Page History

...

In order to support submitting extra more reassignments while existing reassignments are still in-flight.  An extra znode /admin/reassign_partitions_queue  which has the same JSON format as /admin/reassign_partitions.   Three more options --generate-queue  --verify-queue  --execute-queue  will be added to  kafka-reassign-partitions.sh. The controller will be informed of the extra queued reassignments via a ZooKeeper watch.   It will get all topic/partitions from /admin/reassign_partitions_queue  queue  and  add to /admin/reassign_partitions,  then trigger the reassignments onPartitionReassignment()  of the  topic/partitions.   

In case inside the /admin/reassign_partitions_queue,  there are topic/partitions which exist in /admin/reassign_partitions (pending reassignments),  the conflict resolution for those duplicate  topic/partitions is to first  cancel / rollback the reassignments of those topic/partitions in /admin/reassign_partitions,  then submit new reassignments from /admin/reassign_partitions_queuequeue to /admin/reassign_partitions.   This  This approach will be simpler than the algorithm proposed by Tom previously to infer the final replicas assignments for those duplicate  topic/partitions.   After the topic/partition is put in /admin/reassign_partitions,  it will be removed from /admin/reassign_partitions_queue,  and when /admin/reassign_partitions_queue is empty,  the znode will be deleted. 

...