot at index %d", snapshot.Metadata.Index)
index = snapshot.Metadata.Index
}
cfg.Cluster = NewClusterFromStore(cfg.Cluster.token, st)
cfg.Print()
if snapshot != nil {
log.Printf("etcdserver: loaded cluster information from store: %s", cfg.Cluster)
}
if !cfg.ForceNewCluster {
id, n, s, w = restartNode(cfg, index+1, snapshot)
} else {
id, n, s, w = restartAsStandaloneNode(cfg, index+1, snapshot)
}
default:
return nil, fmt.Errorf("unsupported bootstrap config")
}
sstats := &stats.ServerStats{
Name: cfg.Name,
ID: id.String(),
}
lstats := stats.NewLeaderStats(id.String())
srv := &EtcdServer{
cfg: cfg,
store: st,
node: n,
raftStorage: s,
id: id,
attributes: Attributes{Name: cfg.Name, ClientURLs: cfg.ClientURLs.StringSlice()},
Cluster: cfg.Cluster,
storage: NewStorage(w, ss),
stats: sstats,
lstats: lstats,
Ticker: time.Tick(100 * time.Millisecond),
SyncTicker: time.Tick(500 * time.Millisecond),
snapCount: cfg.SnapCount,
}
srv.sendhub = newSendHub(cfg.Transport, cfg.Cluster, srv, sstats, lstats)
for _, m := range getOtherMembers(cfg.Cluster, cfg.Name) {
srv.sendhub.Add(m)
}
return srv, nil
}
func main() {
from := flag.String("data-dir", "", "")
snapfile := flag.String("start-snap", "", "The base name of snapshot file to start dumping")
index := flag.Uint64("start-index", 0, "The index to start dumping")
flag.Parse()
if *from == "" {
log.Fatal("Must provide -data-dir flag.")
}
if *snapfile != "" && *index != 0 {
log.Fatal("start-snap and start-index flags cannot be used together.")
}
var (
walsnap walpb.Snapshot
snapshot *raftpb.Snapshot
err error
)
isIndex := *index != 0
if isIndex {
fmt.Printf("Start dumping log entries from index %d.\n", *index)
walsnap.Index = *index
} else {
if *snapfile == "" {
ss := snap.New(snapDir(*from))
snapshot, err = ss.Load()
} else {
snapshot, err = snap.Read(path.Join(snapDir(*from), *snapfile))
}
switch err {
case nil:
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
nodes := genIDSlice(snapshot.Metadata.ConfState.Nodes)
fmt.Printf("Snapshot:\nterm=%d index=%d nodes=%s\n",
walsnap.Term, walsnap.Index, nodes)
case snap.ErrNoSnapshot:
fmt.Printf("Snapshot:\nempty\n")
default:
log.Fatalf("Failed loading snapshot: %v", err)
}
fmt.Println("Start dupmping log entries from snapshot.")
}
w, err := wal.OpenForRead(walDir(*from), walsnap)
if err != nil {
log.Fatalf("Failed opening WAL: %v", err)
}
wmetadata, state, ents, err := w.ReadAll()
w.Close()
if err != nil && (!isIndex || err != wal.ErrSnapshotNotFound) {
log.Fatalf("Failed reading WAL: %v", err)
}
id, cid := parseWALMetadata(wmetadata)
vid := types.ID(state.Vote)
fmt.Printf("WAL metadata:\nnodeID=%s clusterID=%s term=%d commitIndex=%d vote=%s\n",
id, cid, state.Term, state.Commit, vid)
fmt.Printf("WAL entries:\n")
fmt.Printf("lastIndex=%d\n", ents[len(ents)-1].Index)
fmt.Printf("%4s\t%10s\ttype\tdata\n", "term", "index")
for _, e := range ents {
msg := fmt.Sprintf("%4d\t%10d", e.Term, e.Index)
switch e.Type {
case raftpb.EntryNormal:
msg = fmt.Sprintf("%s\tnorm", msg)
var rr etcdserverpb.InternalRaftRequest
if err := rr.Unmarshal(e.Data); err == nil {
msg = fmt.Sprintf("%s\t%s", msg, rr.String())
break
}
var r etcdserverpb.Request
if err := r.Unmarshal(e.Data); err == nil {
switch r.Method {
case "":
msg = fmt.Sprintf("%s\tnoop", msg)
case "SYNC":
msg = fmt.Sprintf("%s\tmethod=SYNC time=%q", msg, time.Unix(0, r.Time))
case "QGET", "DELETE":
msg = fmt.Sprintf("%s\tmethod=%s path=%s", msg, r.Method, excerpt(r.Path, 64, 64))
default:
msg = fmt.Sprintf("%s\tmethod=%s path=%s val=%s", msg, r.Method, excerpt(r.Path, 64, 64), excerpt(r.Val, 128, 0))
}
break
}
msg = fmt.Sprintf("%s\t???", msg)
case raftpb.EntryConfChange:
msg = fmt.Sprintf("%s\tconf", msg)
var r raftpb.ConfChange
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s\t???", msg)
} else {
msg = fmt.Sprintf("%s\tmethod=%s id=%s", msg, r.Type, types.ID(r.NodeID))
}
}
fmt.Println(msg)
}
}
func rebuild(datadir string) ([]byte, *raftpb.HardState, store.Store, error) {
waldir := path.Join(datadir, "member", "wal")
snapdir := path.Join(datadir, "member", "snap")
ss := snap.New(snapdir)
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
return nil, nil, nil, err
}
var walsnap walpb.Snapshot
if snapshot != nil {
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
}
w, err := wal.OpenForRead(waldir, walsnap)
if err != nil {
return nil, nil, nil, err
}
defer w.Close()
meta, hardstate, ents, err := w.ReadAll()
if err != nil {
return nil, nil, nil, err
}
st := store.New(etcdserver.StoreClusterPrefix, etcdserver.StoreKeysPrefix)
if snapshot != nil {
err := st.Recovery(snapshot.Data)
if err != nil {
return nil, nil, nil, err
}
}
cluster := membership.NewCluster("")
cluster.SetStore(st)
cluster.Recover(func(*semver.Version) {})
applier := etcdserver.NewApplierV2(st, cluster)
for _, ent := range ents {
if ent.Type == raftpb.EntryConfChange {
var cc raftpb.ConfChange
pbutil.MustUnmarshal(&cc, ent.Data)
switch cc.Type {
case raftpb.ConfChangeAddNode:
m := new(membership.Member)
if err := json.Unmarshal(cc.Context, m); err != nil {
return nil, nil, nil, err
}
cluster.AddMember(m)
case raftpb.ConfChangeRemoveNode:
id := types.ID(cc.NodeID)
cluster.RemoveMember(id)
case raftpb.ConfChangeUpdateNode:
m := new(membership.Member)
if err := json.Unmarshal(cc.Context, m); err != nil {
return nil, nil, nil, err
}
cluster.UpdateRaftAttributes(m.ID, m.RaftAttributes)
}
continue
}
var raftReq pb.InternalRaftRequest
if !pbutil.MaybeUnmarshal(&raftReq, ent.Data) { // backward compatible
var r pb.Request
pbutil.MustUnmarshal(&r, ent.Data)
applyRequest(&r, applier)
} else {
if raftReq.V2 != nil {
req := raftReq.V2
applyRequest(req, applier)
}
}
}
return meta, &hardstate, st, nil
}
func main() {
flag.Parse()
if len(*migrateDatadir) == 0 {
glog.Fatal("need to set '--data-dir'")
}
dbpath := path.Join(*migrateDatadir, "member", "snap", "db")
// etcd3 store backend. We will use it to parse v3 data files and extract information.
