// InitAndStartNode gets called after having at least one membership sync with the cluster.
func (n *node) InitAndStartNode(wal *raftwal.Wal) {
restart, err := n.initFromWal(wal)
x.Check(err)
if restart {
fmt.Printf("RESTARTING\n")
n.raft = raft.RestartNode(n.cfg)
} else {
if groups().HasPeer(n.gid) {
n.joinPeers()
n.raft = raft.StartNode(n.cfg, nil)
} else {
peers := []raft.Peer{{ID: n.id}}
n.raft = raft.StartNode(n.cfg, peers)
// Trigger election, so this node can become the leader of this single-node cluster.
n.canCampaign = true
}
}
go n.processCommitCh()
go n.Run()
// TODO: Find a better way to snapshot, so we don't lose the membership
// state information, which isn't persisted.
// go n.snapshotPeriodically()
go n.batchAndSendMessages()
}
// Start implements replication.LogReplicator
func (l *raftLog) Start(lo uint64) error {
inited, err := l.config.Storage.(*raftStorage).IsInitialized()
if err != nil {
return err
}
if inited {
l.config.Applied = lo
l.node = raft.RestartNode(&l.config)
} else {
if lo != 0 {
log.Panicf("storage uninitialized but state machine not fresh: lo = %d", lo)
}
// Add a dummy first entry
hardState, confState, err := l.config.Storage.InitialState()
if err != nil {
return err
}
confNodes := make([]raft.Peer, 0, len(confState.Nodes))
for _, id := range confState.Nodes {
confNodes = append(confNodes, raft.Peer{ID: id})
}
l.config.Storage.(*raftStorage).save(hardState, make([]raftpb.Entry, 1))
l.node = raft.StartNode(&l.config, confNodes)
}
l.leaderHintSet = make(chan bool, COMMITTED_BUFFER)
l.waitCommitted = make(chan replication.LogEntry, COMMITTED_BUFFER)
l.stop = make(chan struct{})
l.stopped = make(chan struct{})
l.grpcDropClient = make(chan uint64)
l.grpcClientCache = make(map[uint64]proto.RaftClient)
go l.run()
return nil
}
func NewRaftNode(name string, c *raft.Config, storage *raft.MemoryStorage, lookup NodeLookup, tr Transporter) *raftNode {
var peers []raft.Peer
node := raft.StartNode(c, peers)
var mu sync.Mutex
removed := make(map[uint64]bool)
h := Jesteress([]byte(name))
return &raftNode{
index: 0,
term: 0,
lead: 0,
hash: h,
lt: time.Now(),
Node: node,
cfg: *c,
mu: mu,
lookup: lookup,
removed: removed,
idgen: idutil.NewGenerator(uint8(h), time.Now()),
w: wait.New(),
ticker: time.Tick(500 * time.Millisecond),
raftStorage: storage,
storage: &NoopStorage{},
transport: tr,
applyc: make(chan apply),
stopped: make(chan struct{}),
done: make(chan struct{}),
}
}
func StartNode(id uint64, peers []raft.Peer, addrs []string, t Transport) *Node {
st := raft.NewMemoryStorage()
c := &raft.Config{
ID: id,
ElectionTick: 10,
HeartbeatTick: 1,
Storage: st,
MaxSizePerMsg: 1024 * 1024,
MaxInflightMsgs: 256,
}
rn := raft.StartNode(c, peers)
n := &Node{
Node: rn,
storage: st,
}
n.t = t
path := fmt.Sprint(DirPath, id)
f, e := os.OpenFile(path, os.O_CREATE|os.O_RDWR, os.ModePerm)
if e != nil {
fmt.Fprintln(os.Stdout, "open the kv file error:", e)
}
n.f = f
go n.t.listen()
go n.start()
return n
}
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
}
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 newCtrl(
self net.Addr,
others []net.Addr, // to join existing cluster, pass nil or empty others
minPeerCount int,
incomingc <-chan raftpb.Message,
outgoingc chan<- raftpb.Message,
unreachablec <-chan uint64,
confchangec <-chan raftpb.ConfChange,
snapshotc chan<- raftpb.Snapshot,
entryc chan<- raftpb.Entry,
proposalc <-chan []byte,
removedc chan<- struct{},
logger mesh.Logger,
) *ctrl {
storage := raft.NewMemoryStorage()
raftLogger := &raft.DefaultLogger{Logger: log.New(ioutil.Discard, "", 0)}
raftLogger.EnableDebug()
nodeConfig := &raft.Config{
ID: makeRaftPeer(self).ID,
ElectionTick: 10,
HeartbeatTick: 1,
Storage: storage,
