[KAFKA-1188] Stale LeaderAndIsr request could be handled by the broker on Controller failover

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Assignee: Guozhang Wang
• Reporter: Guozhang Wang
• Created: 18/Dec/13 19:31
• Updated: 13/Feb/14 22:47
• Resolved: 13/Feb/14 22:47

Description

... which will cause the broker to truncate log as responding in makeFollower, and hence lose data.

One procedure to produce this issue:

3 brokers, 3 partitions, replication factor = 3.

1. Broker 1 is the original controller.
2. Broker 3 lost registration, 1 send LeaderAndIsr to 1 for isr shrinking (leader of partition 1), but not to 2 since it is not the leader.
3. Broker 1 dies, new controller 2 sends all LeaderAndIsr to 2 and 3 (resumed), and 2 handles the makeFollower, truncates data.

Issue Links

• blocks: KAFKA-1211 Hold the produce request with ack > 1 in purgatory until replicas' HW has larger than the produce offset (KIP-101)