Partition rebalancing migrates and reassigns Kafka partitions and replicas across brokers to redistribute workload or remove capacity. Adding a broker does not automatically balance existing data; partitions must be moved, and leadership updated, across the cluster. section 9.2.1
Treat rebalancing as an operational change with measurable costs:
- Moving partitions consumes source-broker CPU and disk I/O, uses network bandwidth, and can cause latency spikes. section 9.2.1
- Removing a broker requires migrating its partitions first; abrupt loss leaves the cluster under-replicated and hot while replicas rebuild. section 9.2.1
- Managed services may automate parts of rebalancing, while other providers require manual broker addition and data redistribution. section 9.2.1
Adding capacity
When a Kafka cluster gains a broker, capacity does not spread automatically. Operators migrate partitions to the new broker, update leadership, and account for transfer time, network use, and possible latency increases. section 9.2.1
