Hi team, I was seeing an issue where a mirror maker attempted to commit an offset for a partition that was ahead of log end offset and in the mean time the leader of the partition was being restarted. In theory only committed messages can be consumed by consumer which means the messages received by mirror maker should be available in all ISR nodes. I want to know why mirror maker gets a message with offset that is larger than the log end offset?
The kafka version is 0.10.0.0 and the replica of the partition is 3 I can see below INFO from the leader for the partition that caused the mirror maker problem [2016-08-09 03:25:43,508] INFO [KafkaApi-1] Closing connection due to error during produce request with correlation id -1428403918 from client id xxxxx with ack=0 Topic and partition to exceptions: topic-0 -> org.apache.kafka.common.errors.NotLeaderForPartitionException (kafka.server.KafkaApis)