// TestRawNodeProposeAndConfChange ensures that RawNode.Propose and RawNode.ProposeConfChange
// send the given proposal and ConfChange to the underlying raft.
func TestRawNodeProposeAndConfChange(t *testing.T) {
s := NewMemoryStorage()
var err error
rawNode, err := NewRawNode(newTestConfig(1, nil, 10, 1, s), []Peer{{ID: 1}})
if err != nil {
t.Fatal(err)
}
rd := rawNode.Ready()
s.Append(rd.Entries)
rawNode.Advance(rd)
rawNode.Campaign()
proposed := false
var (
lastIndex uint64
ccdata []byte
)
for {
rd = rawNode.Ready()
s.Append(rd.Entries)
// Once we are the leader, propose a command and a ConfChange.
if !proposed && rd.SoftState.Lead == rawNode.raft.ID {
rawNode.Propose([]byte("somedata"))
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata, err = cc.Marshal()
if err != nil {
t.Fatal(err)
}
rawNode.ProposeConfChange(cc)
proposed = true
}
rawNode.Advance(rd)
// Exit when we have four entries: one ConfChange, one no-op for the election,
// our proposed command and proposed ConfChange.
lastIndex = s.LastIndex()
if lastIndex >= 4 {
break
}
}
entries, err := s.Entries(lastIndex-1, lastIndex+1, noLimit)
if err != nil {
t.Fatal(err)
}
if len(entries) != 2 {
t.Fatalf("len(entries) = %d, want %d", len(entries), 2)
}
if !bytes.Equal(entries[0].Data, []byte("somedata")) {
t.Errorf("entries[0].Data = %v, want %v", entries[0].Data, []byte("somedata"))
}
if entries[1].Type != raftpb.EntryConfChange {
t.Fatalf("type = %v, want %v", entries[1].Type, raftpb.EntryConfChange)
}
if !bytes.Equal(entries[1].Data, ccdata) {
t.Errorf("data = %v, want %v", entries[1].Data, ccdata)
}
}
// StartNode returns a new Node given configuration and a list of raft peers.
// It appends a ConfChangeAddNode entry for each given peer to the initial log.
func StartNode(c *Config, peers []Peer) Node {
r := newRaft(c)
// become the follower at term 1 and apply initial configuration
// entries of term 1
r.becomeFollower(1, None)
for _, peer := range peers {
cc := pb.ConfChange{Type: pb.ConfChangeAddNode, NodeID: peer.ID, Context: peer.Context}
d, err := cc.Marshal()
if err != nil {
panic("unexpected marshal error")
}
e := pb.Entry{Type: pb.EntryConfChange, Term: 1, Index: r.raftLog.lastIndex() + 1, Data: d}
r.raftLog.append(e)
}
// Mark these initial entries as committed.
// TODO(bdarnell): These entries are still unstable; do we need to preserve
// the invariant that committed < unstable?
r.raftLog.committed = r.raftLog.lastIndex()
// Now apply them, mainly so that the application can call Campaign
// immediately after StartNode in tests. Note that these nodes will
// be added to raft twice: here and when the application's Ready
// loop calls ApplyConfChange. The calls to addNode must come after
// all calls to raftLog.append so progress.next is set after these
// bootstrapping entries (it is an error if we try to append these
// entries since they have already been committed).
// We do not set raftLog.applied so the application will be able
// to observe all conf changes via Ready.CommittedEntries.
for _, peer := range peers {
r.addNode(peer.ID)
}
n := newNode()
go n.run(r)
return &n
}
func (n *node) ProposeConfChange(ctx context.Context, cc pb.ConfChange) error {
data, err := cc.Marshal()
if err != nil {
return err
}
return n.Step(ctx, pb.Message{Type: pb.MsgProp, Entries: []pb.Entry{{Type: pb.EntryConfChange, Data: data}}})
}
func (c *RemoveCommand) Data5() ([]byte, error) {
req5 := raftpb.ConfChange{
ID: 0,
Type: raftpb.ConfChangeRemoveNode,
NodeID: c.id,
}
return req5.Marshal()
}
// TestRawNodeStart ensures that a node can be started correctly. The node should
// start with correct configuration change entries, and can accept and commit
// proposals.
