Page History

...

When a rebalance is triggered, before rejoining the cluster group, all workers will preemptively stop all tasks of all source connectors for which task configurations are present in the config topic after the latest task count record. This step is not necessary to preserve exactly-once delivery guarantees, but should provide a reasonable opportunity for tasks to shut down gracefully before being forcibly fenced out.

Stopping these tasks will be done in parallel, and any tasks that the worker cannot stop within the graceful task shutdown timeout period (which defaults to five seconds but can be controlled via the task.shutdown.graceful.timeout.ms  worker property) will be abandoned and allowed to continue running. They will be disabled later on when the leader fences out their producers during the rebalance (described below).

Because this work will be done in parallel on an executor that currently uses eight threads and the default timeout for task shutdown is fairly low (five seconds), it is highly unlikely that this will hinder a worker's ability to rejoin a group before the rebalance timeout (which defaults to sixty seconds) times out.

Leader

During rebalance, the leader will take on a few extra responsibilities. Once all group members have joined and it is time to perform assignment the leader will:

...