// 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.NewRecorderStream()
p := mockstorage.NewStorageRecorderStream("")
srv := &EtcdServer{
Cfg: &ServerConfig{},
r: raftNode{
Node: newNodeNop(),
raftStorage: s,
storage: p,
},
store: st,
}
srv.kv = mvcc.New(be, &lease.FakeLessor{}, &srv.consistIndex)
srv.be = be
ch := make(chan struct{}, 2)
go func() {
gaction, _ := p.Wait(1)
defer func() { ch <- struct{}{} }()
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])
}
}()
go func() {
gaction, _ := st.Wait(2)
defer func() { ch <- struct{}{} }()
if len(gaction) != 2 {
t.Fatalf("len(action) = %d, want 2", 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])
}
}()
srv.snapshot(1, raftpb.ConfState{Nodes: []uint64{1}})
<-ch
<-ch
}
// 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.applyV2 = &applierV2store{srv}
srv.kv = mvcc.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
}
func newEtcdStore(
proposalc chan<- []byte,
snapshotc <-chan raftpb.Snapshot,
entryc <-chan raftpb.Entry,
confentryc chan<- raftpb.Entry,
logger mesh.Logger,
) *etcdStore {
// It would be much better if we could have a proper in-memory backend. Alas:
// backend.Backend is tightly coupled to bolt.DB, and both are tightly coupled
// to os.Open &c. So we'd need to fork both Bolt and backend. A task for
// another day.
f, err := ioutil.TempFile(os.TempDir(), "mesh_etcd_backend_")
if err != nil {
panic(err)
}
dbPath := f.Name()
f.Close()
logger.Printf("etcd store: using %s", dbPath)
b := backend.NewDefaultBackend(dbPath)
lessor := lease.NewLessor(b)
index := &consistentIndex{0}
kv := mvcc.New(b, lessor, index)
s := &etcdStore{
proposalc: proposalc,
snapshotc: snapshotc,
entryc: entryc,
confentryc: confentryc,
actionc: make(chan func()),
quitc: make(chan struct{}),
terminatedc: make(chan struct{}),
logger: logger,
dbPath: dbPath,
kv: kv,
lessor: lessor,
index: index,
idgen: makeIDGen(),
pending: map[uint64]responseChans{},
}
go s.loop()
return s
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (srv *EtcdServer, err error) {
st := store.New(StoreClusterPrefix, StoreKeysPrefix)
var (
w *wal.WAL
n raft.Node
s *raft.MemoryStorage
id types.ID
cl *membership.RaftCluster
)
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())
if err = os.MkdirAll(cfg.SnapDir(), privateDirMode); err != nil && !os.IsExist(err) {
plog.Fatalf("create snapshot directory error: %v", err)
}
ss := snap.New(cfg.SnapDir())
be := backend.NewDefaultBackend(path.Join(cfg.SnapDir(), databaseFilename))
defer func() {
if err != nil {
be.Close()
}
}()
prt, err := rafthttp.NewRoundTripper(cfg.PeerTLSInfo, cfg.peerDialTimeout())
if err != nil {
return nil, err
}
var remotes []*membership.Member
switch {
case !haveWAL && !cfg.NewCluster:
if err = cfg.VerifyJoinExisting(); err != nil {
return nil, err
}
cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
existingCluster, gerr := GetClusterFromRemotePeers(getRemotePeerURLs(cl, cfg.Name), prt)
if gerr != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", gerr)
}
if err = membership.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)
cl.SetBackend(be)
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 = membership.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
str, err = discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.InitialPeerURLsMap.String())
if err != nil {
return nil, &DiscoveryError{Op: "join", Err: err}
}
var urlsmap types.URLsMap
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 = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, urlsmap); err != nil {
return nil, err
}
}
cl.SetStore(st)
cl.SetBackend(be)
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
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.SetBackend(be)
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{
readych: make(chan struct{}),
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: membership.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),
}
srv.applyV2 = &applierV2store{store: srv.store, cluster: srv.cluster}
srv.be = be
srv.lessor = lease.NewLessor(srv.be)
srv.kv = mvcc.New(srv.be, srv.lessor, &srv.consistIndex)
srv.consistIndex.setConsistentIndex(srv.kv.ConsistentIndex())
srv.authStore = auth.NewAuthStore(srv.be)
if h := cfg.AutoCompactionRetention; h != 0 {
srv.compactor = compactor.NewPeriodic(h, srv.kv, srv)
srv.compactor.Run()
}
if err = srv.restoreAlarms(); err != nil {
return nil, err
}
// 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,
}
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
}
// NewServer creates a new EtcdServer from the supplied configuration. The
// configuration is considered static for the lifetime of the EtcdServer.
func NewServer(cfg *ServerConfig) (srv *EtcdServer, err error) {
st := store.New(StoreClusterPrefix, StoreKeysPrefix)
var (
w *wal.WAL
n raft.Node
s *raft.MemoryStorage
id types.ID
cl *membership.RaftCluster
)
if terr := fileutil.TouchDirAll(cfg.DataDir); terr != nil {
return nil, fmt.Errorf("cannot access data directory: %v", terr)
}
haveWAL := wal.Exist(cfg.WALDir())
if err = fileutil.TouchDirAll(cfg.SnapDir()); err != nil {
plog.Fatalf("create snapshot directory error: %v", err)
}
ss := snap.New(cfg.SnapDir())
bepath := path.Join(cfg.SnapDir(), databaseFilename)
beExist := fileutil.Exist(bepath)
var be backend.Backend
beOpened := make(chan struct{})
go func() {
be = backend.NewDefaultBackend(bepath)
beOpened <- struct{}{}
}()
select {
case <-beOpened:
case <-time.After(time.Second):
plog.Warningf("another etcd process is running with the same data dir and holding the file lock.")
plog.Warningf("waiting for it to exit before starting...")
