Bug: one of the ectd server fails to sent observe responses

[exceptionplayer]

I have a etcd cluster consists of 5 etcd server instances and about 100 clients will use the election.observe() method to get notified when the leader of some election(let's call it electionA) changes.

the clients will connect to the cluster by specifying all the 5 endpoints, it seems that one of the endpoint(etcd server) fails to notify our clients when the leader of electionA changes. When I change the endpoints to let the clients connect to other enpoints, the notification works well, also when I use ectdctl elect -l xxx on that server(who failed to notify the clients), it works well too.

I noticed that the endpoints for all the clients are the same, for example: serverA,severB,serverC,serverD,serverE, so it seems that all of the clients will try to connect to serverA first, so for serverA(the failed server), the traffic may be higher than others.

But the total number of our clients is less than 200.

Do you guys have any ideas about this? Or how can I troubleshoot this issue? Thank you!

After I restarted the server, it worked properly

I am using Etcd 3.5.16.