func (s *EtcdServer) applySnapshot(ep *etcdProgress, apply *apply) {
if raft.IsEmptySnap(apply.snapshot) {
return
}
if apply.snapshot.Metadata.Index <= ep.appliedi {
plog.Panicf("snapshot index [%d] should > appliedi[%d] + 1",
apply.snapshot.Metadata.Index, ep.appliedi)
}
if s.cfg.V3demo {
snapfn, err := s.r.storage.DBFilePath(apply.snapshot.Metadata.Index)
if err != nil {
plog.Panicf("get database snapshot file path error: %v", err)
}
fn := path.Join(s.cfg.SnapDir(), databaseFilename)
if err := os.Rename(snapfn, fn); err != nil {
plog.Panicf("rename snapshot file error: %v", err)
}
newbe := backend.NewDefaultBackend(fn)
newKV := dstorage.New(newbe, &s.consistIndex)
if err := newKV.Restore(); err != nil {
plog.Panicf("restore KV error: %v", err)
}
oldKV := s.swapKV(newKV)
// Closing oldKV might block until all the txns
// on the kv are finished.
// We do not want to wait on closing the old kv.
go func() {
if err := oldKV.Close(); err != nil {
plog.Panicf("close KV error: %v", err)
}
}()
}
if err := s.store.Recovery(apply.snapshot.Data); err != nil {
plog.Panicf("recovery store error: %v", err)
}
s.cluster.Recover()
// recover raft transport
s.r.transport.RemoveAllPeers()
for _, m := range s.cluster.Members() {
if m.ID == s.ID() {
continue
}
s.r.transport.AddPeer(m.ID, m.PeerURLs)
}
ep.appliedi = apply.snapshot.Metadata.Index
ep.snapi = ep.appliedi
ep.confState = apply.snapshot.Metadata.ConfState
plog.Infof("recovered from incoming snapshot at index %d", ep.snapi)
}
// Applied > SnapCount should trigger a SaveSnap event
func TestTriggerSnap(t *testing.T) {
be, tmpPath := backend.NewDefaultTmpBackend()
defer func() {
os.RemoveAll(tmpPath)
}()
snapc := 10
st := mockstore.NewRecorder()
p := mockstorage.NewStorageRecorderStream("")
srv := &EtcdServer{
cfg: &ServerConfig{TickMs: 1},
snapCount: uint64(snapc),
r: raftNode{
Node: newNodeCommitter(),
raftStorage: raft.NewMemoryStorage(),
storage: p,
transport: rafthttp.NewNopTransporter(),
},
store: st,
reqIDGen: idutil.NewGenerator(0, time.Time{}),
}
srv.kv = dstorage.New(be, &lease.FakeLessor{}, &srv.consistIndex)
srv.be = be
srv.start()
donec := make(chan struct{})
go func() {
wcnt := 2 + snapc
gaction, _ := p.Wait(wcnt)
// each operation is recorded as a Save
// (SnapCount+1) * Puts + SaveSnap = (SnapCount+1) * Save + SaveSnap
if len(gaction) != wcnt {
t.Fatalf("len(action) = %d, want %d", len(gaction), wcnt)
}
if !reflect.DeepEqual(gaction[wcnt-1], testutil.Action{Name: "SaveSnap"}) {
t.Errorf("action = %s, want SaveSnap", gaction[wcnt-1])
}
close(donec)
}()
for i := 0; i < snapc+1; i++ {
srv.Do(context.Background(), pb.Request{Method: "PUT"})
}
srv.Stop()
<-donec
}
// snapshot should snapshot the store and cut the persistent
func TestSnapshot(t *testing.T) {
be, tmpPath := backend.NewDefaultTmpBackend()
defer func() {
os.RemoveAll(tmpPath)
}()
s := raft.NewMemoryStorage()
s.Append([]raftpb.Entry{{Index: 1}})
st := mockstore.NewRecorder()
p := mockstorage.NewStorageRecorder("")
srv := &EtcdServer{
cfg: &ServerConfig{},
r: raftNode{
Node: newNodeNop(),
raftStorage: s,
storage: p,
},
store: st,
}
srv.kv = dstorage.New(be, &lease.FakeLessor{}, &srv.consistIndex)
srv.be = be
srv.snapshot(1, raftpb.ConfState{Nodes: []uint64{1}})
gaction, _ := st.Wait(2)
if len(gaction) != 2 {
t.Fatalf("len(action) = %d, want 1", len(gaction))
}
if !reflect.DeepEqual(gaction[0], testutil.Action{Name: "Clone"}) {
t.Errorf("action = %s, want Clone", gaction[0])
}
if !reflect.DeepEqual(gaction[1], testutil.Action{Name: "SaveNoCopy"}) {
t.Errorf("action = %s, want SaveNoCopy", gaction[1])
}
gaction = p.Action()
if len(gaction) != 1 {
t.Fatalf("len(action) = %d, want 1", len(gaction))
}
if !reflect.DeepEqual(gaction[0], testutil.Action{Name: "SaveSnap"}) {
t.Errorf("action = %s, want SaveSnap", gaction[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
n raft.Node
s *raft.MemoryStorage
id types.ID
cl *cluster
)
if terr := fileutil.TouchDirAll(cfg.DataDir); terr != nil {
return nil, fmt.Errorf("cannot access data directory: %v", terr)
}
if !cfg.V3demo && fileutil.Exist(path.Join(cfg.SnapDir(), databaseFilename)) {
return nil, errors.New("experimental-v3demo cannot be disabled once it is enabled")
}
