Page History

...

An earlier iteration of this proposal considered a change to the behavior of group.min.session.timeout.ms and group.max.session.timeout.ms. The idea was to let the coordinator automatically adjust the session timeout provided by the client to be within this range. This would give operators a way to change the allowed session timeout range without causing existing clients to fail. Unfortunately, this does not work gracefully with all clients. In particular, librdkafka-based consumers enforce the session timeout locally. If the session timeout is reached without response from the coordinator, then partitions are automatically revoked and the consumer rejoins the group as a new member. The coordinator, on the other hand, would still enforce the adjusted session timeout, which means that the rebalance would get delayed until the old member could be expired. For this reason, we decided to reject this option for now. We may reconsider it in a separate proposal which allows the coordinator to propagate the adjusted session timeout to the client.