Hi all, Just a gentle reminder about the discussion on KIP-1151: Minimal movement replica balancing algorithm for reassignment. [ https://cwiki.apache.org/confluence/display/KAFKA/KIP-1151%3A+Minimal+movement+replica+balancing+algorithm+for+reassignment ] So far, I have completed the implementation and submitted the PR. The PR also includes a comprehensive set of test cases to help verify the correctness of the algorithm.
I would really appreciate it if anyone could take some time to review the KIP and the PR and share your feedback or suggestions. Best regards Jialun Peng 在 2025年3月29日星期六，peng <p1070048...@gmail.com> 写道： > Hi Kafka Community, > > I would like to start a discussion on a new KIP proposal: KIP-1151: > Minimal movement replica balancing algorithm for reassignment > > Motivation > > Kafka clusters require rebalancing of topic replicas after horizontal > scaling to evenly distribute the load. However, the current approach does > not consider the existing replica distribution, often leading to > unnecessary movements. This KIP proposes an optimized strategy that > minimizes movements while achieving an even distribution. > > Proposal > > The proposal focuses on: > > 1.Minimal Movement: Minimize the number of replica relocations during > rebalancing. > 2.Replica Balancing: Ensure that replicas are evenly distributed across > brokers. > 3.Anti-Affinity Support: Support rack-aware allocation when enabled. > 4.Leader Balancing: Distribute leader replicas evenly across brokers. > 5.ISR Order Optimization: Optimize adjacency relationships to prevent > failover traffic concentration in case of broker failures. > > For a detailed description, please refer to the KIP document: > > KIP-1151: https://cwiki.apache.org/confluence/display/KAFKA/KIP-1151% > 3A+Minimal+movement+replica+balancing+algorithm+for+reassignment > <https://cwiki.apache.org/confluence/display/KAFKA/KIP-1150%3A+Minimal+movement+replica+balancing+algorithm+for+reassignment> > > > I would appreciate any feedback, comments, or suggestions from the > community. > > Best >