be := backend.NewDefaultBackend(dbpath)
tx := be.BatchTx()
// etcd2 store backend. We will use v3 data to update this and then save snapshot to disk.
st := store.New(etcdserver.StoreClusterPrefix, etcdserver.StoreKeysPrefix)
expireTime := time.Now().Add(*ttl)
tx.Lock()
err := tx.UnsafeForEach([]byte("key"), func(k, v []byte) error {
kv := &mvccpb.KeyValue{}
kv.Unmarshal(v)
// This is compact key.
if !strings.HasPrefix(string(kv.Key), "/") {
return nil
}
ttlOpt := store.TTLOptionSet{}
if kv.Lease != 0 {
ttlOpt = store.TTLOptionSet{ExpireTime: expireTime}
}
if !isTombstone(k) {
sk := path.Join(strings.Trim(etcdserver.StoreKeysPrefix, "/"), string(kv.Key))
_, err := st.Set(sk, false, string(kv.Value), ttlOpt)
if err != nil {
return err
}
} else {
st.Delete(string(kv.Key), false, false)
}
return nil
})
if err != nil {
glog.Fatal(err)
}
tx.Unlock()
if err := traverseAndDeleteEmptyDir(st, "/"); err != nil {
glog.Fatal(err)
}
// rebuild cluster state.
metadata, hardstate, oldSt, err := rebuild(*migrateDatadir)
if err != nil {
glog.Fatal(err)
}
// In the following, it's low level logic that saves metadata and data into v2 snapshot.
backupPath := *migrateDatadir + ".rollback.backup"
if err := os.Rename(*migrateDatadir, backupPath); err != nil {
glog.Fatal(err)
}
if err := os.MkdirAll(path.Join(*migrateDatadir, "member", "snap"), 0700); err != nil {
glog.Fatal(err)
}
walDir := path.Join(*migrateDatadir, "member", "wal")
w, err := wal.Create(walDir, metadata)
if err != nil {
glog.Fatal(err)
}
err = w.SaveSnapshot(walpb.Snapshot{Index: hardstate.Commit, Term: hardstate.Term})
if err != nil {
glog.Fatal(err)
}
w.Close()
event, err := oldSt.Get(etcdserver.StoreClusterPrefix, true, false)
if err != nil {
glog.Fatal(err)
}
// nodes (members info) for ConfState
nodes := []uint64{}
traverseMetadata(event.Node, func(n *store.NodeExtern) {
if n.Key != etcdserver.StoreClusterPrefix {
// update store metadata
v := ""
if !n.Dir {
v = *n.Value
}
if n.Key == path.Join(etcdserver.StoreClusterPrefix, "version") {
v = rollbackVersion
}
if _, err := st.Set(n.Key, n.Dir, v, store.TTLOptionSet{}); err != nil {
glog.Fatal(err)
}
// update nodes
fields := strings.Split(n.Key, "/")
if len(fields) == 4 && fields[2] == "members" {
nodeID, err := strconv.ParseUint(fields[3], 16, 64)
if err != nil {
glog.Fatalf("failed to parse member ID (%s): %v", fields[3], err)
}
nodes = append(nodes, nodeID)
}
}
})
data, err := st.Save()
if err != nil {
glog.Fatal(err)
}
raftSnap := raftpb.Snapshot{
Data: data,
Metadata: raftpb.SnapshotMetadata{
Index: hardstate.Commit,
Term: hardstate.Term,
ConfState: raftpb.ConfState{
Nodes: nodes,
},
},
}
snapshotter := snap.New(path.Join(*migrateDatadir, "member", "snap"))
if err := snapshotter.SaveSnap(raftSnap); err != nil {
glog.Fatal(err)
}
fmt.Println("Finished successfully")
}
func (s *snapTransporter) SendSnapshot(m snap.Message) {
ss := snap.New(s.snapDir)
ss.SaveDBFrom(m.ReadCloser, m.Snapshot.Metadata.Index+1)
m.CloseWithError(nil)
s.snapDoneC <- m
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (*EtcdServer, error) {
st := store.New(StoreClusterPrefix, StoreKeysPrefix)
var w *wal.WAL
var n raft.Node
var s *raft.MemoryStorage
var id types.ID
var cl *cluster
// Run the migrations.