func TestRawNodeStart(t *testing.T) {
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata, err := cc.Marshal()
if err != nil {
t.Fatalf("unexpected marshal error: %v", err)
}
wants := []Ready{
{
HardState: raftpb.HardState{Term: 1, Commit: 1, Vote: 0},
Entries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
},
CommittedEntries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
},
},
{
HardState: raftpb.HardState{Term: 2, Commit: 3, Vote: 1},
Entries: []raftpb.Entry{{Term: 2, Index: 3, Data: []byte("foo")}},
CommittedEntries: []raftpb.Entry{{Term: 2, Index: 3, Data: []byte("foo")}},
},
}
storage := NewMemoryStorage()
rawNode, err := NewRawNode(newTestConfig(1, nil, 10, 1, storage), []Peer{{ID: 1}})
if err != nil {
t.Fatal(err)
}
rd := rawNode.Ready()
t.Logf("rd %v", rd)
if !reflect.DeepEqual(rd, wants[0]) {
t.Fatalf("#%d: g = %+v,\n w %+v", 1, rd, wants[0])
} else {
storage.Append(rd.Entries)
rawNode.Advance(rd)
}
storage.Append(rd.Entries)
rawNode.Advance(rd)
rawNode.Campaign()
rd = rawNode.Ready()
storage.Append(rd.Entries)
rawNode.Advance(rd)
rawNode.Propose([]byte("foo"))
if rd = rawNode.Ready(); !reflect.DeepEqual(rd, wants[1]) {
t.Errorf("#%d: g = %+v,\n w %+v", 2, rd, wants[1])
} else {
storage.Append(rd.Entries)
rawNode.Advance(rd)
}
if rawNode.HasReady() {
t.Errorf("unexpected Ready: %+v", rawNode.Ready())
}
}
func (n *nodeConfChangeCommitterRecorder) ProposeConfChange(ctx context.Context, conf raftpb.ConfChange) error {
data, err := conf.Marshal()
if err != nil {
return err
}
n.index++
n.Record(testutil.Action{Name: "ProposeConfChange:" + conf.Type.String()})
n.readyc <- raft.Ready{CommittedEntries: []raftpb.Entry{{Index: n.index, Type: raftpb.EntryConfChange, Data: data}}}
return nil
}
// TestMultiNodeStart ensures that a node can be started correctly. The node should
// start with correct configuration change entries, and can accept and commit
// proposals.
func TestMultiNodeStart(t *testing.T) {
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata, err := cc.Marshal()
if err != nil {
t.Fatalf("unexpected marshal error: %v", err)
}
wants := []Ready{
{
SoftState: &SoftState{Lead: 1, RaftState: StateLeader},
HardState: raftpb.HardState{Term: 2, Commit: 2, Vote: 1},
Entries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
{Term: 2, Index: 2},
},
CommittedEntries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
{Term: 2, Index: 2},
},
},
{
HardState: raftpb.HardState{Term: 2, Commit: 3, Vote: 1},
Entries: []raftpb.Entry{{Term: 2, Index: 3, Data: []byte("foo")}},
CommittedEntries: []raftpb.Entry{{Term: 2, Index: 3, Data: []byte("foo")}},
},
}
mn := StartMultiNode(1)
storage := NewMemoryStorage()
mn.CreateGroup(1, newTestConfig(1, nil, 10, 1, storage), []Peer{{ID: 1}})
mn.Campaign(ctx, 1)
gs := <-mn.Ready()
g := gs[1]
if !reflect.DeepEqual(g, wants[0]) {
t.Fatalf("#%d: g = %+v,\n w %+v", 1, g, wants[0])
} else {
storage.Append(g.Entries)
mn.Advance(gs)
}
mn.Propose(ctx, 1, []byte("foo"))
if gs2 := <-mn.Ready(); !reflect.DeepEqual(gs2[1], wants[1]) {
t.Errorf("#%d: g = %+v,\n w %+v", 2, gs2[1], wants[1])
} else {
storage.Append(gs2[1].Entries)
mn.Advance(gs2)
}
select {
case rd := <-mn.Ready():
t.Errorf("unexpected Ready: %+v", rd)
case <-time.After(time.Millisecond):
}
}
// ProposeConfChange proposes a config change.
