etcdserver: wait for raft is notified on confChange before responding to client

Signed-off-by: Benjamin Wang <wachao@vmware.com>

server/etcdserver/raft.go

@@ -72,6 +72,8 @@ type toApply struct {
 	snapshot raftpb.Snapshot
 	// notifyc synchronizes etcd server applies with the raft node
 	notifyc chan struct{}
+	// confChangeCh synchronizes etcd server applies confChange with raft node
+	confChangeCh chan struct{}
 }
 
 type raftNode struct {
@@ -203,10 +205,12 @@ func (r *raftNode) start(rh *raftReadyHandler) {
 				}
 
 				notifyc := make(chan struct{}, 1)
+				confChangeCh := make(chan struct{}, 1)
 				ap := toApply{
 					entries:      rd.CommittedEntries,
 					snapshot:     rd.Snapshot,
 					notifyc:      notifyc,
+					confChangeCh: confChangeCh,
 				}
 
 				updateCommittedIndex(&ap, rh)
@@ -269,6 +273,14 @@ func (r *raftNode) start(rh *raftReadyHandler) {
 
 				r.raftStorage.Append(rd.Entries)
 
+				waitApply := false
+				for _, ent := range rd.CommittedEntries {
+					if ent.Type == raftpb.EntryConfChange {
+						waitApply = true
+						break
+					}
+				}
+
 				if !islead {
 					// finish processing incoming messages before we signal notifyc chan
 					msgs := r.processMessages(rd.Messages)
@@ -283,13 +295,7 @@ func (r *raftNode) start(rh *raftReadyHandler) {
 					// on its own single-node cluster, before toApply-layer applies the config change.
 					// We simply wait for ALL pending entries to be applied for now.
 					// We might improve this later on if it causes unnecessary long blocking issues.
-					waitApply := false
+
-					for _, ent := range rd.CommittedEntries {
-						if ent.Type == raftpb.EntryConfChange {
-							waitApply = true
-							break
-						}
-					}
 					if waitApply {
 						// blocks until 'applyAll' calls 'applyWait.Trigger'
 						// to be in sync with scheduled config-change job
@@ -310,6 +316,11 @@ func (r *raftNode) start(rh *raftReadyHandler) {
 
 				// gofail: var raftBeforeAdvance struct{}
 				r.Advance()
+
+				if waitApply {
+					// notify etcdserver that raft has already been notified or advanced.
+					confChangeCh <- struct{}{}
+				}
 			case <-r.stopped:
 				return
 			}

server/etcdserver/server.go

@@ -1133,7 +1133,7 @@ func (s *EtcdServer) applyEntries(ep *etcdProgress, apply *toApply) {
 		return
 	}
 	var shouldstop bool
-	if ep.appliedt, ep.appliedi, shouldstop = s.apply(ents, &ep.confState); shouldstop {
+	if ep.appliedt, ep.appliedi, shouldstop = s.apply(ents, &ep.confState, apply.confChangeCh); shouldstop {
 		go s.stopWithDelay(10*100*time.Millisecond, fmt.Errorf("the member has been permanently removed from the cluster"))
 	}
 }
@@ -1810,6 +1810,7 @@ func (s *EtcdServer) sendMergedSnap(merged snap.Message) {
 func (s *EtcdServer) apply(
 	es []raftpb.Entry,
 	confState *raftpb.ConfState,
+	confChangeCh chan struct{},
 ) (appliedt uint64, appliedi uint64, shouldStop bool) {
 	s.lg.Debug("Applying entries", zap.Int("num-entries", len(es)))
 	for i := range es {
@@ -1841,6 +1842,18 @@ func (s *EtcdServer) apply(
 			s.setAppliedIndex(e.Index)
 			s.setTerm(e.Term)
 			shouldStop = shouldStop || removedSelf
+
+			// etcdserver need to ensure the raft has already been notified
+			// or advanced before it responds to the client. Otherwise, the
+			// following config change request may be rejected.
+			// See https://github.com/etcd-io/etcd/issues/15528.
+			select {
+			case <-time.After(500 * time.Millisecond):
+				lg := s.Logger()
+				lg.Warn("timed out waiting for configChange notification")
+			case <-confChangeCh:
+			}
+
 			s.w.Trigger(cc.ID, &confChangeResponse{s.cluster.Members(), err})
 
 		default:

server/etcdserver/server_test.go

@@ -689,7 +689,8 @@ func TestApplyConfigChangeUpdatesConsistIndex(t *testing.T) {
 		Data:  pbutil.MustMarshal(cc),
 	}}
 
-	_, appliedi, _ := srv.apply(ents, &raftpb.ConfState{})
+	confChangeCh := make(chan struct{}, 1)
+	_, appliedi, _ := srv.apply(ents, &raftpb.ConfState{}, confChangeCh)
 	consistIndex := srv.consistIndex.ConsistentIndex()
 	assert.Equal(t, uint64(2), appliedi)
 
@@ -763,7 +764,8 @@ func TestApplyMultiConfChangeShouldStop(t *testing.T) {
 		ents = append(ents, ent)
 	}
 
-	_, _, shouldStop := srv.apply(ents, &raftpb.ConfState{})
+	confChangeCh := make(chan struct{}, 1)
+	_, _, shouldStop := srv.apply(ents, &raftpb.ConfState{}, confChangeCh)
 	if !shouldStop {
 		t.Errorf("shouldStop = %t, want %t", shouldStop, true)
 	}