dataVer, err := version.DetectDataDir(cfg.DataDir)
if err != nil {
return nil, err
}
if err := upgradeDataDir(cfg.DataDir, cfg.Name, dataVer); err != nil {
return nil, err
}
haveWAL := wal.Exist(cfg.WALDir())
ss := snap.New(cfg.SnapDir())
var remotes []*Member
switch {
case !haveWAL && !cfg.NewCluster:
if err := cfg.VerifyJoinExisting(); err != nil {
return nil, err
}
cl, err = newClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
existingCluster, err := GetClusterFromRemotePeers(getRemotePeerURLs(cl, cfg.Name), cfg.Transport)
if err != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", err)
}
if err := ValidateClusterAndAssignIDs(cl, existingCluster); err != nil {
return nil, fmt.Errorf("error validating peerURLs %s: %v", existingCluster, err)
}
if !isCompatibleWithCluster(cl, cl.MemberByName(cfg.Name).ID, cfg.Transport) {
return nil, fmt.Errorf("incomptible with current running cluster")
}
remotes = existingCluster.Members()
cl.SetID(existingCluster.id)
cl.SetStore(st)
cfg.Print()
id, n, s, w = startNode(cfg, cl, nil)
case !haveWAL && cfg.NewCluster:
if err := cfg.VerifyBootstrap(); err != nil {
return nil, err
}
cl, err = newClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
m := cl.MemberByName(cfg.Name)
if isMemberBootstrapped(cl, cfg.Name, cfg.Transport) {
return nil, fmt.Errorf("member %s has already been bootstrapped", m.ID)
}
if cfg.ShouldDiscover() {
str, err := discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.InitialPeerURLsMap.String())
if err != nil {
return nil, err
}
urlsmap, err := types.NewURLsMap(str)
if err != nil {
return nil, err
}
if checkDuplicateURL(urlsmap) {
return nil, fmt.Errorf("discovery cluster %s has duplicate url", urlsmap)
}
if cl, err = newClusterFromURLsMap(cfg.InitialClusterToken, urlsmap); err != nil {
return nil, err
}
}
cl.SetStore(st)
cfg.PrintWithInitial()
id, n, s, w = startNode(cfg, cl, cl.MemberIDs())
case haveWAL:
if err := fileutil.IsDirWriteable(cfg.DataDir); err != nil {
return nil, fmt.Errorf("cannot write to data directory: %v", err)
}
if err := fileutil.IsDirWriteable(cfg.MemberDir()); err != nil {
return nil, fmt.Errorf("cannot write to member directory: %v", err)
}
if cfg.ShouldDiscover() {
plog.Warningf("discovery token ignored since a cluster has already been initialized. Valid log found at %q", cfg.WALDir())
}
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
return nil, err
}
if snapshot != nil {
if err := st.Recovery(snapshot.Data); err != nil {
plog.Panicf("recovered store from snapshot error: %v", err)
}
plog.Infof("recovered store from snapshot at index %d", snapshot.Metadata.Index)
}
cfg.Print()
if snapshot != nil {
plog.Infof("loaded cluster information from store: %s", cl)
}
if !cfg.ForceNewCluster {
id, cl, n, s, w = restartNode(cfg, snapshot)
} else {
id, cl, n, s, w = restartAsStandaloneNode(cfg, snapshot)
}
cl.SetStore(st)
cl.Recover()
default:
return nil, fmt.Errorf("unsupported bootstrap config")
}
sstats := &stats.ServerStats{
Name: cfg.Name,
ID: id.String(),
}
sstats.Initialize()
lstats := stats.NewLeaderStats(id.String())
srv := &EtcdServer{
cfg: cfg,
snapCount: cfg.SnapCount,
errorc: make(chan error, 1),
store: st,
r: raftNode{
Node: n,
ticker: time.Tick(time.Duration(cfg.TickMs) * time.Millisecond),
raftStorage: s,
storage: NewStorage(w, ss),
},
id: id,
attributes: Attributes{Name: cfg.Name, ClientURLs: cfg.ClientURLs.StringSlice()},
cluster: cl,
stats: sstats,
lstats: lstats,
SyncTicker: time.Tick(500 * time.Millisecond),
reqIDGen: idutil.NewGenerator(uint8(id), time.Now()),
forceVersionC: make(chan struct{}),
}
// TODO: move transport initialization near the definition of remote
tr := rafthttp.NewTransporter(cfg.Transport, id, cl.ID(), srv, srv.errorc, sstats, lstats)
// add all remotes into transport
for _, m := range remotes {
if m.ID != id {
tr.AddRemote(m.ID, m.PeerURLs)
}
}
for _, m := range cl.Members() {
if m.ID != id {
tr.AddPeer(m.ID, m.PeerURLs)
}
}
srv.r.transport = tr
return srv, nil
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) *EtcdServer {
m := cfg.Cluster.FindName(cfg.Name)
if m == nil {
// Should never happen
log.Fatalf("could not find name %v in cluster!", cfg.Name)
}
snapdir := path.Join(cfg.DataDir, "snap")
if err := os.MkdirAll(snapdir, privateDirMode); err != nil {
log.Fatalf("etcdserver: cannot create snapshot directory: %v", err)
}
ss := snap.New(snapdir)
st := store.New()
var w *wal.WAL
var n raft.Node
var err error
waldir := path.Join(cfg.DataDir, "wal")
if !wal.Exist(waldir) {
if w, err = wal.Create(waldir); err != nil {
log.Fatal(err)
}
n = raft.StartNode(m.ID, cfg.Cluster.IDs(), 10, 1)
} else {
var index int64
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
log.Fatal(err)
}
if snapshot != nil {
log.Printf("etcdserver: restart from snapshot at index %d", snapshot.Index)
st.Recovery(snapshot.Data)
index = snapshot.Index
}
// restart a node from previous wal
if w, err = wal.OpenAtIndex(waldir, index); err != nil {
log.Fatal(err)
}
wid, st, ents, err := w.ReadAll()
if err != nil {
log.Fatal(err)
}
// TODO(xiangli): save/recovery nodeID?
if wid != 0 {
log.Fatalf("unexpected nodeid %d: nodeid should always be zero until we save nodeid into wal", wid)
}
n = raft.RestartNode(m.ID, cfg.Cluster.IDs(), 10, 1, snapshot, st, ents)
}
cls := NewClusterStore(st, *cfg.Cluster)
s := &EtcdServer{
store: st,
node: n,
name: cfg.Name,
storage: struct {
*wal.WAL
*snap.Snapshotter
}{w, ss},
send: Sender(cfg.Transport, cls),
clientURLs: cfg.ClientURLs,
ticker: time.Tick(100 * time.Millisecond),
syncTicker: time.Tick(500 * time.Millisecond),
snapCount: cfg.SnapCount,
ClusterStore: cls,
}
return s
}
func Migrate4To2(dataDir string, name string) error {
// prep new directories
sd2 := snapDir2(dataDir)
if err := os.MkdirAll(sd2, 0700); err != nil {
return fmt.Errorf("failed creating snapshot directory %s: %v", sd2, err)
}
// read v0.4 data
snap4, err := DecodeLatestSnapshot4FromDir(snapDir4(dataDir))
if err != nil {
return err
}
cfg4, err := DecodeConfig4FromFile(cfgFile4(dataDir))
if err != nil {
return err
}
ents4, err := DecodeLog4FromFile(logFile4(dataDir))
if err != nil {
return err
}
nodeIDs := ents4.NodeIDs()
nodeID := GuessNodeID(nodeIDs, snap4, cfg4, name)
if nodeID == 0 {
return fmt.Errorf("Couldn't figure out the node ID from the log or flags, cannot convert")
}
metadata := pbutil.MustMarshal(&pb.Metadata{NodeID: nodeID, ClusterID: 0x04add5})
wd2 := walDir2(dataDir)
w, err := wal.Create(wd2, metadata)
if err != nil {
return fmt.Errorf("failed initializing wal at %s: %v", wd2, err)
}
defer w.Close()
// transform v0.4 data
var snap2 *raftpb.Snapshot
if snap4 == nil {
log.Printf("No snapshot found")
} else {
log.Printf("Found snapshot: lastIndex=%d", snap4.LastIndex)
snap2 = snap4.Snapshot2()
}
st2 := cfg4.HardState2()
// If we've got the most recent snapshot, we can use it's committed index. Still likely less than the current actual index, but worth it for the replay.