// 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())
prt, err := rafthttp.NewRoundTripper(cfg.PeerTLSInfo, cfg.peerDialTimeout())
if err != nil {
return nil, err
}
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), prt)
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, prt) {
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, prt, cfg.bootstrapTimeout()) {
return nil, fmt.Errorf("member %s has already been bootstrapped", m.ID)
}
if cfg.ShouldDiscover() {
var str string
var err error
str, err = discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.InitialPeerURLsMap.String())
if err != nil {
return nil, &DiscoveryError{Op: "join", Err: 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.MemberDir()); err != nil {
return nil, fmt.Errorf("cannot write to member directory: %v", err)
}
if err := fileutil.IsDirWriteable(cfg.WALDir()); err != nil {
return nil, fmt.Errorf("cannot write to WAL 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())
}
var snapshot *raftpb.Snapshot
var err error
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 !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")
}
if terr := fileutil.TouchDirAll(cfg.MemberDir()); terr != nil {
return nil, fmt.Errorf("cannot access member directory: %v", terr)
}
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),
peerRt: prt,
reqIDGen: idutil.NewGenerator(uint16(id), time.Now()),
forceVersionC: make(chan struct{}),
msgSnapC: make(chan raftpb.Message, maxInFlightMsgSnap),
}
if cfg.V3demo {
srv.be = backend.NewDefaultBackend(path.Join(cfg.SnapDir(), databaseFilename))
srv.lessor = lease.NewLessor(srv.be)
srv.kv = dstorage.New(srv.be, srv.lessor, &srv.consistIndex)
srv.authStore = auth.NewAuthStore(srv)
if h := cfg.AutoCompactionRetention; h != 0 {
srv.compactor = compactor.NewPeriodic(h, srv.kv, srv)
srv.compactor.Run()
}
}
// TODO: move transport initialization near the definition of remote
tr := &rafthttp.Transport{
TLSInfo: cfg.PeerTLSInfo,
DialTimeout: cfg.peerDialTimeout(),
ID: id,
URLs: cfg.PeerURLs,
ClusterID: cl.ID(),
Raft: srv,
Snapshotter: ss,
ServerStats: sstats,
LeaderStats: lstats,
ErrorC: srv.errorc,
V3demo: cfg.V3demo,
}
if err := tr.Start(); err != nil {
return nil, err
}
// 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
}
// TestConcurrentApplyAndSnapshotV3 will send out snapshots concurrently with
// proposals.
func TestConcurrentApplyAndSnapshotV3(t *testing.T) {
const (
// snapshots that may queue up at once without dropping
maxInFlightMsgSnap = 16
)
n := newNopReadyNode()
st := store.New()
cl := membership.NewCluster("abc")
cl.SetStore(st)
testdir, err := ioutil.TempDir(os.TempDir(), "testsnapdir")
if err != nil {
t.Fatalf("Couldn't open tempdir (%v)", err)
}
defer os.RemoveAll(testdir)
if err := os.MkdirAll(testdir+"/member/snap", 0755); err != nil {
t.Fatalf("Couldn't make snap dir (%v)", err)
}
rs := raft.NewMemoryStorage()
tr, snapDoneC := rafthttp.NewSnapTransporter(testdir)
s := &EtcdServer{
cfg: &ServerConfig{
DataDir: testdir,
},
r: raftNode{
Node: n,
transport: tr,
storage: mockstorage.NewStorageRecorder(testdir),
raftStorage: rs,
},
store: st,
cluster: cl,
msgSnapC: make(chan raftpb.Message, maxInFlightMsgSnap),
}
be, tmpPath := backend.NewDefaultTmpBackend()
defer func() {
os.RemoveAll(tmpPath)
}()
s.kv = dstorage.New(be, &lease.FakeLessor{}, &s.consistIndex)
s.be = be
s.start()
defer s.Stop()
// submit applied entries and snap entries
idx := uint64(0)
outdated := 0
accepted := 0
for k := 1; k <= 101; k++ {
idx++
ch := s.w.Register(uint64(idx))
req := &pb.Request{Method: "QGET", ID: uint64(idx)}
ent := raftpb.Entry{Index: uint64(idx), Data: pbutil.MustMarshal(req)}
ready := raft.Ready{Entries: []raftpb.Entry{ent}}
n.readyc <- ready
ready = raft.Ready{CommittedEntries: []raftpb.Entry{ent}}
n.readyc <- ready
// "idx" applied
<-ch
// one snapshot for every two messages
if k%2 != 0 {
continue
}
n.readyc <- raft.Ready{Messages: []raftpb.Message{{Type: raftpb.MsgSnap}}}
// get the snapshot sent by the transport
snapMsg := <-snapDoneC
// If the snapshot trails applied records, recovery will panic
// since there's no allocated snapshot at the place of the
// snapshot record. This only happens when the applier and the
// snapshot sender get out of sync.
