// TestApplyRepeat tests that server handles repeat raft messages gracefully
func TestApplyRepeat(t *testing.T) {
	n := newNodeConfChangeCommitterStream()
	n.readyc <- raft.Ready{
		SoftState: &raft.SoftState{RaftState: raft.StateLeader},
	}
	cl := newTestCluster(nil)
	st := store.New()
	cl.SetStore(store.New())
	cl.AddMember(&membership.Member{ID: 1234})
	s := &EtcdServer{
		r: raftNode{
			Node:        n,
			raftStorage: raft.NewMemoryStorage(),
			storage:     mockstorage.NewStorageRecorder(""),
			transport:   rafthttp.NewNopTransporter(),
		},
		cfg:      &ServerConfig{},
		store:    st,
		cluster:  cl,
		reqIDGen: idutil.NewGenerator(0, time.Time{}),
	}
	s.applyV2 = &applierV2store{s}
	s.start()
	req := &pb.Request{Method: "QGET", ID: uint64(1)}
	ents := []raftpb.Entry{{Index: 1, Data: pbutil.MustMarshal(req)}}
	n.readyc <- raft.Ready{CommittedEntries: ents}
	// dup msg
	n.readyc <- raft.Ready{CommittedEntries: ents}

	// use a conf change to block until dup msgs are all processed
	cc := &raftpb.ConfChange{Type: raftpb.ConfChangeRemoveNode, NodeID: 2}
	ents = []raftpb.Entry{{
		Index: 2,
		Type:  raftpb.EntryConfChange,
		Data:  pbutil.MustMarshal(cc),
	}}
	n.readyc <- raft.Ready{CommittedEntries: ents}
	// wait for conf change message
	act, err := n.Wait(1)
	// wait for stop message (async to avoid deadlock)
	stopc := make(chan error)
	go func() {
		_, werr := n.Wait(1)
		stopc <- werr
	}()
	s.Stop()

	// only want to confirm etcdserver won't panic; no data to check

	if err != nil {
		t.Fatal(err)
	}
	if len(act) == 0 {
		t.Fatalf("expected len(act)=0, got %d", len(act))
	}

	if err = <-stopc; err != nil {
		t.Fatalf("error on stop (%v)", err)
	}
}

func (enc *messageEncoder) encode(m raftpb.Message) error {
	if err := binary.Write(enc.w, binary.BigEndian, uint64(m.Size())); err != nil {
		return err
	}
	_, err := enc.w.Write(pbutil.MustMarshal(&m))
	return err
}

func (s *Snapshotter) save(snapshot *raftpb.Snapshot) error {
	start := time.Now()

	fname := fmt.Sprintf("%016x-%016x%s", snapshot.Metadata.Term, snapshot.Metadata.Index, snapSuffix)
	b := pbutil.MustMarshal(snapshot)
	crc := crc32.Update(0, crcTable, b)
	snap := snappb.Snapshot{Crc: crc, Data: b}
	d, err := snap.Marshal()
	if err != nil {
		return err
	} else {
		marshallingDurations.Observe(float64(time.Since(start)) / float64(time.Second))
	}

	err = pioutil.WriteAndSyncFile(path.Join(s.dir, fname), d, 0666)
	if err == nil {
		saveDurations.Observe(float64(time.Since(start)) / float64(time.Second))
	} else {
		err1 := os.Remove(path.Join(s.dir, fname))
		if err1 != nil {
			plog.Errorf("failed to remove broken snapshot file %s", path.Join(s.dir, fname))
		}
	}
	return err
}

func (s *sender) handle() {
	defer s.wg.Done()
	for m := range s.q {
		start := time.Now()
		err := s.post(pbutil.MustMarshal(m))
		end := time.Now()

		s.mu.Lock()
		if err != nil {
			if s.errored == nil || s.errored.Error() != err.Error() {
				log.Printf("sender: error posting to %s: %v", s.id, err)
				s.errored = err
			}
			if s.active {
				log.Printf("sender: the connection with %s becomes inactive", s.id)
				s.active = false
			}
			if m.Type == raftpb.MsgApp {
				s.fs.Fail()
			}
		} else {
			if !s.active {
				log.Printf("sender: the connection with %s becomes active", s.id)
				s.active = true
				s.errored = nil
			}
			if m.Type == raftpb.MsgApp {
				s.fs.Succ(end.Sub(start))
			}
		}
		s.mu.Unlock()
	}
}

func (p *pipeline) handle() {
	defer p.wg.Done()
	for m := range p.msgc {
		start := time.Now()
		err := p.post(pbutil.MustMarshal(&m))
		if err == errStopped {
			return
		}
		end := time.Now()