if snap2 != nil {
st2.Commit = snap2.Metadata.Index
}
ents2, err := Entries4To2(ents4)
if err != nil {
return err
}
ents2Len := len(ents2)
log.Printf("Found %d log entries: firstIndex=%d lastIndex=%d", ents2Len, ents2[0].Index, ents2[ents2Len-1].Index)
// set the state term to the biggest term we have ever seen,
// so term of future entries will not be the same with term of old ones.
st2.Term = ents2[ents2Len-1].Term
// explicitly prepend an empty entry as the WAL code expects it
ents2 = append(make([]raftpb.Entry, 1), ents2...)
if err = w.Save(st2, ents2); err != nil {
return err
}
log.Printf("Log migration successful")
// migrate snapshot (if necessary) and logs
var walsnap walpb.Snapshot
if snap2 != nil {
walsnap.Index, walsnap.Term = snap2.Metadata.Index, snap2.Metadata.Term
ss := snap.New(sd2)
if err := ss.SaveSnap(*snap2); err != nil {
return err
}
log.Printf("Snapshot migration successful")
}
if err = w.SaveSnapshot(walsnap); err != nil {
return err
}
return nil
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) *EtcdServer {
if err := os.MkdirAll(cfg.SnapDir(), privateDirMode); err != nil {
log.Fatalf("etcdserver: cannot create snapshot directory: %v", err)
}
ss := snap.New(cfg.SnapDir())
st := store.New()
var w *wal.WAL
var n raft.Node
var id uint64
haveWAL := wal.Exist(cfg.WALDir())
switch {
case !haveWAL && cfg.ClusterState == ClusterStateValueExisting:
cl, err := GetClusterFromPeers(cfg.Cluster.PeerURLs())
if err != nil {
log.Fatal(err)
}
if err := cfg.Cluster.ValidateAndAssignIDs(cl.Members()); err != nil {
log.Fatalf("etcdserver: %v", err)
}
cfg.Cluster.SetID(cl.id)
cfg.Cluster.SetStore(st)
id, n, w = startNode(cfg, nil)
case !haveWAL && cfg.ClusterState == ClusterStateValueNew:
if err := cfg.VerifyBootstrapConfig(); err != nil {
log.Fatalf("etcdserver: %v", err)
}
m := cfg.Cluster.MemberByName(cfg.Name)
if cfg.ShouldDiscover() {
d, err := discovery.New(cfg.DiscoveryURL, m.ID, cfg.Cluster.String())
if err != nil {
log.Fatalf("etcdserver: cannot init discovery %v", err)
}
s, err := d.Discover()
if err != nil {
log.Fatalf("etcdserver: %v", err)
}
if cfg.Cluster, err = NewClusterFromString(cfg.Cluster.name, s); err != nil {
log.Fatalf("etcdserver: %v", err)
}
}
cfg.Cluster.SetStore(st)
id, n, w = startNode(cfg, cfg.Cluster.MemberIDs())
case haveWAL:
if cfg.ShouldDiscover() {
log.Printf("etcdserver: warn: ignoring discovery: etcd has already been initialized and has a valid log in %q", cfg.WALDir())
}
var index uint64
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
log.Fatal(err)
}
if snapshot != nil {
log.Printf("etcdserver: recovering from snapshot at index %d", snapshot.Index)
st.Recovery(snapshot.Data)
index = snapshot.Index
}
cfg.Cluster = NewClusterFromStore(cfg.Cluster.name, st)
id, n, w = restartNode(cfg, index, snapshot)
default:
log.Fatalf("etcdserver: unsupported bootstrap config")
}
sstats := &stats.ServerStats{
Name: cfg.Name,
ID: strutil.IDAsHex(id),
}
lstats := stats.NewLeaderStats(strutil.IDAsHex(id))
s := &EtcdServer{
store: st,
node: n,
id: id,
attributes: Attributes{Name: cfg.Name, ClientURLs: cfg.ClientURLs.StringSlice()},
Cluster: cfg.Cluster,
storage: struct {
*wal.WAL
*snap.Snapshotter
}{w, ss},
stats: sstats,
lstats: lstats,
send: Sender(cfg.Transport, cfg.Cluster, sstats, lstats),
Ticker: time.Tick(100 * time.Millisecond),
SyncTicker: time.Tick(500 * time.Millisecond),
snapCount: cfg.SnapCount,
}
return s
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (*EtcdServer, error) {
st := store.New(StoreAdminPrefix, StoreKeysPrefix)
var w *wal.WAL
var n raft.Node
var s *raft.MemoryStorage
var id types.ID
walVersion, err := wal.DetectVersion(cfg.DataDir)
if err != nil {
return nil, err
}
if walVersion == wal.WALUnknown {
return nil, fmt.Errorf("unknown wal version in data dir %s", cfg.DataDir)
}
haveWAL := walVersion != wal.WALNotExist
ss := snap.New(cfg.SnapDir())
switch {
case !haveWAL && !cfg.NewCluster:
us := getOtherPeerURLs(cfg.Cluster, cfg.Name)
existingCluster, err := GetClusterFromPeers(us, cfg.Transport)
if err != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", err)
}
if err := ValidateClusterAndAssignIDs(cfg.Cluster, existingCluster); err != nil {
return nil, fmt.Errorf("error validating peerURLs %s: %v", existingCluster, err)
}
cfg.Cluster.SetID(existingCluster.id)
cfg.Cluster.SetStore(st)
cfg.Print()
id, n, s, w = startNode(cfg, nil)
case !haveWAL && cfg.NewCluster:
if err := cfg.VerifyBootstrapConfig(); err != nil {
return nil, err
}
m := cfg.Cluster.MemberByName(cfg.Name)
if isBootstrapped(cfg) {
return nil, fmt.Errorf("member %s has already been bootstrapped", m.ID)
}
if cfg.ShouldDiscover() {
s, err := discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.Cluster.String())
if err != nil {
return nil, err
}
if cfg.Cluster, err = NewClusterFromString(cfg.Cluster.token, s); err != nil {
return nil, err
}
}
cfg.Cluster.SetStore(st)
cfg.PrintWithInitial()
id, n, s, w = startNode(cfg, cfg.Cluster.MemberIDs())
case haveWAL:
if walVersion != wal.WALv0_5 {
if err := upgradeWAL(cfg, walVersion); err != nil {
return nil, err
}
}
if cfg.ShouldDiscover() {
log.Printf("etcdserver: discovery token ignored since a cluster has already been initialized. Valid log found at %q", cfg.WALDir())
}
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
return nil, err
}
if snapshot != nil {
if err := st.Recovery(snapshot.Data); err != nil {
log.Panicf("etcdserver: recovered store from snapshot error: %v", err)
}
log.Printf("etcdserver: recovered store from snapshot at index %d", snapshot.Metadata.Index)
}
cfg.Cluster = NewClusterFromStore(cfg.Cluster.token, st)
cfg.Print()
if snapshot != nil {
log.Printf("etcdserver: loaded cluster information from store: %s", cfg.Cluster)
