// 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
}
func readWAL(waldir string, index uint64) (w *wal.WAL, id, cid types.ID, st raftpb.HardState, ents []raftpb.Entry) {
var err error
if w, err = wal.OpenAtIndex(waldir, index); err != nil {
log.Fatalf("etcdserver: open wal error: %v", err)
}
var wmetadata []byte
if wmetadata, st, ents, err = w.ReadAll(); err != nil {
log.Fatalf("etcdserver: read wal error: %v", err)
}
var metadata pb.Metadata
pbutil.MustUnmarshal(&metadata, wmetadata)
id = types.ID(metadata.NodeID)
cid = types.ID(metadata.ClusterID)
return
}
func dump5(dataDir string) ([]raftpb.Entry, error) {
wd5 := walDir5(dataDir)
if !wal.Exist(wd5) {
return nil, fmt.Errorf("No wal exists at %s", wd5)
}
w, err := wal.OpenAtIndex(wd5, 0)
if err != nil {
return nil, err
}
defer w.Close()
_, _, ents, err := w.ReadAll()
return ents, 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 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.OpenAtIndex(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)
}
}
func restartNode(cfg *ServerConfig, index uint64, snapshot *raftpb.Snapshot) (id uint64, n raft.Node, w *wal.WAL) {
var err error
// restart a node from previous wal
if w, err = wal.OpenAtIndex(cfg.WALDir(), index); err != nil {
log.Fatal(err)
}
wmetadata, st, ents, err := w.ReadAll()
if err != nil {
log.Fatal(err)
}
var metadata pb.Metadata
pbutil.MustUnmarshal(&metadata, wmetadata)
id = metadata.NodeID
cfg.Cluster.SetID(metadata.ClusterID)
log.Printf("etcdserver: restart member %x in cluster %x at commit index %d", id, cfg.Cluster.ID(), st.Commit)
n = raft.RestartNode(id, 10, 1, snapshot, st, ents)
return
}
// 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
}