if snapMsg.Snapshot.Metadata.Index == idx {
idx++
snapMsg.Snapshot.Metadata.Index = idx
ready = raft.Ready{Snapshot: snapMsg.Snapshot}
n.readyc <- ready
accepted++
} else {
outdated++
}
// don't wait for the snapshot to complete, move to next message
}
if accepted != 50 {
t.Errorf("accepted=%v, want 50", accepted)
}
if outdated != 0 {
t.Errorf("outdated=%v, want 0", outdated)
}
}
func (s *EtcdServer) run() {
snap, err := s.r.raftStorage.Snapshot()
if err != nil {
plog.Panicf("get snapshot from raft storage error: %v", err)
}
confState := snap.Metadata.ConfState
snapi := snap.Metadata.Index
appliedi := snapi
s.r.start(s)
defer func() {
s.r.stop()
close(s.done)
}()
var shouldstop bool
for {
select {
case apply := <-s.r.apply():
// apply snapshot
if !raft.IsEmptySnap(apply.snapshot) {
if apply.snapshot.Metadata.Index <= appliedi {
plog.Panicf("snapshot index [%d] should > appliedi[%d] + 1",
apply.snapshot.Metadata.Index, appliedi)
}
if s.cfg.V3demo {
if err := s.kv.Close(); err != nil {
plog.Panicf("close KV error: %v", err)
}
snapfn, err := s.r.raftStorage.snapStore.getSnapFilePath(apply.snapshot.Metadata.Index)
if err != nil {
plog.Panicf("get snapshot file path error: %v", err)
}
fn := path.Join(s.cfg.StorageDir(), databaseFilename)
if err := os.Rename(snapfn, fn); err != nil {
plog.Panicf("rename snapshot file error: %v", err)
}
s.kv = dstorage.New(fn, &s.consistIndex)
if err := s.kv.Restore(); err != nil {
plog.Panicf("restore KV error: %v", err)
}
}
if err := s.store.Recovery(apply.snapshot.Data); err != nil {
plog.Panicf("recovery store error: %v", err)
}
s.cluster.Recover()
// recover raft transport
s.r.transport.RemoveAllPeers()
for _, m := range s.cluster.Members() {
if m.ID == s.ID() {
continue
}
s.r.transport.AddPeer(m.ID, m.PeerURLs)
}
appliedi = apply.snapshot.Metadata.Index
snapi = appliedi
confState = apply.snapshot.Metadata.ConfState
plog.Infof("recovered from incoming snapshot at index %d", snapi)
}
// apply entries
if len(apply.entries) != 0 {
firsti := apply.entries[0].Index
if firsti > appliedi+1 {
plog.Panicf("first index of committed entry[%d] should <= appliedi[%d] + 1", firsti, appliedi)
}
var ents []raftpb.Entry
if appliedi+1-firsti < uint64(len(apply.entries)) {
ents = apply.entries[appliedi+1-firsti:]
}
if appliedi, shouldstop = s.apply(ents, &confState); shouldstop {
go s.stopWithDelay(10*100*time.Millisecond, fmt.Errorf("the member has been permanently removed from the cluster"))
}
}
// wait for the raft routine to finish the disk writes before triggering a
// snapshot. or applied index might be greater than the last index in raft
// storage, since the raft routine might be slower than apply routine.
apply.done <- struct{}{}
// trigger snapshot
if appliedi-snapi > s.snapCount {
plog.Infof("start to snapshot (applied: %d, lastsnap: %d)", appliedi, snapi)
s.snapshot(appliedi, confState)
snapi = appliedi
}
case <-s.r.raftStorage.reqsnap():
s.r.raftStorage.raftsnap() <- s.createRaftSnapshot(appliedi, confState)
plog.Infof("requested snapshot created at %d", snapi)
case err := <-s.errorc:
plog.Errorf("%s", err)
plog.Infof("the data-dir used by this member must be removed.")
return
case <-s.stop:
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, 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
if terr := fileutil.TouchDirAll(cfg.DataDir); terr != nil {
return nil, fmt.Errorf("cannot access data directory: %v", terr)
}
// 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, &DiscoveryError{Op: "join", Err: 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.MemberDir()); err != nil {
return nil, fmt.Errorf("cannot write to member directory: %v", err)
}
if err := fileutil.IsDirWriteable(cfg.WALDir()); err != nil {
return nil, fmt.Errorf("cannot write to WAL 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")
}
if terr := fileutil.TouchDirAll(cfg.MemberDir()); terr != nil {
return nil, fmt.Errorf("cannot access member directory: %v", terr)
}
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{}),
}
if cfg.V3demo {
srv.kv = dstorage.New(path.Join(cfg.DataDir, "member", "v3demo"))
} else {
// we do not care about the error of the removal
os.RemoveAll(path.Join(cfg.DataDir, "member", "v3demo"))
}
// 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
}