an issue from async-raft

[pfwang80s]

how to repro:

1. start a single node 0 cluster
2. add non-vote node 1 to the cluster
3. add non-vote node 2 to the cluster
4. change membership {0, 1, 2}, now all these 3 nodes are in one cluster
5. kill node 0, the leader
6. a lot of "ERROR async_raft::replication: error sending AppendEntries RPC to target" errors are issued on node 1 or 2.
7. connect to the new leader, change membership to {1, 2} to evict node 0, but 6 still continues

it seems there is a ReplicationStream to node 0 after step 7, it runs forever as RaftNetwork.append_entries to node 0 returns an network error, but target_state of it is not updated, check async-raft/src/replication/mod.rs:223.

Does openraft have the same problem?

thanks.

[github-actions]

👋 Thanks for opening this issue!

Get help or engage by:

• /help : to print help messages.
• /assignme : to assign this issue to you.

[drmingdrmer]

Yes Openraft has fixed this issue.

Every time replication changes, such as leader establish or quit, or membership change, the replication stream will be completely shutdown and restart again. All replication update is dealt with here:

openraft/openraft/src/core/raft_core.rs

Lines 1632 to 1644 in d8e0417

	Command::RebuildReplicationStreams { targets } => {
	self.remove_all_replication().await;
	for (target, matching) in targets.iter() {
	let handle = self.spawn_replication_stream(*target, *matching).await;
	if let Some(l) = &mut self.leader_data {
	l.replications.insert(*target, handle);
	} else {
	unreachable!("it has to be a leader!!!");
	}
	}
	}

[pfwang80s]

I do love OpenRaft now, thank you very much~

[pfwang80s]

issue closed.