Exemplo n.º 1
0
// newStreamClient starts and returns a new started stream client.
// The caller should call stop when finished, to shut it down.
func newStreamReader(id, to, cid types.ID, term uint64, tr http.RoundTripper, u string, r Raft) (*streamReader, error) {
	s := &streamReader{
		id:   id,
		to:   to,
		term: term,
		r:    r,
		done: make(chan struct{}),
	}

	uu, err := url.Parse(u)
	if err != nil {
		return nil, fmt.Errorf("parse url %s error: %v", u, err)
	}
	uu.Path = path.Join(RaftStreamPrefix, s.id.String())
	req, err := http.NewRequest("GET", uu.String(), nil)
	if err != nil {
		return nil, fmt.Errorf("new request to %s error: %v", u, err)
	}
	req.Header.Set("X-Etcd-Cluster-ID", cid.String())
	req.Header.Set("X-Raft-To", s.to.String())
	req.Header.Set("X-Raft-Term", strconv.FormatUint(s.term, 10))
	resp, err := tr.RoundTrip(req)
	if err != nil {
		return nil, fmt.Errorf("error posting to %q: %v", u, err)
	}
	if resp.StatusCode != http.StatusOK {
		resp.Body.Close()
		return nil, fmt.Errorf("unhandled http status %d", resp.StatusCode)
	}
	s.closer = resp.Body
	go s.handle(resp.Body)
	log.Printf("rafthttp: starting client stream to %s at term %d", s.to, s.term)
	return s, nil
}
Exemplo n.º 2
0
func isCompatibleWithVers(vers map[string]*version.Versions, local types.ID, minV, maxV *semver.Version) bool {
	var ok bool
	for id, v := range vers {
		// ignore comparasion with local version
		if id == local.String() {
			continue
		}
		if v == nil {
			continue
		}
		clusterv, err := semver.NewVersion(v.Cluster)
		if err != nil {
			plog.Errorf("cannot understand the cluster version of member %s (%v)", id, err)
			continue
		}
		if clusterv.LessThan(*minV) {
			plog.Warningf("the running cluster version(%v) is lower than the minimal cluster version(%v) supported", clusterv.String(), minV.String())
			return false
		}
		if maxV.LessThan(*clusterv) {
			plog.Warningf("the running cluster version(%v) is higher than the maximum cluster version(%v) supported", clusterv.String(), maxV.String())
			return false
		}
		ok = true
	}
	return ok
}
Exemplo n.º 3
0
// the caller of this function must have the peers mutex.
func (t *transport) removePeer(id types.ID) {
	if peer, ok := t.peers[id]; ok {
		peer.Stop()
	} else {
		log.Panicf("rafthttp: unexpected removal of unknown peer '%d'", id)
	}
	delete(t.peers, id)
	delete(t.leaderStats.Followers, id.String())
}
Exemplo n.º 4
0
func (t *transport) AddPeer(id types.ID, us []string) {
	t.mu.Lock()
	defer t.mu.Unlock()
	if _, ok := t.peers[id]; ok {
		return
	}
	urls, err := types.NewURLs(us)
	if err != nil {
		plog.Panicf("newURLs %+v should never fail: %+v", us, err)
	}
	fs := t.leaderStats.Follower(id.String())
	t.peers[id] = startPeer(t.roundTripper, urls, t.id, id, t.clusterID, t.raft, fs, t.errorc, t.term)
}
Exemplo n.º 5
0
func (t *transport) AddPeer(id types.ID, urls []string) {
	t.mu.Lock()
	defer t.mu.Unlock()
	if _, ok := t.peers[id]; ok {
		return
	}
	// TODO: considering how to switch between all available peer urls
	peerURL := urls[0]
	u, err := url.Parse(peerURL)
	if err != nil {
		log.Panicf("unexpect peer url %s", peerURL)
	}
	u.Path = path.Join(u.Path, RaftPrefix)
	fs := t.leaderStats.Follower(id.String())
	t.peers[id] = NewPeer(t.roundTripper, u.String(), id, t.clusterID, t.raft, fs, t.errorc)
}
Exemplo n.º 6
0
func (t *transport) UpdatePeer(id types.ID, us []string) {
	t.mu.Lock()
	defer t.mu.Unlock()
	// TODO: return error or just panic?
	if _, ok := t.peers[id]; !ok {
		return
	}
	urls, err := types.NewURLs(us)
	if err != nil {
		plog.Panicf("newURLs %+v should never fail: %+v", us, err)
	}
	t.peers[id].Update(urls)

	t.prober.Remove(id.String())
	addPeerToProber(t.prober, id.String(), us)
}
Exemplo n.º 7
0
func newEntryReader(r io.Reader, id types.ID) *entryReader {
	return &entryReader{
		r:         r,
		id:        id,
		ents:      metrics.GetMap("rafthttp.stream.entries_received").NewCounter(id.String()),
		bytes:     metrics.GetMap("rafthttp.stream.bytes_received").NewCounter(id.String()),
		lastIndex: metrics.GetMap("rafthttp.stream.last_index_received").NewGauge(id.String()),
	}
}
Exemplo n.º 8
0
func newEntryWriter(w io.Writer, id types.ID) *entryWriter {
	ew := &entryWriter{
		w:         w,
		id:        id,
		ents:      metrics.GetMap("rafthttp.stream.entries_sent").NewCounter(id.String()),
		bytes:     metrics.GetMap("rafthttp.stream.bytes_sent").NewCounter(id.String()),
		lastIndex: metrics.GetMap("rafthttp.stream.last_index_sent").NewGauge(id.String()),
	}
	return ew
}
Exemplo n.º 9
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, 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
}
Exemplo n.º 10
0
func removedMemberStoreKey(id types.ID) string {
	return path.Join(storeRemovedMembersPrefix, id.String())
}