		if err != nil {
			reportSentFailure(pipelineMsg, m)
			p.status.deactivate(failureType{source: pipelineMsg, action: "write"}, err.Error())
			if m.Type == raftpb.MsgApp && p.fs != nil {
				p.fs.Fail()
			}
			p.r.ReportUnreachable(m.To)
			if isMsgSnap(m) {
				p.r.ReportSnapshot(m.To, raft.SnapshotFailure)
			}
		} else {
			p.status.activate()
			if m.Type == raftpb.MsgApp && p.fs != nil {
				p.fs.Succ(end.Sub(start))
			}
			if isMsgSnap(m) {
				p.r.ReportSnapshot(m.To, raft.SnapshotFinish)
			}
			reportSentDuration(pipelineMsg, m, time.Since(start))
		}
	}
}

func (p *peer) handle() {
	defer p.wg.Done()
	for m := range p.q {
		start := time.Now()
		err := p.post(pbutil.MustMarshal(m))
		end := time.Now()

		p.Lock()
		if err != nil {
			if p.errored == nil || p.errored.Error() != err.Error() {
				log.Printf("sender: error posting to %s: %v", p.id, err)
				p.errored = err
			}
			if p.active {
				log.Printf("sender: the connection with %s becomes inactive", p.id)
				p.active = false
			}
			if m.Type == raftpb.MsgApp {
				p.fs.Fail()
			}
		} else {
			if !p.active {
				log.Printf("sender: the connection with %s becomes active", p.id)
				p.active = true
				p.errored = nil
			}
			if m.Type == raftpb.MsgApp {
				p.fs.Succ(end.Sub(start))
			}
		}
		p.Unlock()
	}
}

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
}

// TestApplyMultiConfChangeShouldStop ensures that apply will return shouldStop
// if the local member is removed along with other conf updates.
func TestApplyMultiConfChangeShouldStop(t *testing.T) {
	cl := membership.NewCluster("")
	cl.SetStore(store.New())
	for i := 1; i <= 5; i++ {
		cl.AddMember(&membership.Member{ID: types.ID(i)})
	}
	srv := &EtcdServer{
		id: 2,
		r: raftNode{
			Node:      newNodeNop(),
			transport: rafthttp.NewNopTransporter(),
		},
		cluster: cl,
		w:       wait.New(),
	}
	ents := []raftpb.Entry{}
	for i := 1; i <= 4; i++ {
		ent := raftpb.Entry{
			Term:  1,
			Index: uint64(i),
			Type:  raftpb.EntryConfChange,
			Data: pbutil.MustMarshal(
				&raftpb.ConfChange{
					Type:   raftpb.ConfChangeRemoveNode,
					NodeID: uint64(i)}),
		}
		ents = append(ents, ent)
	}

	_, shouldStop := srv.apply(ents, &raftpb.ConfState{})
	if !shouldStop {
		t.Errorf("shouldStop = %t, want %t", shouldStop, true)
	}
}

// TODO (xiangli): reasonable retry logic
func (s *sender) Send(m raftpb.Message) error {
	s.maybeStopStream(m.Term)
	if shouldInitStream(m) && !s.hasStreamClient() {
		s.initStream(types.ID(m.From), types.ID(m.To), m.Term)
		s.batcher.Reset(time.Now())
	}
	if canBatch(m) && s.hasStreamClient() {
		if s.batcher.ShouldBatch(time.Now()) {
			return nil
		}
	}
	if canUseStream(m) {
		if ok := s.tryStream(m); ok {
			return nil
		}
	}
	// TODO: don't block. we should be able to have 1000s
	// of messages out at a time.
	data := pbutil.MustMarshal(&m)
	select {
	case s.q <- data:
		return nil
	default:
		log.Printf("sender: reach the maximal serving to %s", s.u)
		return fmt.Errorf("reach maximal serving")
	}
}

// createConfigChangeEnts creates a series of Raft entries (i.e.
// EntryConfChange) to remove the set of given IDs from the cluster. The ID
// `self` is _not_ removed, even if present in the set.
// If `self` is not inside the given ids, it creates a Raft entry to add a
// default member with the given `self`.
func createConfigChangeEnts(ids []uint64, self uint64, term, index uint64) []raftpb.Entry {
	ents := make([]raftpb.Entry, 0)
	next := index + 1
	found := false
	for _, id := range ids {
		if id == self {
			found = true
			continue
		}
		cc := &raftpb.ConfChange{
			Type:   raftpb.ConfChangeRemoveNode,
			NodeID: id,
		}
		e := raftpb.Entry{
			Type:  raftpb.EntryConfChange,
			Data:  pbutil.MustMarshal(cc),
			Term:  term,
			Index: next,
		}
		ents = append(ents, e)
		next++
	}
	if !found {
		m := Member{
			ID:             types.ID(self),
			RaftAttributes: RaftAttributes{PeerURLs: []string{"http://localhost:7001", "http://localhost:2380"}},
		}
		ctx, err := json.Marshal(m)
		if err != nil {
			plog.Panicf("marshal member should never fail: %v", err)
		}
		cc := &raftpb.ConfChange{
			Type:    raftpb.ConfChangeAddNode,
			NodeID:  self,
			Context: ctx,
		}
		e := raftpb.Entry{
			Type:  raftpb.EntryConfChange,
			Data:  pbutil.MustMarshal(cc),
			Term:  term,
			Index: next,
		}
		ents = append(ents, e)
	}
	return ents
}