func (rn *RawNode) ProposeConfChange(cc pb.ConfChange) error {
data, err := cc.Marshal()
if err != nil {
return err
}
return rn.raft.Step(pb.Message{
Type: pb.MsgProp,
Entries: []pb.Entry{
{Type: pb.EntryConfChange, Data: data},
},
})
}
func (mn *multiNode) ProposeConfChange(ctx context.Context, group uint64, cc pb.ConfChange) error {
data, err := cc.Marshal()
if err != nil {
return err
}
return mn.Step(ctx, group,
pb.Message{
Type: pb.MsgProp,
Entries: []pb.Entry{
{Type: pb.EntryConfChange, Data: data},
},
})
}
// TestMultiNodeProposeConfig ensures that multiNode.ProposeConfChange
// sends the given configuration proposal to the underlying raft.
func TestMultiNodeProposeConfig(t *testing.T) {
mn := newMultiNode(1)
go mn.run()
s := NewMemoryStorage()
mn.CreateGroup(1, newTestConfig(1, nil, 10, 1, s), []Peer{{ID: 1}})
mn.Campaign(context.TODO(), 1)
proposed := false
var lastIndex uint64
var ccdata []byte
for {
rds := <-mn.Ready()
rd := rds[1]
s.Append(rd.Entries)
// change the step function to appendStep until this raft becomes leader
if !proposed && rd.SoftState.Lead == mn.id {
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
var err error
ccdata, err = cc.Marshal()
if err != nil {
t.Fatal(err)
}
mn.ProposeConfChange(context.TODO(), 1, cc)
proposed = true
}
mn.Advance(rds)
var err error
lastIndex, err = s.LastIndex()
if err != nil {
t.Fatal(err)
}
if lastIndex >= 3 {
break
}
}
mn.Stop()
entries, err := s.Entries(lastIndex, lastIndex+1, noLimit)
if err != nil {
t.Fatal(err)
}
if len(entries) != 1 {
t.Fatalf("len(entries) = %d, want %d", len(entries), 1)
}
if entries[0].Type != raftpb.EntryConfChange {
t.Fatalf("type = %v, want %v", entries[0].Type, raftpb.EntryConfChange)
}
if !bytes.Equal(entries[0].Data, ccdata) {
t.Errorf("data = %v, want %v", entries[0].Data, ccdata)
}
}
// makeWAL creates a WAL for the initial cluster
func makeWAL(waldir string, cl *membership.RaftCluster) {
if err := os.MkdirAll(waldir, 0755); err != nil {
ExitWithError(ExitIO, err)
}
m := cl.MemberByName(restoreName)
md := &etcdserverpb.Metadata{NodeID: uint64(m.ID), ClusterID: uint64(cl.ID())}
metadata, merr := md.Marshal()
if merr != nil {
ExitWithError(ExitInvalidInput, merr)
}
w, walerr := wal.Create(waldir, metadata)
if walerr != nil {
ExitWithError(ExitIO, walerr)
}
defer w.Close()
peers := make([]raft.Peer, len(cl.MemberIDs()))
for i, id := range cl.MemberIDs() {
ctx, err := json.Marshal((*cl).Member(id))
if err != nil {
ExitWithError(ExitInvalidInput, err)
}
peers[i] = raft.Peer{ID: uint64(id), Context: ctx}
}
ents := make([]raftpb.Entry, len(peers))
for i, p := range peers {
cc := raftpb.ConfChange{
Type: raftpb.ConfChangeAddNode,
NodeID: p.ID,
Context: p.Context}
d, err := cc.Marshal()
if err != nil {
ExitWithError(ExitInvalidInput, err)
}
e := raftpb.Entry{
Type: raftpb.EntryConfChange,
Term: 1,
Index: uint64(i + 1),
Data: d,
}
ents[i] = e
}
w.Save(raftpb.HardState{
Term: 1,
Vote: peers[0].ID,
Commit: uint64(len(ents))}, ents)
}
func TestNode(t *testing.T) {
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata, err := cc.Marshal()
if err != nil {
t.Fatalf("unexpected marshal error: %v", err)
}
wants := []Ready{
{
SoftState: &SoftState{Lead: 1, Nodes: []uint64{1}, RaftState: StateLeader},
HardState: raftpb.HardState{Term: 1, Commit: 2},
Entries: []raftpb.Entry{
{},
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
{Term: 1, Index: 2},
},
CommittedEntries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
{Term: 1, Index: 2},
},
},
{
HardState: raftpb.HardState{Term: 1, Commit: 3},
Entries: []raftpb.Entry{{Term: 1, Index: 3, Data: []byte("foo")}},
CommittedEntries: []raftpb.Entry{{Term: 1, Index: 3, Data: []byte("foo")}},
},
}
n := StartNode(1, []Peer{{ID: 1}}, 10, 1)
n.ApplyConfChange(cc)
n.Campaign(ctx)
if g := <-n.Ready(); !reflect.DeepEqual(g, wants[0]) {
t.Errorf("#%d: g = %+v,\n w %+v", 1, g, wants[0])
}
n.Propose(ctx, []byte("foo"))
if g := <-n.Ready(); !reflect.DeepEqual(g, wants[1]) {
t.Errorf("#%d: g = %+v,\n w %+v", 2, g, wants[1])
}
select {
case rd := <-n.Ready():
t.Errorf("unexpected Ready: %+v", rd)
default:
}
}
// NewRawNode returns a new RawNode given configuration and a list of raft peers.