}
if !cfg.ForceNewCluster {
id, n, s, w = restartNode(cfg, snapshot)
} else {
id, n, s, w = restartAsStandaloneNode(cfg, snapshot)
}
default:
return nil, fmt.Errorf("unsupported bootstrap config")
}
sstats := &stats.ServerStats{
Name: cfg.Name,
ID: id.String(),
}
lstats := stats.NewLeaderStats(id.String())
srv := &EtcdServer{
cfg: cfg,
errorc: make(chan error, 1),
store: st,
r: raftNode{
Node: n,
snapCount: cfg.SnapCount,
ticker: time.Tick(time.Duration(cfg.TickMs) * time.Millisecond),
raftStorage: s,
storage: NewStorage(w, ss),
},
id: id,
attributes: Attributes{Name: cfg.Name, ClientURLs: cfg.ClientURLs.StringSlice()},
Cluster: cfg.Cluster,
stats: sstats,
lstats: lstats,
SyncTicker: time.Tick(500 * time.Millisecond),
reqIDGen: idutil.NewGenerator(uint8(id), time.Now()),
}
tr := rafthttp.NewTransporter(cfg.Transport, id, cfg.Cluster.ID(), srv, srv.errorc, sstats, lstats)
// add all the remote members into sendhub
for _, m := range cfg.Cluster.Members() {
if m.ID != id {
tr.AddPeer(m.ID, m.PeerURLs)
}
}
srv.r.transport = tr
return srv, nil
}
func rebuildStoreV2() (store.Store, uint64) {
var index uint64
cl := membership.NewCluster("")
waldir := migrateWALdir
if len(waldir) == 0 {
waldir = path.Join(migrateDatadir, "member", "wal")
}
snapdir := path.Join(migrateDatadir, "member", "snap")
ss := snap.New(snapdir)
snapshot, err := ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
ExitWithError(ExitError, err)
}
var walsnap walpb.Snapshot
if snapshot != nil {
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
index = snapshot.Metadata.Index
}
w, err := wal.OpenForRead(waldir, walsnap)
if err != nil {
ExitWithError(ExitError, err)
}
defer w.Close()
_, _, ents, err := w.ReadAll()
if err != nil {
ExitWithError(ExitError, err)
}
st := store.New()
if snapshot != nil {
err := st.Recovery(snapshot.Data)
if err != nil {
ExitWithError(ExitError, err)
}
}
cl.SetStore(st)
cl.Recover(api.UpdateCapability)
applier := etcdserver.NewApplierV2(st, cl)
for _, ent := range ents {
if ent.Type == raftpb.EntryConfChange {
var cc raftpb.ConfChange
pbutil.MustUnmarshal(&cc, ent.Data)
applyConf(cc, cl)
continue
}
var raftReq pb.InternalRaftRequest
if !pbutil.MaybeUnmarshal(&raftReq, ent.Data) { // backward compatible
var r pb.Request
pbutil.MustUnmarshal(&r, ent.Data)
applyRequest(&r, applier)
} else {
if raftReq.V2 != nil {
req := raftReq.V2
applyRequest(req, applier)
}
}
if ent.Index > index {
index = ent.Index
}
}
return st, index
}
// makeWAL creates a WAL for the initial cluster
func makeWALAndSnap(waldir, snapdir string, cl *membership.RaftCluster) {
if err := fileutil.CreateDirAll(waldir); err != nil {
ExitWithError(ExitIO, err)
}
// add members again to persist them to the store we create.
st := store.New(etcdserver.StoreClusterPrefix, etcdserver.StoreKeysPrefix)
cl.SetStore(st)
for _, m := range cl.Members() {
cl.AddMember(m)
}
m := cl.MemberByName(restoreName)
md := &etcdserverpb.Metadata{NodeID: uint64(m.ID), ClusterID: uint64(cl.ID())}
metadata, merr := md.Marshal()
if merr != nil {
ExitWithError(ExitInvalidInput, merr)
}
w, walerr := wal.Create(waldir, metadata)
if walerr != nil {
ExitWithError(ExitIO, walerr)
}
defer w.Close()
peers := make([]raft.Peer, len(cl.MemberIDs()))
for i, id := range cl.MemberIDs() {
ctx, err := json.Marshal((*cl).Member(id))
if err != nil {
ExitWithError(ExitInvalidInput, err)
}
peers[i] = raft.Peer{ID: uint64(id), Context: ctx}
}
ents := make([]raftpb.Entry, len(peers))
nodeIDs := make([]uint64, len(peers))
for i, p := range peers {
nodeIDs[i] = p.ID
cc := raftpb.ConfChange{
Type: raftpb.ConfChangeAddNode,
NodeID: p.ID,
Context: p.Context}
d, err := cc.Marshal()
if err != nil {
ExitWithError(ExitInvalidInput, err)
}
e := raftpb.Entry{
Type: raftpb.EntryConfChange,
Term: 1,
Index: uint64(i + 1),
Data: d,
}
ents[i] = e
}
commit, term := uint64(len(ents)), uint64(1)
if err := w.Save(raftpb.HardState{
Term: term,
Vote: peers[0].ID,
Commit: commit}, ents); err != nil {
ExitWithError(ExitIO, err)
}
b, berr := st.Save()
if berr != nil {
ExitWithError(ExitError, berr)
}
raftSnap := raftpb.Snapshot{
Data: b,
Metadata: raftpb.SnapshotMetadata{
Index: commit,
Term: term,
ConfState: raftpb.ConfState{
Nodes: nodeIDs,
},
},
}
snapshotter := snap.New(snapdir)
if err := snapshotter.SaveSnap(raftSnap); err != nil {
panic(err)
}
if err := w.SaveSnapshot(walpb.Snapshot{Index: commit, Term: term}); err != nil {
ExitWithError(ExitIO, err)
}
}
func Migrate4To5(dataDir string, name string) error {
// prep new directories
sd5 := snapDir5(dataDir)
if err := os.MkdirAll(sd5, 0700); err != nil {
return fmt.Errorf("failed creating snapshot directory %s: %v", sd5, err)
}
// read v0.4 data
snap4, err := DecodeLatestSnapshot4FromDir(snapDir4(dataDir))
if err != nil {
return err
}
cfg4, err := DecodeConfig4FromFile(cfgFile4(dataDir))
if err != nil {
return err
}
ents4, err := DecodeLog4FromFile(logFile4(dataDir))
if err != nil {
return err
}
nodeIDs := ents4.NodeIDs()
nodeID := GuessNodeID(nodeIDs, snap4, cfg4, name)
if nodeID == 0 {
return fmt.Errorf("Couldn't figure out the node ID from the log or flags, cannot convert")
}
metadata := pbutil.MustMarshal(&pb.Metadata{NodeID: nodeID, ClusterID: 0x04add5})
wd5 := walDir5(dataDir)
w, err := wal.Create(wd5, metadata)
if err != nil {
return fmt.Errorf("failed initializing wal at %s: %v", wd5, err)
}
defer w.Close()
// transform v0.4 data
var snap5 *raftpb.Snapshot
if snap4 == nil {
log.Printf("No snapshot found")
} else {
log.Printf("Found snapshot: lastIndex=%d", snap4.LastIndex)
snap5 = snap4.Snapshot5()
}
st5 := cfg4.HardState5()
// If we've got the most recent snapshot, we can use it's committed index. Still likely less than the current actual index, but worth it for the replay.
