KIP-693 — Client-side Circuit Breaker for Partition Write Errors

Discussion Producer

Introduces a client-side per-partition circuit breaker in the Kafka producer that stops routing new records to partitions with persistent write errors, redirecting traffic to healthy partitions. When disk failures cause a subset of partitions to have high write latency, the shared producer buffer fills up with blocked batches, eventually blocking writes to all partitions including healthy ones.

Details