This message was deleted.

We are waiting for the fixes too :) There are other issues which may cause the infinite loop, connected with 3.1 client (workaround is to use latest 2.x client)

Thanks for your response. We rely on topic compaction heavily. Does it make sense then to use the latest 2.x client and keep all the cluster services (Broker, etc.) at version 3.x? The environment we use is very standard, os4, a private network, Pulsar 3.1.1 and 3 pods for each brokers, bookkeeper and zookeeper, when I ran the scenario we had replication between identical clusters and as I mentioned it seems like the bug is more likely to happen when the backlog is not very short.

The issue may be not with the compaction. 2.x clients should be compatible with 3.1, try to change it in your application and see if the issue occurs. I would also recommend opening the GH issue with the description on how to reproduce the issue. Add as much details as possible, Pulsar team is really helpful in fixing the issues, but they need to know how to reproduce the issue.

Yes, it happens with exclusive subscription. It also happens only when I readAsCompacted, could that still mean that the bug is on the clients side (and not with the compaction itself or somewhere in the Brokers side)? I guess I'll try client 2.x anyway. We will open a bug on GitHub.

cc @Cong Zhao

https://github.com/apache/pulsar/pull/21187 fixed https://github.com/apache/pulsar/issues/21074, I will cherry-pick this fix to other release branches.

Sounds good! Do you have an estimate in what release version this bug will be fixed?

@Cong Zhao ☝️

This fix will release in release/2.10.6、release/2.11.4、release/3.0.3、release/3.1.3