if snap5 != nil {
st5.Commit = snap5.Metadata.Index
}
ents5, err := Entries4To5(ents4)
if err != nil {
return err
}
ents5Len := len(ents5)
log.Printf("Found %d log entries: firstIndex=%d lastIndex=%d", ents5Len, ents5[0].Index, ents5[ents5Len-1].Index)
// explicitly prepend an empty entry as the WAL code expects it
ents5 = append(make([]raftpb.Entry, 1), ents5...)
if err = w.Save(st5, ents5); err != nil {
return err
}
log.Printf("Log migration successful")
// migrate snapshot (if necessary) and logs
if snap5 != nil {
ss := snap.New(sd5)
if err := ss.SaveSnap(*snap5); err != nil {
return err
}
log.Printf("Snapshot migration successful")
}
return nil
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (srv *EtcdServer, err error) {
st := store.New(StoreClusterPrefix, StoreKeysPrefix)
var (
w *wal.WAL
n raft.Node
s *raft.MemoryStorage
id types.ID
cl *membership.RaftCluster
)
if terr := fileutil.TouchDirAll(cfg.DataDir); terr != nil {
return nil, fmt.Errorf("cannot access data directory: %v", terr)
}
haveWAL := wal.Exist(cfg.WALDir())
if err = fileutil.TouchDirAll(cfg.SnapDir()); err != nil {
plog.Fatalf("create snapshot directory error: %v", err)
}
ss := snap.New(cfg.SnapDir())
bepath := path.Join(cfg.SnapDir(), databaseFilename)
beExist := fileutil.Exist(bepath)
var be backend.Backend
beOpened := make(chan struct{})
go func() {
be = backend.NewDefaultBackend(bepath)
beOpened <- struct{}{}
}()
select {
case <-beOpened:
case <-time.After(time.Second):
plog.Warningf("another etcd process is running with the same data dir and holding the file lock.")
plog.Warningf("waiting for it to exit before starting...")
<-beOpened
}
defer func() {
if err != nil {
be.Close()
}
}()
prt, err := rafthttp.NewRoundTripper(cfg.PeerTLSInfo, cfg.peerDialTimeout())
if err != nil {
return nil, err
}
var (
remotes []*membership.Member
snapshot *raftpb.Snapshot
)
switch {
case !haveWAL && !cfg.NewCluster:
if err = cfg.VerifyJoinExisting(); err != nil {
return nil, err
}
cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
existingCluster, gerr := GetClusterFromRemotePeers(getRemotePeerURLs(cl, cfg.Name), prt)
if gerr != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", gerr)
}
if err = membership.ValidateClusterAndAssignIDs(cl, existingCluster); err != nil {
return nil, fmt.Errorf("error validating peerURLs %s: %v", existingCluster, err)
}
if !isCompatibleWithCluster(cl, cl.MemberByName(cfg.Name).ID, prt) {
return nil, fmt.Errorf("incompatible with current running cluster")
}
remotes = existingCluster.Members()
cl.SetID(existingCluster.ID())
cl.SetStore(st)
cl.SetBackend(be)
cfg.Print()
id, n, s, w = startNode(cfg, cl, nil)
case !haveWAL && cfg.NewCluster:
if err = cfg.VerifyBootstrap(); err != nil {
return nil, err
}
cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
m := cl.MemberByName(cfg.Name)
if isMemberBootstrapped(cl, cfg.Name, prt, cfg.bootstrapTimeout()) {
return nil, fmt.Errorf("member %s has already been bootstrapped", m.ID)
}
if cfg.ShouldDiscover() {
var str string
str, err = discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.InitialPeerURLsMap.String())
if err != nil {
return nil, &DiscoveryError{Op: "join", Err: err}
}
var urlsmap types.URLsMap
urlsmap, err = types.NewURLsMap(str)
if err != nil {
return nil, err
}
if checkDuplicateURL(urlsmap) {
return nil, fmt.Errorf("discovery cluster %s has duplicate url", urlsmap)
}
if cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, urlsmap); err != nil {
return nil, err
}
}
cl.SetStore(st)
cl.SetBackend(be)
cfg.PrintWithInitial()
id, n, s, w = startNode(cfg, cl, cl.MemberIDs())
case haveWAL:
if err = fileutil.IsDirWriteable(cfg.MemberDir()); err != nil {
return nil, fmt.Errorf("cannot write to member directory: %v", err)
}
if err = fileutil.IsDirWriteable(cfg.WALDir()); err != nil {
return nil, fmt.Errorf("cannot write to WAL directory: %v", err)
}
if cfg.ShouldDiscover() {
plog.Warningf("discovery token ignored since a cluster has already been initialized. Valid log found at %q", cfg.WALDir())
}
snapshot, err = ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
return nil, err
}
if snapshot != nil {
if err = st.Recovery(snapshot.Data); err != nil {
plog.Panicf("recovered store from snapshot error: %v", err)
}
plog.Infof("recovered store from snapshot at index %d", snapshot.Metadata.Index)
}
cfg.Print()
if !cfg.ForceNewCluster {
id, cl, n, s, w = restartNode(cfg, snapshot)
} else {
id, cl, n, s, w = restartAsStandaloneNode(cfg, snapshot)
}
cl.SetStore(st)
cl.SetBackend(be)
cl.Recover(api.UpdateCapability)
if cl.Version() != nil && !cl.Version().LessThan(semver.Version{Major: 3}) && !beExist {
os.RemoveAll(bepath)
return nil, fmt.Errorf("database file (%v) of the backend is missing", bepath)
}
default:
return nil, fmt.Errorf("unsupported bootstrap config")
}
if terr := fileutil.TouchDirAll(cfg.MemberDir()); terr != nil {
return nil, fmt.Errorf("cannot access member directory: %v", terr)
}
sstats := &stats.ServerStats{
Name: cfg.Name,
ID: id.String(),
}
sstats.Initialize()
lstats := stats.NewLeaderStats(id.String())
heartbeat := time.Duration(cfg.TickMs) * time.Millisecond
srv = &EtcdServer{
readych: make(chan struct{}),
Cfg: cfg,
snapCount: cfg.SnapCount,
errorc: make(chan error, 1),
store: st,
r: raftNode{
isIDRemoved: func(id uint64) bool { return cl.IsIDRemoved(types.ID(id)) },
Node: n,
ticker: time.Tick(heartbeat),
// set up contention detectors for raft heartbeat message.
