What happens when a Kafka broker goes down?

When a broker fails: 1) The controller detects the failure via heartbeat timeout. 2) For partitions where the failed broker was leader, the controller elects a new leader from the ISR. 3) For partitions where it was a follower, it's removed from the ISR. Producer metadata becomes stale and is refreshed on the next request. Consumers reconnect to the new leaders. If the failed broker was the controller itself, a new controller is elected. When the broker restarts, it catches up and rejoins the ISR.

Expected key concepts: controller, heartbeat, new leader, ISR, election, stale metadata, refresh, follower, removed, restart, catch up