Example #1
0
// startRaft starts a raft node from the given wal dir.
// If the wal dir does not exist, startRaft will start a new raft node.
// If the wal dir exists, startRaft will restart the previous raft node.
// startRaft returns the started raft node and the opened wal.
func startRaft(id int64, peerIDs []int64, waldir string) (raft.Node, *wal.WAL) {
	if !wal.Exist(waldir) {
		w, err := wal.Create(waldir)
		if err != nil {
			log.Fatal(err)
		}
		n := raft.Start(id, peerIDs, 10, 1)
		return n, w
	}

	// restart a node from previous wal
	// TODO(xiangli): check snapshot; not open from one
	w, err := wal.OpenAtIndex(waldir, 1)
	if err != nil {
		log.Fatal(err)
	}
	wid, st, ents, err := w.ReadAll()
	// 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)
	}
	if err != nil {
		log.Fatal(err)
	}
	n := raft.Restart(id, peerIDs, 10, 1, st, ents)
	return n, w
}
Example #2
0
func startNode(cfg *ServerConfig, ids []types.ID) (id types.ID, n raft.Node, s *raft.MemoryStorage, w *wal.WAL) {
	var err error
	member := cfg.Cluster.MemberByName(cfg.Name)
	metadata := pbutil.MustMarshal(
		&pb.Metadata{
			NodeID:    uint64(member.ID),
			ClusterID: uint64(cfg.Cluster.ID()),
		},
	)
	if err := os.MkdirAll(cfg.SnapDir(), privateDirMode); err != nil {
		log.Fatalf("etcdserver create snapshot directory error: %v", err)
	}
	if w, err = wal.Create(cfg.WALDir(), metadata); err != nil {
		log.Fatalf("etcdserver: create wal error: %v", err)
	}
	peers := make([]raft.Peer, len(ids))
	for i, id := range ids {
		ctx, err := json.Marshal((*cfg.Cluster).Member(id))
		if err != nil {
			log.Panicf("marshal member should never fail: %v", err)
		}
		peers[i] = raft.Peer{ID: uint64(id), Context: ctx}
	}
	id = member.ID
	log.Printf("etcdserver: start member %s in cluster %s", id, cfg.Cluster.ID())
	s = raft.NewMemoryStorage()
	n = raft.StartNode(uint64(id), peers, 10, 1, s)
	return
}
Example #3
0
// 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)
	}
}
Example #4
0
// Create returns a new WAL object with the given encrypters and decrypters.
func (wc walCryptor) Create(dirpath string, metadata []byte) (WAL, error) {
	w, err := wal.Create(dirpath, metadata)
	if err != nil {
		return nil, err
	}
	return &wrappedWAL{
		WAL:       w,
		encrypter: wc.encrypter,
		decrypter: wc.decrypter,
	}, nil
}
Example #5
0
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
}
Example #6
0
// makeWAL creates a WAL for the initial cluster
func makeWAL(waldir string, cl *membership.RaftCluster) {
	if err := os.MkdirAll(waldir, 0755); err != nil {
		ExitWithError(ExitIO, err)
	}

	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))
	for i, p := range peers {
		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
	}

	w.Save(raftpb.HardState{
		Term:   1,
		Vote:   peers[0].ID,
		Commit: uint64(len(ents))}, ents)
}
Example #7
0
// 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)
	}
}
Example #8
0
// 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)
	}
}
Example #9
0
File: raft.go Project: oywc410/MYPG
func startNode(cfg *ServerConfig, cl *cluster, ids []types.ID) (id types.ID, n raft.Node, s *raft.MemoryStorage, w *wal.WAL) {
	var err error
	member := cl.MemberByName(cfg.Name)
	metadata := pbutil.MustMarshal(
		&pb.Metadata{
			NodeID:    uint64(member.ID),
			ClusterID: uint64(cl.ID()),
		},
	)

