func newMemberCollection(ms []*etcdserver.Member) httptypes.MemberCollection {
c := httptypes.MemberCollection(make([]httptypes.Member, len(ms)))
for i, m := range ms {
tm := httptypes.Member{
ID: strutil.IDAsHex(m.ID),
Name: m.Name,
PeerURLs: make([]string, len(m.PeerURLs)),
ClientURLs: make([]string, len(m.ClientURLs)),
}
copy(tm.PeerURLs, m.PeerURLs)
copy(tm.ClientURLs, m.ClientURLs)
c[i] = tm
}
return c
}
// send uses the given client to send a message to a member in the given
// ClusterStore, retrying up to 3 times for each message. The given
// ServerStats and LeaderStats are updated appropriately
func send(c *http.Client, cl *Cluster, m raftpb.Message, ss *stats.ServerStats, ls *stats.LeaderStats) {
cid := cl.ID()
// TODO (xiangli): reasonable retry logic
for i := 0; i < 3; i++ {
memb := cl.Member(m.To)
if memb == nil {
// TODO: unknown peer id.. what do we do? I
// don't think his should ever happen, need to
// look into this further.
log.Printf("etcdhttp: no member for %d", m.To)
return
}
u := fmt.Sprintf("%s%s", memb.PickPeerURL(), raftPrefix)
// TODO: don't block. we should be able to have 1000s
// of messages out at a time.
data, err := m.Marshal()
if err != nil {
log.Println("etcdhttp: dropping message:", err)
return // drop bad message
}
if m.Type == raftpb.MsgApp {
ss.SendAppendReq(len(data))
}
to := strutil.IDAsHex(m.To)
fs := ls.Follower(to)
start := time.Now()
sent := httpPost(c, u, cid, data)
end := time.Now()
if sent {
fs.Succ(end.Sub(start))
return
}
fs.Fail()
// TODO: backoff
}
}
func removedMemberStoreKey(id uint64) string {
return path.Join(storeRemovedMembersPrefix, strutil.IDAsHex(id))
}
func (s *EtcdServer) UpdateRecvApp(from uint64, length int64) {
s.stats.RecvAppendReq(strutil.IDAsHex(from), int(length))
}
// 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
}