Applied: 0, // starting fresh
MaxSizePerMsg: 4096, // TODO(pb): looks like bytes; confirm that
MaxInflightMsgs: 256, // TODO(pb): copied from docs; confirm that
CheckQuorum: true, // leader steps down if quorum is not active for an electionTimeout
Logger: raftLogger,
}
startPeers := makeRaftPeers(others)
if len(startPeers) == 0 {
startPeers = nil // special case: join existing
}
node := raft.StartNode(nodeConfig, startPeers)
c := &ctrl{
self: makeRaftPeer(self),
minPeerCount: minPeerCount,
incomingc: incomingc,
outgoingc: outgoingc,
unreachablec: unreachablec,
confchangec: confchangec,
snapshotc: snapshotc,
entryc: entryc,
proposalc: proposalc,
stopc: make(chan struct{}),
removedc: removedc,
terminatedc: make(chan struct{}),
storage: storage,
node: node,
logger: logger,
}
go c.driveRaft() // analagous to raftexample serveChannels
return c
}
func startNode(id uint64, peers []raft.Peer, iface iface) *node {
st := raft.NewMemoryStorage()
rn := raft.StartNode(id, peers, 10, 1, st)
n := &node{
Node: rn,
id: id,
storage: st,
iface: iface,
}
n.start()
return n
}
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
}
func (rc *raftNode) startRaft() {
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 {
rc.node = raft.StartNode(c, rpeers)
}
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 newNode(id uint64, peers []raft.Peer) *node {
store := raft.NewMemoryStorage()
n := &node{
id: id,
store: store,
cfg: &raft.Config{
ID: uint64(id),
ElectionTick: 3,
HeartbeatTick: 1,
Storage: store,
MaxSizePerMsg: 4096,
MaxInflightMsgs: 256,
},
data: make(map[string]string),
ctx: context.TODO(),
}
n.raft = raft.StartNode(n.cfg, peers)
return n
}
// NewNode generates a new Raft node based on an unique
// ID, an address and optionally: a handler and receive
// only channel to send event when an entry is committed
// to the logs
func NewNode(id uint64, addr string, cfg *raft.Config, apply ApplyCommand) (*Node, error) {
if cfg == nil {
cfg = DefaultNodeConfig()
}
store := raft.NewMemoryStorage()
peers := []raft.Peer{{ID: id}}
n := &Node{
ID: id,
Ctx: context.TODO(),
Cluster: NewCluster(),
Store: store,
Address: addr,
Cfg: &raft.Config{
ID: id,
ElectionTick: cfg.ElectionTick,
HeartbeatTick: cfg.HeartbeatTick,
Storage: store,
MaxSizePerMsg: cfg.MaxSizePerMsg,
MaxInflightMsgs: cfg.MaxInflightMsgs,
Logger: cfg.Logger,
},
PStore: make(map[string]string),
ticker: time.NewTicker(time.Second),
stopChan: make(chan struct{}),
pauseChan: make(chan bool),
apply: apply,
}
n.Cluster.AddPeer(
&Peer{
NodeInfo: &NodeInfo{
ID: id,
Addr: addr,
},
},
)
n.Node = raft.StartNode(n.Cfg, peers)
return n, nil
}
func startNode(id uint64, peers []raft.Peer, iface iface) *node {
st := raft.NewMemoryStorage()
c := &raft.Config{
ID: id,
ElectionTick: 10,
HeartbeatTick: 1,
Storage: st,
MaxSizePerMsg: 1024 * 1024,
MaxInflightMsgs: 256,
}
rn := raft.StartNode(c, peers)
n := &node{
Node: rn,
id: id,
storage: st,
iface: iface,
pausec: make(chan bool),
}
n.start()
return n
}
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
}
func newNode(id uint64, addr string, peers []raft.Peer) *node {
store := raft.NewMemoryStorage()
n := &node{
id: id,
addr: addr,
ctx: context.TODO(),
store: store,
cfg: &raft.Config{
ID: id,
ElectionTick: 5 * hb,
HeartbeatTick: hb,
Storage: store,
MaxSizePerMsg: math.MaxUint16,
MaxInflightMsgs: 256,
},
pstore: make(map[string]string),
ticker: time.Tick(time.Second),
done: make(chan struct{}),
}
n.raft = raft.StartNode(n.cfg, peers)
return n
}
// JoinAndStart joins and starts the raft server
func (n *Node) JoinAndStart(ctx context.Context) (err error) {
ctx, cancel := n.WithContext(ctx)
defer func() {
cancel()
if err != nil {
n.done()
}
}()