	//创建记录
	if err = os.MkdirAll(cfg.SnapDir(), privateDirMode); err != nil {
		plog.Fatalf("create snapshot directory error: %v", err)
	}
	if w, err = wal.Create(cfg.WALDir(), metadata); err != nil {
		plog.Fatalf("create wal error: %v", err)
	}
	//获取节点信息
	peers := make([]raft.Peer, len(ids))
	for i, id := range ids {
		ctx, err := json.Marshal((*cl).Member(id))
		if err != nil {
			plog.Panicf("marshal member should never fail: %v", err)
		}
		peers[i] = raft.Peer{ID: uint64(id), Context: ctx}
	}
	id = member.ID
	plog.Infof("starting member %s in cluster %s", id, cl.ID())
	s = raft.NewMemoryStorage()
	c := &raft.Config{
		ID:              uint64(id),
		ElectionTick:    cfg.ElectionTicks,
		HeartbeatTick:   1,
		Storage:         s, //存储
		MaxSizePerMsg:   maxSizePerMsg,
		MaxInflightMsgs: maxInflightMsgs,
		CheckQuorum:     true,
	}

	n = raft.StartNode(c, peers)
	raftStatusMu.Lock()
	raftStatus = n.Status
	raftStatusMu.Unlock()
	advanceTicksForElection(n, c.ElectionTick)
	return
}
Example #10
0
func (n *Node) createWAL(nodeID string) (raft.Peer, error) {
	raftNode := &api.RaftMember{
		RaftID: n.Config.ID,
		NodeID: nodeID,
		Addr:   n.Address,
	}
	metadata, err := raftNode.Marshal()
	if err != nil {
		return raft.Peer{}, fmt.Errorf("error marshalling raft node: %v", err)
	}
	n.wal, err = wal.Create(n.walDir(), metadata)
	if err != nil {
		return raft.Peer{}, fmt.Errorf("create WAL error: %v", err)
	}

	n.cluster.AddMember(&membership.Member{RaftMember: raftNode})
	return raft.Peer{ID: n.Config.ID, Context: metadata}, nil
}
Example #11
0
func (n *Node) createWAL(nodeID string) (raft.Peer, error) {
	raftNode := &api.RaftMember{
		RaftID: n.Config.ID,
		NodeID: nodeID,
		Addr:   n.opts.Addr,
	}
	metadata, err := raftNode.Marshal()
	if err != nil {
		return raft.Peer{}, errors.Wrap(err, "error marshalling raft node")
	}
	n.wal, err = wal.Create(n.walDir(), metadata)
	if err != nil {
		return raft.Peer{}, errors.Wrap(err, "failed to create WAL")
	}

	n.cluster.AddMember(&membership.Member{RaftMember: raftNode})
	return raft.Peer{ID: n.Config.ID, Context: metadata}, nil
}
Example #12
0
File: raft.go Project: ikatson/etcd
// openWAL returns a WAL ready for reading.
func (rc *raftNode) openWAL() *wal.WAL {
	if wal.Exist(rc.waldir) == false {
		if err := os.Mkdir(rc.waldir, 0750); err != nil {
			log.Fatalf("raftexample: cannot create dir for wal (%v)", err)
		}

		w, err := wal.Create(rc.waldir, nil)
		if err != nil {
			log.Fatalf("raftexample: create wal error (%v)", err)
		}
		w.Close()
	}

	w, err := wal.Open(rc.waldir, walpb.Snapshot{})
	if err != nil {
		log.Fatalf("raftexample: error loading wal (%v)", err)
	}

	return w
}
Example #13
0
func startNode(cfg *ServerConfig, ids []uint64) (id uint64, n raft.Node, w *wal.WAL) {
	var err error
	// TODO: remove the discoveryURL when it becomes part of the source for
	// generating nodeID.
	member := cfg.Cluster.MemberByName(cfg.Name)
	metadata := pbutil.MustMarshal(&pb.Metadata{NodeID: member.ID, ClusterID: cfg.Cluster.ID()})
	if w, err = wal.Create(cfg.WALDir(), metadata); err != nil {
		log.Fatal(err)
	}
	peers := make([]raft.Peer, len(ids))
	for i, id := range ids {
		ctx, err := json.Marshal((*cfg.Cluster).Member(id))
		if err != nil {
			log.Fatal(err)
		}
		peers[i] = raft.Peer{ID: id, Context: ctx}
	}
	id = member.ID
	log.Printf("etcdserver: start node %x in cluster %x", id, cfg.Cluster.ID())
	n = raft.StartNode(id, peers, 10, 1)
	return
}
Example #14
0
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")
}
Example #15
0
// 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)
	}
}
Example #16
0
// 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
}
Example #17
0
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
}
Example #18
0
func (o originalWAL) Create(dirpath string, metadata []byte) (WAL, error) {
	return wal.Create(dirpath, metadata)
}
Example #19
0
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
}