func (w *WAL) SaveState(s *raftpb.HardState) error {
	if raft.IsEmptyHardState(*s) {
		return nil
	}
	b := pbutil.MustMarshal(s)
	rec := &walpb.Record{Type: stateType, Data: b}
	return w.encoder.encode(rec)
}

func (w *WAL) SaveEntry(e *raftpb.Entry) error {
	b := pbutil.MustMarshal(e)
	rec := &walpb.Record{Type: entryType, Data: b}
	if err := w.encoder.encode(rec); err != nil {
		return err
	}
	w.enti = e.Index
	return nil
}

func (w *WAL) saveEntry(e *raftpb.Entry) error {
	// TODO: add MustMarshalTo to reduce one allocation.
	b := pbutil.MustMarshal(e)
	rec := &walpb.Record{Type: entryType, Data: b}
	if err := w.encoder.encode(rec); err != nil {
		return err
	}
	w.enti = e.Index
	return nil
}

// handleBackup handles a request that intends to do a backup.
func handleBackup(c *cli.Context) {
	srcSnap := path.Join(c.String("data-dir"), "member", "snap")
	destSnap := path.Join(c.String("backup-dir"), "member", "snap")
	srcWAL := path.Join(c.String("data-dir"), "member", "wal")
	destWAL := path.Join(c.String("backup-dir"), "member", "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 walsnap walpb.Snapshot
	if snapshot != nil {
		walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
		newss := snap.New(destSnap)
		if err = newss.SaveSnap(*snapshot); err != nil {
			log.Fatal(err)
		}
	}

	w, err := wal.OpenForRead(srcWAL, walsnap)
	if err != nil {
		log.Fatal(err)
	}
	defer w.Close()
	wmetadata, state, ents, err := w.ReadAll()
	switch err {
	case nil:
	case wal.ErrSnapshotNotFound:
		fmt.Printf("Failed to find the match snapshot record %+v in wal %v.", walsnap, srcWAL)
		fmt.Printf("etcdctl will add it back. Start auto fixing...")
	default:
		log.Fatal(err)
	}
	var metadata etcdserverpb.Metadata
	pbutil.MustUnmarshal(&metadata, wmetadata)
	idgen := idutil.NewGenerator(0, time.Now())
	metadata.NodeID = idgen.Next()
	metadata.ClusterID = idgen.Next()

	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)
	}
	if err := neww.SaveSnapshot(walsnap); err != nil {
		log.Fatal(err)
	}
}

func (s *Snapshotter) save(snapshot *raftpb.Snapshot) error {
	fname := fmt.Sprintf("%016x-%016x%s", snapshot.Metadata.Term, snapshot.Metadata.Index, snapSuffix)
	b := pbutil.MustMarshal(snapshot)
	crc := crc32.Update(0, crcTable, b)
	snap := snappb.Snapshot{Crc: crc, Data: b}
	d, err := snap.Marshal()
	if err != nil {
		return err
	}
	return ioutil.WriteFile(path.Join(s.dir, fname), d, 0666)
}

func TestGetIDs(t *testing.T) {
	addcc := &raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 2}
	addEntry := raftpb.Entry{Type: raftpb.EntryConfChange, Data: pbutil.MustMarshal(addcc)}
	removecc := &raftpb.ConfChange{Type: raftpb.ConfChangeRemoveNode, NodeID: 2}
	removeEntry := raftpb.Entry{Type: raftpb.EntryConfChange, Data: pbutil.MustMarshal(removecc)}
	normalEntry := raftpb.Entry{Type: raftpb.EntryNormal}
	updatecc := &raftpb.ConfChange{Type: raftpb.ConfChangeUpdateNode, NodeID: 2}
	updateEntry := raftpb.Entry{Type: raftpb.EntryConfChange, Data: pbutil.MustMarshal(updatecc)}

