ignore old leader's leases revoking request

Signed-off-by: Benjamin Wang <[email protected]>

server/etcdserver/server.go

@@ -858,8 +858,18 @@ func (s *EtcdServer) run() {
 }
 
 func (s *EtcdServer) revokeExpiredLeases(leases []*lease.Lease) {
+	// We shouldn't revoke any leases if current member isn't a leader,
+	// because the operation should only be performed by the leader. When
+	// the leader gets blocked on the raft loop, such as writing WAL entries,
+	// it can't process any events or messages from raft. It may think it
+	// is still the leader even the leader has already changed.
+	// Refer to https://github.com/etcd-io/etcd/issues/15247
+	lg := s.Logger()
+	if !s.ensureLeadership() {
+		lg.Warn("Ignore the lease revoking request because current member isn't a leader")
+		return
+	}
 	s.GoAttach(func() {
-		lg := s.Logger()
 		// Increases throughput of expired leases deletion process through parallelization
 		c := make(chan struct{}, maxPendingRevokes)
 		for _, curLease := range leases {
@@ -892,6 +902,26 @@ func (s *EtcdServer) revokeExpiredLeases(leases []*lease.Lease) {
 	})
 }
 
+// ensureLeadership checks whether current member is still the leader.
+func (s *EtcdServer) ensureLeadership() bool {
+	lg := s.Logger()
+	if err := s.linearizableReadNotify(s.ctx); err != nil {
+		lg.Warn("Failed to check current member's leadership",
+			zap.Error(err))
+		return false
+	}
+
+	newLeaderId := s.raftStatus().Lead
+	if newLeaderId != uint64(s.MemberId()) {
+		lg.Warn("Current member isn't a leader",
+			zap.Uint64("local-member-id", uint64(s.MemberId())),
+			zap.Uint64("new-lead", newLeaderId))
+		return false
+	}
+
+	return true
+}
+
 // Cleanup removes allocated objects by EtcdServer.NewServer in
 // situation that EtcdServer::Start was not called (that takes care of cleanup).
 func (s *EtcdServer) Cleanup() {