Eager vs Cooperative Rebalancing

Adding one consumer may require moving only a few partitions. Yet a rebalance can interrupt work on partitions that will stay with their current owners. The difference comes from how the group gives up ownership before establishing the next assignment.

Eager rebalancing releases all partitions first. Cooperative rebalancing lets members retain partitions that do not need to move and transfers the others in stages. That changes both the cost of a rebalance and what application callbacks must do.

We’ll compare the two approaches using a KRaft-based Kafka 4.3 cluster and the Apache Kafka Java consumer with subscribe(). The detailed walkthrough uses group.protocol=classic, where the chosen assignor determines eager or cooperative behavior. We’ll also distinguish this from the newer consumer group protocol.