func NewRawNode(config *Config, peers []Peer) (*RawNode, error) {
if config.ID == 0 {
panic("config.ID must not be zero")
}
r := newRaft(config)
rn := &RawNode{
raft: r,
}
lastIndex, err := config.Storage.LastIndex()
if err != nil {
panic(err) // TODO(bdarnell)
}
// If the log is empty, this is a new RawNode (like StartNode); otherwise it's
// restoring an existing RawNode (like RestartNode).
// TODO(bdarnell): rethink RawNode initialization and whether the application needs
// to be able to tell us when it expects the RawNode to exist.
if lastIndex == 0 {
r.becomeFollower(1, None)
ents := make([]pb.Entry, len(peers))
for i, peer := range peers {
cc := pb.ConfChange{Type: pb.ConfChangeAddNode, NodeID: peer.ID, Context: peer.Context}
data, err := cc.Marshal()
if err != nil {
panic("unexpected marshal error")
}
ents[i] = pb.Entry{Type: pb.EntryConfChange, Term: 1, Index: uint64(i + 1), Data: data}
}
r.raftLog.append(ents...)
r.raftLog.committed = uint64(len(ents))
for _, peer := range peers {
r.addNode(peer.ID)
}
}
// Set the initial hard and soft states after performing all initialization.
rn.prevSoftSt = r.softState()
if lastIndex == 0 {
rn.prevHardSt = emptyState
} else {
rn.prevHardSt = r.hardState()
}
return rn, nil
}
// StartNode returns a new Node given a unique raft id, a list of raft peers, and
// the election and heartbeat timeouts in units of ticks.
// It also builds ConfChangeAddNode entry for each peer and puts them at the head of the log.
func StartNode(id uint64, peers []Peer, election, heartbeat int) Node {
n := newNode()
r := newRaft(id, nil, election, heartbeat)
for _, peer := range peers {
cc := pb.ConfChange{Type: pb.ConfChangeAddNode, NodeID: peer.ID, Context: peer.Context}
d, err := cc.Marshal()
if err != nil {
panic("unexpected marshal error")
}
e := pb.Entry{Type: pb.EntryConfChange, Term: 1, Index: r.raftLog.lastIndex() + 1, Data: d}
r.raftLog.append(r.raftLog.lastIndex(), e)
}
r.raftLog.committed = r.raftLog.lastIndex()
go n.run(r)
return &n
}
// StartNode returns a new Node given a unique raft id, a list of raft peers, and
// the election and heartbeat timeouts in units of ticks.
// It appends a ConfChangeAddNode entry for each given peer to the initial log.
func StartNode(id uint64, peers []Peer, election, heartbeat int, storage Storage) Node {
n := newNode()
r := newRaft(id, nil, election, heartbeat, storage)
for _, peer := range peers {
cc := pb.ConfChange{Type: pb.ConfChangeAddNode, NodeID: peer.ID, Context: peer.Context}
d, err := cc.Marshal()
if err != nil {
panic("unexpected marshal error")
}
e := pb.Entry{Type: pb.EntryConfChange, Term: 1, Index: r.raftLog.lastIndex() + 1, Data: d}
r.raftLog.append(r.raftLog.lastIndex(), e)
}
// Mark these initial entries as committed.
