coordinator: fix too much cursor causes by ensemble breaking

coordinator/impl/shard_controller.go

@@ -368,6 +368,9 @@ func (s *shardController) electLeader() error {
 
 func (s *shardController) deletingRemovedNodes() error {
 	for _, ds := range s.shardMetadata.RemovedNodes {
+		if _, err := s.newTerm(s.ctx, ds); err != nil {
+			return err
+		}
 		if _, err := s.rpc.DeleteShard(s.ctx, ds, &proto.DeleteShardRequest{
 			Namespace: s.namespace,
 			ShardId:   s.shard,
@@ -497,8 +500,8 @@ func (s *shardController) internalNewTermAndAddFollower(ctx context.Context, nod
 func (s *shardController) newTermQuorum() (map[model.ServerAddress]*proto.EntryId, error) { //nolint:revive
 	timer := s.newTermQuorumLatency.Timer()
 
-	fencingQuorum := mergeLists(s.shardMetadata.Ensemble, s.shardMetadata.RemovedNodes)
-	fencingQuorumSize := len(fencingQuorum)
+	ensembleQuorum := s.shardMetadata.Ensemble
+	fencingQuorumSize := len(ensembleQuorum)
 	majority := fencingQuorumSize/2 + 1
 
 	// Use a new context, so we can cancel the pending requests
@@ -512,7 +515,7 @@ func (s *shardController) newTermQuorum() (map[model.ServerAddress]*proto.EntryI
 		error
 	}, fencingQuorumSize)
 
-	for _, sa := range fencingQuorum {
+	for _, sa := range ensembleQuorum {
 		// We need to save the address because it gets modified in the loop
 		serverAddress := sa
 		go common.DoWithLabels(
@@ -559,11 +562,8 @@ func (s *shardController) newTermQuorum() (map[model.ServerAddress]*proto.EntryI
 		totalResponses++
 		if r.error == nil {
 			successResponses++
-
 			// We don't consider the removed nodes as candidates for leader/followers
-			if listContains(s.shardMetadata.Ensemble, r.ServerAddress) {
-				res[r.ServerAddress] = r.EntryId
-			}
+			res[r.ServerAddress] = r.EntryId
 		} else {
 			err = multierr.Append(err, r.error)
 		}