Under-Replicated Partitions Signal Broker Degradation

critical

ReplicationUpdated Dec 20, 2025

When ISR count drops below the configured replication factor, partitions become under-replicated, indicating broker health issues, network problems, or disk saturation that compromise data durability.

Sources

Lambda timeout issue with confluent kafka esm | AWS re:Postrepost.aws

Monitoring | Apache Kafkakafka.apache.org

Technologies:

Apache KafkaSymptoms of this issue are visible in Apache Kafka metrics and logs

kafka.replication.under_replicated_partitions

kafka.server.ReplicaManager.OfflineReplicaCount

kafka.replication.isr_expands.rate

kafka.server.ReplicaManager.IsrExpandsPerSec

How to detect:

Monitor kafka.replication.under_replicated_partitions > 0 or kafka.server.ReplicaManager.OfflineReplicaCount > 0. Cross-reference with per-broker metrics to identify which brokers are falling out of ISR across multiple partitions.

Recommended action:

Check broker pod/instance health and resource utilization (CPU, memory, disk). Verify network connectivity between brokers. If one broker consistently falls behind, restart it or investigate disk I/O bottlenecks. Ensure min.insync.replicas configuration doesn't exceed available healthy replicas.