// TODO(bdarnell): These entries are still unstable; do we need to preserve
// the invariant that committed < unstable?
r.raftLog.committed = r.raftLog.lastIndex()
go n.run(r)
return &n
}
// TestNodeProposeConfig ensures that node.ProposeConfChange sends the given configuration proposal
// to the underlying raft.
func TestNodeProposeConfig(t *testing.T) {
msgs := []raftpb.Message{}
appendStep := func(r *raft, m raftpb.Message) {
msgs = append(msgs, m)
}
n := newNode()
s := NewMemoryStorage()
r := newTestRaft(1, []uint64{1}, 10, 1, s)
go n.run(r)
n.Campaign(context.TODO())
for {
rd := <-n.Ready()
s.Append(rd.Entries)
// change the step function to appendStep until this raft becomes leader
if rd.SoftState.Lead == r.id {
r.step = appendStep
n.Advance()
break
}
n.Advance()
}
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata, err := cc.Marshal()
if err != nil {
t.Fatal(err)
}
n.ProposeConfChange(context.TODO(), cc)
n.Stop()
if len(msgs) != 1 {
t.Fatalf("len(msgs) = %d, want %d", len(msgs), 1)
}
if msgs[0].Type != raftpb.MsgProp {
t.Errorf("msg type = %d, want %d", msgs[0].Type, raftpb.MsgProp)
}
if !bytes.Equal(msgs[0].Entries[0].Data, ccdata) {
t.Errorf("data = %v, want %v", msgs[0].Entries[0].Data, ccdata)
}
}
// TestNodeProposeAddDuplicateNode ensures that two proposes to add the same node should
// not affect the later propose to add new node.
func TestNodeProposeAddDuplicateNode(t *testing.T) {
n := newNode()
s := NewMemoryStorage()
r := newTestRaft(1, []uint64{1}, 10, 1, s)
go n.run(r)
n.Campaign(context.TODO())
rdyEntries := make([]raftpb.Entry, 0)
ticker := time.NewTicker(time.Millisecond * 100)
done := make(chan struct{})
stop := make(chan struct{})
applyConfChan := make(chan struct{})
go func() {
defer close(done)
for {
select {
case <-stop:
return
case <-ticker.C:
n.Tick()
case rd := <-n.Ready():
s.Append(rd.Entries)
for _, e := range rd.Entries {
rdyEntries = append(rdyEntries, e)
switch e.Type {
case raftpb.EntryNormal:
case raftpb.EntryConfChange:
var cc raftpb.ConfChange
cc.Unmarshal(e.Data)
n.ApplyConfChange(cc)
applyConfChan <- struct{}{}
}
}
n.Advance()
}
}
}()
cc1 := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata1, _ := cc1.Marshal()
n.ProposeConfChange(context.TODO(), cc1)
<-applyConfChan
// try add the same node again
n.ProposeConfChange(context.TODO(), cc1)
<-applyConfChan
// the new node join should be ok
cc2 := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 2}
ccdata2, _ := cc2.Marshal()
n.ProposeConfChange(context.TODO(), cc2)
<-applyConfChan
close(stop)
<-done
if len(rdyEntries) != 4 {
t.Errorf("len(entry) = %d, want %d, %v\n", len(rdyEntries), 3, rdyEntries)
}
if !bytes.Equal(rdyEntries[1].Data, ccdata1) {
t.Errorf("data = %v, want %v", rdyEntries[1].Data, ccdata1)
}
if !bytes.Equal(rdyEntries[3].Data, ccdata2) {
t.Errorf("data = %v, want %v", rdyEntries[3].Data, ccdata2)
}
n.Stop()
}
// TestNodeStart ensures that a node can be started correctly. The node should
// start with correct configuration change entries, and can accept and commit
// proposals.