<-beOpened
}
defer func() {
if err != nil {
be.Close()
}
}()
prt, err := rafthttp.NewRoundTripper(cfg.PeerTLSInfo, cfg.peerDialTimeout())
if err != nil {
return nil, err
}
var (
remotes []*membership.Member
snapshot *raftpb.Snapshot
)
switch {
case !haveWAL && !cfg.NewCluster:
if err = cfg.VerifyJoinExisting(); err != nil {
return nil, err
}
cl, err = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, cfg.InitialPeerURLsMap)
if err != nil {
return nil, err
}
existingCluster, gerr := GetClusterFromRemotePeers(getRemotePeerURLs(cl, cfg.Name), prt)
if gerr != nil {
return nil, fmt.Errorf("cannot fetch cluster info from peer urls: %v", gerr)
}
if err = membership.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("incompatible with current running cluster")
}
remotes = existingCluster.Members()
cl.SetID(existingCluster.ID())
cl.SetStore(st)
cl.SetBackend(be)
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 = membership.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
str, err = discovery.JoinCluster(cfg.DiscoveryURL, cfg.DiscoveryProxy, m.ID, cfg.InitialPeerURLsMap.String())
if err != nil {
return nil, &DiscoveryError{Op: "join", Err: err}
}
var urlsmap types.URLsMap
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 = membership.NewClusterFromURLsMap(cfg.InitialClusterToken, urlsmap); err != nil {
return nil, err
}
}
cl.SetStore(st)
cl.SetBackend(be)
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 !cfg.ForceNewCluster {
id, cl, n, s, w = restartNode(cfg, snapshot)
} else {
id, cl, n, s, w = restartAsStandaloneNode(cfg, snapshot)
}
cl.SetStore(st)
cl.SetBackend(be)
cl.Recover(api.UpdateCapability)
if cl.Version() != nil && !cl.Version().LessThan(semver.Version{Major: 3}) && !beExist {
os.RemoveAll(bepath)
return nil, fmt.Errorf("database file (%v) of the backend is missing", bepath)
}
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())
heartbeat := time.Duration(cfg.TickMs) * time.Millisecond
srv = &EtcdServer{
readych: make(chan struct{}),
Cfg: cfg,
snapCount: cfg.SnapCount,
errorc: make(chan error, 1),
store: st,
r: raftNode{
isIDRemoved: func(id uint64) bool { return cl.IsIDRemoved(types.ID(id)) },
Node: n,
ticker: time.Tick(heartbeat),
// set up contention detectors for raft heartbeat message.
// expect to send a heartbeat within 2 heartbeat intervals.
td: contention.NewTimeoutDetector(2 * heartbeat),
heartbeat: heartbeat,
raftStorage: s,
storage: NewStorage(w, ss),
msgSnapC: make(chan raftpb.Message, maxInFlightMsgSnap),
readStateC: make(chan raft.ReadState, 1),
},
id: id,
attributes: membership.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{}),
}
srv.applyV2 = &applierV2store{store: srv.store, cluster: srv.cluster}
srv.be = be
minTTL := time.Duration((3*cfg.ElectionTicks)/2) * heartbeat
// always recover lessor before kv. When we recover the mvcc.KV it will reattach keys to its leases.
// If we recover mvcc.KV first, it will attach the keys to the wrong lessor before it recovers.
srv.lessor = lease.NewLessor(srv.be, int64(math.Ceil(minTTL.Seconds())))
srv.kv = mvcc.New(srv.be, srv.lessor, &srv.consistIndex)
if beExist {
kvindex := srv.kv.ConsistentIndex()
// TODO: remove kvindex != 0 checking when we do not expect users to upgrade
// etcd from pre-3.0 release.
if snapshot != nil && kvindex < snapshot.Metadata.Index {
if kvindex != 0 {
return nil, fmt.Errorf("database file (%v index %d) does not match with snapshot (index %d).", bepath, kvindex, snapshot.Metadata.Index)
}
plog.Warningf("consistent index never saved (snapshot index=%d)", snapshot.Metadata.Index)
}
}
srv.consistIndex.setConsistentIndex(srv.kv.ConsistentIndex())
srv.authStore = auth.NewAuthStore(srv.be)
if h := cfg.AutoCompactionRetention; h != 0 {
srv.compactor = compactor.NewPeriodic(h, srv.kv, srv)
srv.compactor.Run()
}
srv.applyV3Base = &applierV3backend{srv}
if err = srv.restoreAlarms(); err != nil {
return nil, err
}
// 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,
}
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),
}
s.applyV2 = &applierV2store{s}
be, tmpPath := backend.NewDefaultTmpBackend()
defer func() {
os.RemoveAll(tmpPath)
}()
s.kv = mvcc.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)
}
}
