[ https://issues.apache.org/jira/browse/KAFKA-6794?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16777149#comment-16777149 ]
GEORGE LI edited comment on KAFKA-6794 at 2/25/19 7:56 PM: ----------------------------------------------------------- Hi, [~viktorsomogyi], When I rebalance the whole cluster, I generate the reassignment plan json with a list of topic/partitions with its new_replicas/original_replicas, and sort them by their size, so try to group them in batches of similar sizes for execution, so that they are expected to complete reassignment using about the same amount of time. Say there are 1000 reassignments, and 50 per batch. That will be at least 20 batches/buckets to put in for execution. there could be > 20 batches because a reassignment like (1,2,3,4) => (5,6,7,8) can be split into 4 reassignments in 4 batches. A batch will be submitted, and an execution program will keep checking the existence of /admin/reassign_partitions before submitting the next batch. Comparing the new_replicas Vs. original_replicas set, the algorithm can detect if there is more than 1 new replica in the new_replicas, if yes, then break it and put in different batch/bucket. There are other considerations of the reassignments in the same batch: e.g. for different topic/partition, try to spread the load and not to overwhelm a Leader. e.g. the Leadership bytes within the same batch for reassignments should be balanced across all brokers/leaders in the cluster as much as possible. Same for new follower (spread across the cluster not to overwhelm a particular follower). I think this (optimal executions of reassignment plans in batches) can be achieved outside of Kafka. was (Author: sql_consulting): Hi, [~viktorsomogyi], When I rebalance the whole cluster, I generate the reassignment plan json with a list of topic/partitions with its new_replicas/original_replicas, and sort them by their size, so try to group them in batches of similar sizes for execution, so that they are expected to complete reassignment using about the same amount of time. Say there are 1000 reassignments, and 50 per batch. That will be at least 20 batches/buckets to put in for execution. Comparing the new_replicas Vs. original_replicas set, the algorithm can detect if there is more than 1 new replica in the new_replicas, if yes, then break it and put in different batch/bucket. There are other considerations of the reassignments in the same batch: e.g. for different topic/partition, try to spread the load and not to overwhelm a Leader. e.g. the Leadership bytes within the same batch for reassignments should be balanced across all brokers/leaders in the cluster as much as possible. I think this (optimal executions of reassignment plans in batches) can only be achieved outside of Kafka. > Support for incremental replica reassignment > -------------------------------------------- > > Key: KAFKA-6794 > URL: https://issues.apache.org/jira/browse/KAFKA-6794 > Project: Kafka > Issue Type: Improvement > Reporter: Jason Gustafson > Assignee: Viktor Somogyi-Vass > Priority: Major > > Say you have a replication factor of 4 and you trigger a reassignment which > moves all replicas to new brokers. Now 8 replicas are fetching at the same > time which means you need to account for 8 times the current producer load > plus the catch-up replication. To make matters worse, the replicas won't all > become in-sync at the same time; in the worst case, you could have 7 replicas > in-sync while one is still catching up. Currently, the old replicas won't be > disabled until all new replicas are in-sync. This makes configuring the > throttle tricky since ISR traffic is not subject to it. > Rather than trying to bring all 4 new replicas online at the same time, a > friendlier approach would be to do it incrementally: bring one replica > online, bring it in-sync, then remove one of the old replicas. Repeat until > all replicas have been changed. This would reduce the impact of a > reassignment and make configuring the throttle easier at the cost of a slower > overall reassignment. -- This message was sent by Atlassian JIRA (v7.6.3#76005)