func TestNodeStart(t *testing.T) {
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
cc := raftpb.ConfChange{Type: raftpb.ConfChangeAddNode, NodeID: 1}
ccdata, err := cc.Marshal()
if err != nil {
t.Fatalf("unexpected marshal error: %v", err)
}
wants := []Ready{
{
HardState: raftpb.HardState{Term: 1, Commit: 1, Vote: 0},
Entries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
},
CommittedEntries: []raftpb.Entry{
{Type: raftpb.EntryConfChange, Term: 1, Index: 1, Data: ccdata},
},
},
{
HardState: raftpb.HardState{Term: 2, Commit: 3, Vote: 1},
Entries: []raftpb.Entry{{Term: 2, Index: 3, Data: []byte("foo")}},
CommittedEntries: []raftpb.Entry{{Term: 2, Index: 3, Data: []byte("foo")}},
},
}
storage := NewMemoryStorage()
c := &Config{
ID: 1,
ElectionTick: 10,
HeartbeatTick: 1,
Storage: storage,
MaxSizePerMsg: noLimit,
MaxInflightMsgs: 256,
}
n := StartNode(c, []Peer{{ID: 1}})
defer n.Stop()
g := <-n.Ready()
if !reflect.DeepEqual(g, wants[0]) {
t.Fatalf("#%d: g = %+v,\n w %+v", 1, g, wants[0])
} else {
storage.Append(g.Entries)
n.Advance()
}
n.Campaign(ctx)
rd := <-n.Ready()
storage.Append(rd.Entries)
n.Advance()
n.Propose(ctx, []byte("foo"))
if g2 := <-n.Ready(); !reflect.DeepEqual(g2, wants[1]) {
t.Errorf("#%d: g = %+v,\n w %+v", 2, g2, wants[1])
} else {
storage.Append(g2.Entries)
n.Advance()
}
select {
case rd := <-n.Ready():
t.Errorf("unexpected Ready: %+v", rd)
case <-time.After(time.Millisecond):
}
}
// makeWAL creates a WAL for the initial cluster
func makeWALAndSnap(waldir, snapdir string, cl *membership.RaftCluster) {
if err := fileutil.CreateDirAll(waldir); err != nil {
ExitWithError(ExitIO, err)
}
// add members again to persist them to the store we create.
st := store.New(etcdserver.StoreClusterPrefix, etcdserver.StoreKeysPrefix)
cl.SetStore(st)
for _, m := range cl.Members() {
cl.AddMember(m)
}
m := cl.MemberByName(restoreName)
md := &etcdserverpb.Metadata{NodeID: uint64(m.ID), ClusterID: uint64(cl.ID())}
metadata, merr := md.Marshal()
if merr != nil {
ExitWithError(ExitInvalidInput, merr)
}
w, walerr := wal.Create(waldir, metadata)
if walerr != nil {
ExitWithError(ExitIO, walerr)
}
defer w.Close()
peers := make([]raft.Peer, len(cl.MemberIDs()))
for i, id := range cl.MemberIDs() {
ctx, err := json.Marshal((*cl).Member(id))
if err != nil {
ExitWithError(ExitInvalidInput, err)
}
peers[i] = raft.Peer{ID: uint64(id), Context: ctx}
}
ents := make([]raftpb.Entry, len(peers))
nodeIDs := make([]uint64, len(peers))
for i, p := range peers {
nodeIDs[i] = p.ID
cc := raftpb.ConfChange{
Type: raftpb.ConfChangeAddNode,
NodeID: p.ID,
Context: p.Context}
d, err := cc.Marshal()
if err != nil {
ExitWithError(ExitInvalidInput, err)
}
e := raftpb.Entry{
Type: raftpb.EntryConfChange,
Term: 1,
Index: uint64(i + 1),
Data: d,
}
ents[i] = e
}
commit, term := uint64(len(ents)), uint64(1)
if err := w.Save(raftpb.HardState{
Term: term,
Vote: peers[0].ID,
Commit: commit}, ents); err != nil {
ExitWithError(ExitIO, err)
}
b, berr := st.Save()
if berr != nil {
ExitWithError(ExitError, berr)
}
raftSnap := raftpb.Snapshot{
Data: b,
Metadata: raftpb.SnapshotMetadata{
Index: commit,
Term: term,
ConfState: raftpb.ConfState{
Nodes: nodeIDs,
},
},
}
snapshotter := snap.New(snapdir)
if err := snapshotter.SaveSnap(raftSnap); err != nil {
panic(err)
}
if err := w.SaveSnapshot(walpb.Snapshot{Index: commit, Term: term}); err != nil {
ExitWithError(ExitIO, err)
}
}
func (mn *multiNode) run() {
groups := map[uint64]*groupState{}
rds := map[uint64]Ready{}
var advancec chan map[uint64]Ready
for {
// Only select readyc if we have something to report and we are not
// currently waiting for an advance.
readyc := mn.readyc
if len(rds) == 0 || advancec != nil {
readyc = nil
}
// group points to the group that was touched on this iteration (if any)
var group *groupState
select {
case gc := <-mn.groupc:
gc.config.ID = mn.id
r := newRaft(gc.config)
group = &groupState{
id: gc.id,
raft: r,
}
groups[gc.id] = group
lastIndex, err := gc.config.Storage.LastIndex()
if err != nil {
panic(err) // TODO(bdarnell)
}
// If the log is empty, this is a new group (like StartNode); otherwise it's
// restoring an existing group (like RestartNode).