	tests := []struct {
		confState *raftpb.ConfState
		ents      []raftpb.Entry

		widSet []uint64
	}{
		{nil, []raftpb.Entry{}, []uint64{}},
		{&raftpb.ConfState{Nodes: []uint64{1}},
			[]raftpb.Entry{}, []uint64{1}},
		{&raftpb.ConfState{Nodes: []uint64{1}},
			[]raftpb.Entry{addEntry}, []uint64{1, 2}},
		{&raftpb.ConfState{Nodes: []uint64{1}},
			[]raftpb.Entry{addEntry, removeEntry}, []uint64{1}},
		{&raftpb.ConfState{Nodes: []uint64{1}},
			[]raftpb.Entry{addEntry, normalEntry}, []uint64{1, 2}},
		{&raftpb.ConfState{Nodes: []uint64{1}},
			[]raftpb.Entry{addEntry, normalEntry, updateEntry}, []uint64{1, 2}},
		{&raftpb.ConfState{Nodes: []uint64{1}},
			[]raftpb.Entry{addEntry, removeEntry, normalEntry}, []uint64{1}},
	}

	for i, tt := range tests {
		var snap raftpb.Snapshot
		if tt.confState != nil {
			snap.Metadata.ConfState = *tt.confState
		}
		idSet := getIDs(&snap, tt.ents)
		if !reflect.DeepEqual(idSet, tt.widSet) {
			t.Errorf("#%d: idset = %#v, want %#v", i, idSet, tt.widSet)
		}
	}
}

func TestNew(t *testing.T) {
	p, err := ioutil.TempDir(os.TempDir(), "waltest")
	if err != nil {
		t.Fatal(err)
	}
	defer os.RemoveAll(p)

	w, err := Create(p, []byte("somedata"))
	if err != nil {
		t.Fatalf("err = %v, want nil", err)
	}
	if g := path.Base(w.tail().Name()); g != walName(0, 0) {
		t.Errorf("name = %+v, want %+v", g, walName(0, 0))
	}
	defer w.Close()

	// file is preallocated to segment size; only read data written by wal
	off, err := w.tail().Seek(0, os.SEEK_CUR)
	if err != nil {
		t.Fatal(err)
	}
	gd := make([]byte, off)
	f, err := os.Open(w.tail().Name())
	if err != nil {
		t.Fatal(err)
	}
	defer f.Close()
	if _, err = io.ReadFull(f, gd); err != nil {
		t.Fatalf("err = %v, want nil", err)
	}

	var wb bytes.Buffer
	e := newEncoder(&wb, 0)
	err = e.encode(&walpb.Record{Type: crcType, Crc: 0})
	if err != nil {
		t.Fatalf("err = %v, want nil", err)
	}
	err = e.encode(&walpb.Record{Type: metadataType, Data: []byte("somedata")})
	if err != nil {
		t.Fatalf("err = %v, want nil", err)
	}
	r := &walpb.Record{
		Type: snapshotType,
		Data: pbutil.MustMarshal(&walpb.Snapshot{}),
	}
	if err = e.encode(r); err != nil {
		t.Fatalf("err = %v, want nil", err)
	}
	e.flush()
	if !reflect.DeepEqual(gd, wb.Bytes()) {
		t.Errorf("data = %v, want %v", gd, wb.Bytes())
	}
}

func (w *WAL) SaveSnapshot(e walpb.Snapshot) error {
	b := pbutil.MustMarshal(&e)
	rec := &walpb.Record{Type: snapshotType, Data: b}
	if err := w.encoder.encode(rec); err != nil {
		return err
	}
	// update enti only when snapshot is ahead of last index
	if w.enti < e.Index {
		w.enti = e.Index
	}
	return w.sync()
}

// sync proposes a SYNC request and is non-blocking.
// This makes no guarantee that the request will be proposed or performed.
// The request will be canceled after the given timeout.
func (s *EtcdServer) sync(timeout time.Duration) {
	ctx, cancel := context.WithTimeout(context.Background(), timeout)
	req := pb.Request{
		Method: "SYNC",
		ID:     s.reqIDGen.Next(),
		Time:   time.Now().UnixNano(),
	}
	data := pbutil.MustMarshal(&req)
	// There is no promise that node has leader when do SYNC request,
	// so it uses goroutine to propose.
	go func() {
		s.r.Propose(ctx, data)
		cancel()
	}()
}

func (w *WAL) SaveSnapshot(e walpb.Snapshot) error {
	w.mu.Lock()
	defer w.mu.Unlock()

	b := pbutil.MustMarshal(&e)
	rec := &walpb.Record{Type: snapshotType, Data: b}
	if err := w.encoder.encode(rec); err != nil {
		return err
	}
	// update enti only when snapshot is ahead of last index
	if w.enti < e.Index {
		w.enti = e.Index
	}
	lastIndexSaved.Set(float64(w.enti))
	return w.sync()
}