// expect to send a heartbeat within 2 heartbeat intervals.
td: contention.NewTimeoutDetector(2 * heartbeat),
heartbeat: heartbeat,
raftStorage: s,
storage: NewStorage(w, ss),
msgSnapC: make(chan raftpb.Message, maxInFlightMsgSnap),
readStateC: make(chan raft.ReadState, 1),
},
id: id,
attributes: membership.Attributes{Name: cfg.Name, ClientURLs: cfg.ClientURLs.StringSlice()},
cluster: cl,
stats: sstats,
lstats: lstats,
SyncTicker: time.Tick(500 * time.Millisecond),
peerRt: prt,
reqIDGen: idutil.NewGenerator(uint16(id), time.Now()),
forceVersionC: make(chan struct{}),
}
srv.applyV2 = &applierV2store{store: srv.store, cluster: srv.cluster}
srv.be = be
minTTL := time.Duration((3*cfg.ElectionTicks)/2) * heartbeat
// always recover lessor before kv. When we recover the mvcc.KV it will reattach keys to its leases.
// If we recover mvcc.KV first, it will attach the keys to the wrong lessor before it recovers.
srv.lessor = lease.NewLessor(srv.be, int64(math.Ceil(minTTL.Seconds())))
srv.kv = mvcc.New(srv.be, srv.lessor, &srv.consistIndex)
if beExist {
kvindex := srv.kv.ConsistentIndex()
// TODO: remove kvindex != 0 checking when we do not expect users to upgrade
// etcd from pre-3.0 release.
if snapshot != nil && kvindex < snapshot.Metadata.Index {
if kvindex != 0 {
return nil, fmt.Errorf("database file (%v index %d) does not match with snapshot (index %d).", bepath, kvindex, snapshot.Metadata.Index)
}
plog.Warningf("consistent index never saved (snapshot index=%d)", snapshot.Metadata.Index)
}
}
srv.consistIndex.setConsistentIndex(srv.kv.ConsistentIndex())
srv.authStore = auth.NewAuthStore(srv.be)
if h := cfg.AutoCompactionRetention; h != 0 {
srv.compactor = compactor.NewPeriodic(h, srv.kv, srv)
srv.compactor.Run()
}
srv.applyV3Base = &applierV3backend{srv}
if err = srv.restoreAlarms(); err != nil {
return nil, err
}
// TODO: move transport initialization near the definition of remote
tr := &rafthttp.Transport{
TLSInfo: cfg.PeerTLSInfo,
DialTimeout: cfg.peerDialTimeout(),
ID: id,
URLs: cfg.PeerURLs,
ClusterID: cl.ID(),
Raft: srv,
Snapshotter: ss,
ServerStats: sstats,
LeaderStats: lstats,
ErrorC: srv.errorc,
}
if err = tr.Start(); err != nil {
return nil, err
}
// add all remotes into transport
for _, m := range remotes {
if m.ID != id {
tr.AddRemote(m.ID, m.PeerURLs)
}
}
for _, m := range cl.Members() {
if m.ID != id {
tr.AddPeer(m.ID, m.PeerURLs)
}
}
srv.r.transport = tr
return srv, nil
}
func main() {
from := flag.String("data-dir", "", "")
flag.Parse()
if *from == "" {
log.Fatal("Must provide -data-dir flag")
}
ss := snap.New(snapDir(*from))
snapshot, err := ss.Load()
var walsnap walpb.Snapshot
switch err {
case nil:
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
nodes := genIDSlice(snapshot.Metadata.ConfState.Nodes)
fmt.Printf("Snapshot:\nterm=%d index=%d nodes=%s\n",
walsnap.Term, walsnap.Index, nodes)
case snap.ErrNoSnapshot:
fmt.Printf("Snapshot:\nempty\n")
default:
log.Fatalf("Failed loading snapshot: %v", err)
}
w, err := wal.Open(walDir(*from), walsnap)
if err != nil {
log.Fatalf("Failed opening WAL: %v", err)
}
wmetadata, state, ents, err := w.ReadAll()
w.Close()
if err != nil {
log.Fatalf("Failed reading WAL: %v", err)
}
id, cid := parseWALMetadata(wmetadata)
vid := types.ID(state.Vote)
fmt.Printf("WAL metadata:\nnodeID=%s clusterID=%s term=%d commitIndex=%d vote=%s\n",
id, cid, state.Term, state.Commit, vid)
fmt.Printf("WAL entries:\n")
fmt.Printf("lastIndex=%d\n", ents[len(ents)-1].Index)
fmt.Printf("%4s\t%10s\ttype\tdata\n", "term", "index")
for _, e := range ents {
msg := fmt.Sprintf("%4d\t%10d", e.Term, e.Index)
switch e.Type {
case raftpb.EntryNormal:
msg = fmt.Sprintf("%s\tnorm", msg)
var r etcdserverpb.Request
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s\t???", msg)
break
}
switch r.Method {
case "":
msg = fmt.Sprintf("%s\tnoop", msg)
case "SYNC":
msg = fmt.Sprintf("%s\tmethod=SYNC time=%q", msg, time.Unix(0, r.Time))
case "QGET", "DELETE":
msg = fmt.Sprintf("%s\tmethod=%s path=%s", msg, r.Method, excerpt(r.Path, 64, 64))
default:
msg = fmt.Sprintf("%s\tmethod=%s path=%s val=%s", msg, r.Method, excerpt(r.Path, 64, 64), excerpt(r.Val, 128, 0))
}
case raftpb.EntryConfChange:
msg = fmt.Sprintf("%s\tconf", msg)
var r raftpb.ConfChange
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s\t???", msg)
} else {
msg = fmt.Sprintf("%s\tmethod=%s id=%s", msg, r.Type, types.ID(r.NodeID))
}
}
fmt.Println(msg)
}
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (srv *EtcdServer, err error) {
st := store.New(StoreClusterPrefix, StoreKeysPrefix)
var (
w *wal.WAL
n raft.Node
s *raft.MemoryStorage
id types.ID
cl *membership.RaftCluster
)
if terr := fileutil.TouchDirAll(cfg.DataDir); terr != nil {
return nil, fmt.Errorf("cannot access data directory: %v", terr)
}
// Run the migrations.