loadAndStartErr := n.loadAndStart(ctx, n.opts.ForceNewCluster)
if loadAndStartErr != nil && loadAndStartErr != errNoWAL {
return loadAndStartErr
}
snapshot, err := n.raftStore.Snapshot()
// Snapshot never returns an error
if err != nil {
panic("could not get snapshot of raft store")
}
n.confState = snapshot.Metadata.ConfState
n.appliedIndex = snapshot.Metadata.Index
n.snapshotIndex = snapshot.Metadata.Index
if loadAndStartErr == errNoWAL {
if n.opts.JoinAddr != "" {
c, err := n.ConnectToMember(n.opts.JoinAddr, 10*time.Second)
if err != nil {
return err
}
client := api.NewRaftMembershipClient(c.Conn)
defer func() {
_ = c.Conn.Close()
}()
joinCtx, joinCancel := context.WithTimeout(ctx, 10*time.Second)
defer joinCancel()
resp, err := client.Join(joinCtx, &api.JoinRequest{
Addr: n.opts.Addr,
})
if err != nil {
return err
}
n.Config.ID = resp.RaftID
if _, err := n.createWAL(n.opts.ID); err != nil {
return err
}
n.raftNode = raft.StartNode(n.Config, []raft.Peer{})
if err := n.registerNodes(resp.Members); err != nil {
if walErr := n.wal.Close(); err != nil {
log.G(ctx).WithError(walErr).Error("raft: error closing WAL")
}
return err
}
} else {
// First member in the cluster, self-assign ID
n.Config.ID = uint64(rand.Int63()) + 1
peer, err := n.createWAL(n.opts.ID)
if err != nil {
return err
}
n.raftNode = raft.StartNode(n.Config, []raft.Peer{peer})
n.campaignWhenAble = true
}
atomic.StoreUint32(&n.isMember, 1)
return nil
}
if n.opts.JoinAddr != "" {
log.G(ctx).Warning("ignoring request to join cluster, because raft state already exists")
}
n.campaignWhenAble = true
n.raftNode = raft.RestartNode(n.Config)
atomic.StoreUint32(&n.isMember, 1)
return nil
}
// JoinAndStart joins and starts the raft server
func (n *Node) JoinAndStart() error {
loadAndStartErr := n.loadAndStart(n.Ctx, n.opts.ForceNewCluster)
if loadAndStartErr != nil && loadAndStartErr != errNoWAL {
n.ticker.Stop()
return loadAndStartErr
}
snapshot, err := n.raftStore.Snapshot()
// Snapshot never returns an error
if err != nil {
panic("could not get snapshot of raft store")
}
n.confState = snapshot.Metadata.ConfState
n.appliedIndex = snapshot.Metadata.Index
n.snapshotIndex = snapshot.Metadata.Index
if loadAndStartErr == errNoWAL {
if n.joinAddr != "" {
c, err := n.ConnectToMember(n.joinAddr, 10*time.Second)
if err != nil {
return err
}
client := api.NewRaftMembershipClient(c.Conn)
defer func() {
_ = c.Conn.Close()
}()
ctx, cancel := context.WithTimeout(n.Ctx, 10*time.Second)
defer cancel()
resp, err := client.Join(ctx, &api.JoinRequest{
Addr: n.Address,
})
if err != nil {
return err
}
n.Config.ID = resp.RaftID
if _, err := n.createWAL(n.opts.ID); err != nil {
return err
}
n.Node = raft.StartNode(n.Config, []raft.Peer{})
if err := n.registerNodes(resp.Members); err != nil {
return err
}
} else {
// First member in the cluster, self-assign ID
n.Config.ID = uint64(rand.Int63()) + 1
peer, err := n.createWAL(n.opts.ID)
if err != nil {
return err
}
n.Node = raft.StartNode(n.Config, []raft.Peer{peer})
if err := n.Campaign(n.Ctx); err != nil {
return err
}
}
atomic.StoreUint32(&n.isMember, 1)
return nil
}
if n.joinAddr != "" {
n.Config.Logger.Warning("ignoring request to join cluster, because raft state already exists")
}
n.Node = raft.RestartNode(n.Config)
atomic.StoreUint32(&n.isMember, 1)
return nil
}
// NewNode generates a new Raft node
func NewNode(ctx context.Context, opts NewNodeOptions) (*Node, error) {
cfg := opts.Config
if cfg == nil {
cfg = DefaultNodeConfig()
}
if opts.TickInterval == 0 {
opts.TickInterval = time.Second
}
raftID, err := identity.ParseNodeID(opts.ID)
if err != nil {
return nil, err
}
raftStore := raft.NewMemoryStorage()
ctx, cancel := context.WithCancel(ctx)
n := &Node{
Ctx: ctx,
cancel: cancel,
