KIP-112 — Handle disk failure for JBOD

Accepted Kafka 1.0 Broker

Enables Kafka brokers to handle individual disk failures in a JBOD configuration by taking the failed disk's log directories offline and continuing to serve partitions on healthy disks, with controllers reassigning replicas away from the failed disk. Without this, a single disk failure on a multi-disk broker kills all partitions on that broker, requiring 4x storage overhead via RAID-10 to achieve the same fault tolerance at higher cost.

Protocol Impact

Produce · Fetch · Metadata · LeaderAndIsr · StopReplica · UpdateMetadata

Details