// TODO(bdarnell): rethink group initialization and whether the application needs
// to be able to tell us when it expects the group to exist.
if lastIndex == 0 {
r.becomeFollower(1, None)
ents := make([]pb.Entry, len(gc.peers))
for i, peer := range gc.peers {
cc := pb.ConfChange{Type: pb.ConfChangeAddNode, NodeID: peer.ID, Context: peer.Context}
data, err := cc.Marshal()
if err != nil {
panic("unexpected marshal error")
}
ents[i] = pb.Entry{Type: pb.EntryConfChange, Term: 1, Index: uint64(i + 1), Data: data}
}
r.raftLog.append(ents...)
r.raftLog.committed = uint64(len(ents))
for _, peer := range gc.peers {
r.addNode(peer.ID)
}
}
// Set the initial hard and soft states after performing all initialization.
group.prevSoftSt = r.softState()
group.prevHardSt = r.HardState
close(gc.done)
case gr := <-mn.rmgroupc:
delete(groups, gr.id)
delete(rds, gr.id)
close(gr.done)
case mm := <-mn.propc:
// TODO(bdarnell): single-node impl doesn't read from propc unless the group
// has a leader; we can't do that since we have one propc for many groups.
// We'll have to buffer somewhere on a group-by-group basis, or just let
// raft.Step drop any such proposals on the floor.
mm.msg.From = mn.id
var ok bool
if group, ok = groups[mm.group]; ok {
group.raft.Step(mm.msg)
}
case mm := <-mn.recvc:
group = groups[mm.group]
if _, ok := group.raft.prs[mm.msg.From]; ok || !IsResponseMsg(mm.msg) {
group.raft.Step(mm.msg)
}
case mcc := <-mn.confc:
group = groups[mcc.group]
if mcc.msg.NodeID == None {
group.raft.resetPendingConf()
select {
case mcc.ch <- pb.ConfState{Nodes: group.raft.nodes()}:
case <-mn.done:
}
break
}
switch mcc.msg.Type {
case pb.ConfChangeAddNode:
group.raft.addNode(mcc.msg.NodeID)
case pb.ConfChangeRemoveNode:
group.raft.removeNode(mcc.msg.NodeID)
case pb.ConfChangeUpdateNode:
group.raft.resetPendingConf()
default:
panic("unexpected conf type")
}
select {
case mcc.ch <- pb.ConfState{Nodes: group.raft.nodes()}:
case <-mn.done:
}
case <-mn.tickc:
// TODO(bdarnell): instead of calling every group on every tick,
// we should have a priority queue of groups based on their next
// time-based event.
for _, g := range groups {
g.raft.tick()
rd := g.newReady()
if rd.containsUpdates() {
rds[g.id] = rd
}
}
case readyc <- rds:
// Clear outgoing messages as soon as we've passed them to the application.
for g := range rds {
groups[g].raft.msgs = nil
}
rds = map[uint64]Ready{}
advancec = mn.advancec
case advs := <-advancec:
for groupID, rd := range advs {
g, ok := groups[groupID]
if !ok {
continue
}
g.commitReady(rd)
// We've been accumulating new entries in rds which may now be obsolete.
// Drop the old Ready object and create a new one if needed.
delete(rds, groupID)
newRd := g.newReady()
if newRd.containsUpdates() {
rds[groupID] = newRd
}
}
advancec = nil
case ms := <-mn.status:
if g, ok := groups[ms.group]; ok {
s := getStatus(g.raft)
ms.ch <- &s
} else {
ms.ch <- nil
}
case <-mn.stop:
close(mn.done)
return
}
if group != nil {
rd := group.newReady()
if rd.containsUpdates() {
rds[group.id] = rd
}
}
}
}