dataVer, err := version.DetectDataDir(cfg.DataDir)
if err != nil {
return nil, err
}
if err = upgradeDataDir(cfg.DataDir, cfg.Name, dataVer); err != nil {
return nil, err
}
haveWAL := wal.Exist(cfg.WALDir())
if err = os.MkdirAll(cfg.SnapDir(), privateDirMode); err != nil && !os.IsExist(err) {
plog.Fatalf("create snapshot directory error: %v", err)
}
ss := snap.New(cfg.SnapDir())
be := backend.NewDefaultBackend(path.Join(cfg.SnapDir(), databaseFilename))
defer func() {
if err != nil {
be.Close()
}
}()
prt, err := rafthttp.NewRoundTripper(cfg.PeerTLSInfo, cfg.peerDialTimeout())
if err != nil {
return nil, err
}
var remotes []*membership.Member
switch {
case !haveWAL && !cfg.NewCluster:
if err = cfg.VerifyJoinExisting(); err != nil {
return nil, err
}
cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
existingCluster, gerr := GetClusterFromRemotePeers(getRemotePeerURLs(cl, cfg.Name), prt)
if gerr != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", gerr)
}
if err = membership.ValidateClusterAndAssignIDs(cl, existingCluster); err != nil {
return nil, fmt.Errorf("error validating peerURLs %s: %v", existingCluster, err)
}
if !isCompatibleWithCluster(cl, cl.MemberByName(cfg.Name).ID, prt) {
return nil, fmt.Errorf("incomptible with current running cluster")
}
remotes = existingCluster.Members()
cl.SetID(existingCluster.ID())
cl.SetStore(st)
cl.SetBackend(be)
cfg.Print()
id, n, s, w = startNode(cfg, cl, nil)
case !haveWAL && cfg.NewCluster:
if err = cfg.VerifyBootstrap(); err != nil {
return nil, err
}
cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
m := cl.MemberByName(cfg.Name)
if isMemberBootstrapped(cl, cfg.Name, prt, cfg.bootstrapTimeout()) {
return nil, fmt.Errorf("member %s has already been bootstrapped", m.ID)
}
if cfg.ShouldDiscover() {
var str string
str, err = discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.InitialPeerURLsMap.String())
if err != nil {
return nil, &DiscoveryError{Op: "join", Err: err}
}
var urlsmap types.URLsMap
urlsmap, err = types.NewURLsMap(str)
if err != nil {
return nil, err
}
if checkDuplicateURL(urlsmap) {
return nil, fmt.Errorf("discovery cluster %s has duplicate url", urlsmap)
}
if cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, urlsmap); err != nil {
return nil, err
}
}
cl.SetStore(st)
cl.SetBackend(be)
cfg.PrintWithInitial()
id, n, s, w = startNode(cfg, cl, cl.MemberIDs())
case haveWAL:
if err = fileutil.IsDirWriteable(cfg.MemberDir()); err != nil {
return nil, fmt.Errorf("cannot write to member directory: %v", err)
}
if err = fileutil.IsDirWriteable(cfg.WALDir()); err != nil {
return nil, fmt.Errorf("cannot write to WAL directory: %v", err)
}
if cfg.ShouldDiscover() {
plog.Warningf("discovery token ignored since a cluster has already been initialized. Valid log found at %q", cfg.WALDir())
}
var snapshot *raftpb.Snapshot
snapshot, err = ss.Load()
if err != nil && err != snap.ErrNoSnapshot {
return nil, err
}
if snapshot != nil {
if err = st.Recovery(snapshot.Data); err != nil {
plog.Panicf("recovered store from snapshot error: %v", err)
}
plog.Infof("recovered store from snapshot at index %d", snapshot.Metadata.Index)
}
cfg.Print()
if !cfg.ForceNewCluster {
id, cl, n, s, w = restartNode(cfg, snapshot)
} else {
id, cl, n, s, w = restartAsStandaloneNode(cfg, snapshot)
}
cl.SetStore(st)
cl.SetBackend(be)
cl.Recover()
default:
return nil, fmt.Errorf("unsupported bootstrap config")
}
if terr := fileutil.TouchDirAll(cfg.MemberDir()); terr != nil {
return nil, fmt.Errorf("cannot access member directory: %v", terr)
}
sstats := &stats.ServerStats{
Name: cfg.Name,
ID: id.String(),
}
sstats.Initialize()
lstats := stats.NewLeaderStats(id.String())
srv = &EtcdServer{
readych: make(chan struct{}),
Cfg: cfg,
snapCount: cfg.SnapCount,
errorc: make(chan error, 1),
store: st,
r: raftNode{
Node: n,
ticker: time.Tick(time.Duration(cfg.TickMs) * time.Millisecond),
raftStorage: s,
storage: NewStorage(w, ss),
},
id: id,
attributes: membership.Attributes{Name: cfg.Name, ClientURLs: cfg.ClientURLs.StringSlice()},
cluster: cl,
stats: sstats,
lstats: lstats,
SyncTicker: time.Tick(500 * time.Millisecond),
peerRt: prt,
reqIDGen: idutil.NewGenerator(uint16(id), time.Now()),
forceVersionC: make(chan struct{}),
msgSnapC: make(chan raftpb.Message, maxInFlightMsgSnap),
}
srv.applyV2 = &applierV2store{store: srv.store, cluster: srv.cluster}
srv.be = be
srv.lessor = lease.NewLessor(srv.be)
srv.kv = mvcc.New(srv.be, srv.lessor, &srv.consistIndex)
srv.consistIndex.setConsistentIndex(srv.kv.ConsistentIndex())
srv.authStore = auth.NewAuthStore(srv.be)
if h := cfg.AutoCompactionRetention; h != 0 {
srv.compactor = compactor.NewPeriodic(h, srv.kv, srv)
srv.compactor.Run()
}
if err = srv.restoreAlarms(); err != nil {
return nil, err
}
// TODO: move transport initialization near the definition of remote
tr := &rafthttp.Transport{
TLSInfo: cfg.PeerTLSInfo,
DialTimeout: cfg.peerDialTimeout(),
ID: id,
URLs: cfg.PeerURLs,
ClusterID: cl.ID(),
Raft: srv,
Snapshotter: ss,
ServerStats: sstats,
LeaderStats: lstats,
ErrorC: srv.errorc,
}
if err = tr.Start(); err != nil {
return nil, err
}
// add all remotes into transport
for _, m := range remotes {
if m.ID != id {
tr.AddRemote(m.ID, m.PeerURLs)
}
}
for _, m := range cl.Members() {
if m.ID != id {
tr.AddPeer(m.ID, m.PeerURLs)
}
}
srv.r.transport = tr
return srv, nil
}
func (o originalSnap) New(dirpath string) Snapshotter {
return snap.New(dirpath)
}
