Page History

...

1. The stream clients shall not downgrade to group_coordinator:v1 once they upgrade to group_coordinator:v2, even if the leader instructs them to do so. This is because, in the DescribeFeatures response that contains group_coordinator:v2, the epoch for the features is higher than earlier response. The client uses the epoch to ignore the earlier response from the broker. (see this section to learn about the monotonically increasing epoch number). The solution is to persist the EOS model flag as part of the consumers' metadata when the rebalance completes, so that on a second rebalance as long as some members are already on the beta version, the leader of the group will not kick off another feature request because it knows an upgrade was ongoing in last generation, based on the eventual consistency guarantee. It will just continue to inform all the members to upgrade to beta version. 

2. It is important to note that the user is only upgrading the cluster, from enabling group_coordinator:v1 to enabling group_coordinator:v1-v2. Therefore, once a client sees that v2 is enabled on any broker, it should be a clear signal that all broker binaries within the cluster would support group_coordinator:v2 semantic, it’s just a matter of time before feature version update metadata propagates to every broker from controller.

...