After doing some partition reassignments, I've ended up with some partitions that have both the old and new brokers assigned.
The output of kafka-topics.sh --describe looks like this: Topic:cs-es-indexer-a PartitionCount:30 ReplicationFactor:2 Configs: retention.ms=10800000 ... Topic: cs-es-indexer-a Partition: 11 Leader: 11 Replicas: 11,12 Isr: 11,12 Topic: cs-es-indexer-a Partition: 12 Leader: 15 Replicas: 15,17,12 Isr: 12,17,15 ... Partition 12 has 3 replicas, but the topic has a replication factor of 2. It was reassigned from [12,17] to [15,17]. During the reassignment, the controller changed from one broker to another, and the resigning controller deadlocked with https://issues.apache.org/jira/browse/KAFKA-1317. Is this expected, or a known issue? I've had to alter some of my monitoring tools to handle different-sized replica lists for partitions of the same topic.