Description
We hit a situation where the connect worker seems to have been requesting group rejoins repeatedly for some reason. In theory the same thing could happen to the consumer if a user repeatedly changes the subscribed topics in a tight loop. Currently AbstractCoordinator would not backoff in these cases because the rebalances may all be completing successfully. It probably makes sense to enforce a minimum time between locally triggered rebalances in order to avoid overloading the broker with repeated group rebalances.
Attachments
Issue Links
- is related to
-
KAFKA-9678 Introduce bounded exponential backoff in clients
-
- Resolved
-