An under-replicated partition has fewer active replicas than its configured replication factor. It usually signals that a broker is unavailable or a replica has fallen out of sync, reducing redundancy until recovery completes. section 5.3 section 15.6
Use the partition and ISR state to distinguish replication loss from an immediate availability failure:
- Run kafka-topics.sh --describe --under-replicated-partitions to find affected partitions across topics. section 5.3
- Compare Replicas with Isr: a missing replica makes the partition under-replicated, while the remaining in-sync replicas may still support continued operation. section 5.3
- Check under-minimum-ISR status separately; an under-replicated partition can still retain enough ISRs to satisfy min.insync.replicas. section 5.3
- After the failed broker returns, allow its replica to synchronize before treating redundancy as restored; preferred-leader reassignment is a separate step. section 5.3
A broker failure in a three-replica topic
The products.prices.changelog.replication-3 topic uses replication factor 3 and min.insync.replicas=2. When Broker 3 stops, its partitions show Replicas containing three brokers but Isr containing only Brokers 1 and 2. The partitions are therefore under-replicated, yet two ISRs remain. After Broker 3 restarts and catches up, it rejoins the ISR; a preferred-leader election can then restore it as leader where required.