cluster: membership.NewCluster(),
tlsCredentials: opts.TLSCredentials,
raftStore: raftStore,
Address: opts.Addr,
Config: &raft.Config{
ElectionTick: cfg.ElectionTick,
HeartbeatTick: cfg.HeartbeatTick,
Storage: raftStore,
MaxSizePerMsg: cfg.MaxSizePerMsg,
MaxInflightMsgs: cfg.MaxInflightMsgs,
Logger: cfg.Logger,
ID: raftID,
},
forceNewCluster: opts.ForceNewCluster,
stopCh: make(chan struct{}),
doneCh: make(chan struct{}),
StateDir: opts.StateDir,
joinAddr: opts.JoinAddr,
sendTimeout: 2 * time.Second,
leadershipBroadcast: events.NewBroadcaster(),
}
n.memoryStore = store.NewMemoryStore(n)
if opts.ClockSource == nil {
n.ticker = clock.NewClock().NewTicker(opts.TickInterval)
} else {
n.ticker = opts.ClockSource.NewTicker(opts.TickInterval)
}
if opts.SendTimeout != 0 {
n.sendTimeout = opts.SendTimeout
}
if err := n.loadAndStart(ctx, opts.ForceNewCluster); err != nil {
n.ticker.Stop()
return nil, err
}
snapshot, err := raftStore.Snapshot()
// Snapshot never returns an error
if err != nil {
panic("could not get snapshot of raft store")
}
n.confState = snapshot.Metadata.ConfState
n.appliedIndex = snapshot.Metadata.Index
n.snapshotIndex = snapshot.Metadata.Index
n.reqIDGen = idutil.NewGenerator(uint16(n.Config.ID), time.Now())
n.wait = newWait()
if n.startNodePeers != nil {
if n.joinAddr != "" {
c, err := n.ConnectToMember(n.joinAddr, 10*time.Second)
if err != nil {
return nil, err
}
defer func() {
_ = c.Conn.Close()
}()
ctx, cancel := context.WithTimeout(n.Ctx, 10*time.Second)
defer cancel()
resp, err := c.Join(ctx, &api.JoinRequest{
Addr: n.Address,
})
if err != nil {
return nil, err
}
n.Node = raft.StartNode(n.Config, []raft.Peer{})
if err := n.registerNodes(resp.Members); err != nil {
return nil, err
}
} else {
n.Node = raft.StartNode(n.Config, n.startNodePeers)
if err := n.Campaign(n.Ctx); err != nil {
return nil, err
}
}
return n, nil
}
if n.joinAddr != "" {
n.Config.Logger.Warning("ignoring request to join cluster, because raft state already exists")
}
n.Node = raft.RestartNode(n.Config)
return n, nil
}
func startNode(id int, chans []chan raftpb.Message) {
l := log.WithField("id", id)
storage := raft.NewMemoryStorage()
c := &raft.Config{
ID: uint64(id),
ElectionTick: 10,
HeartbeatTick: 1,
Storage: storage,
MaxSizePerMsg: 4096,
MaxInflightMsgs: 256,
}
var peers []raft.Peer
for i := 1; i <= 3; i++ {
if id == i {
continue
}
peer := raft.Peer{ID: uint64(i)}
peers = append(peers, peer)
}
l.WithField("peers", peers).Debug("Peers")
n := raft.StartNode(c, peers)
tick := time.Tick(3 * time.Second)
for count := 0; ; count++ {
l.Debug("Waiting for something to happen")
select {
case <-tick:
l.Debug("Got a tick")
n.Tick()
{
r := 1 + rand.Intn(3)
l.WithField("r", r).Debugf("Got rand value")
if id == r {
// I should send some data.
data := fmt.Sprintf("This is me: %v at count %v", id, count)
l.WithField("data", data).Debug("Proposing some data")
ctx, _ := context.WithTimeout(context.Background(), 5*time.Second)
go n.Propose(ctx, []byte(data))
}
}
case rd := <-n.Ready():
l.Debug("Got ready")
// Save initial state.
if err := save(rd, storage); err != nil {
log.WithField("error", err).Error("While saving")
return
}
// Send all messages to other nodes.
for i := 1; i <= 3; i++ {
if id == i {
continue
}
for _, msg := range rd.Messages {
chans[i-1] <- msg
}
}
// Apply Snapshot (already done) and CommittedEntries
storage.Append(rd.CommittedEntries) // No config change in test.
n.Advance()
case msg := <-chans[id-1]:
l.WithField("msg", msg.String()).Debug("GOT MESSAGE")
ctx, _ := context.WithTimeout(context.Background(), 5*time.Second)
n.Step(ctx, 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) *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
}
