KafkaUnderReplicatedPartitions

One or more partitions of a topic have replicas that are not keeping up with the leader.


Severity	warning
Source	danielqsj/kafka_exporter 1.7+
Key metrics	`kafka_topic_partition_under_replicated_partition`, `kafka_topic_partition_in_sync_replica`, `kafka_topic_partition_replicas`

What it means

A partition is under-replicated when its in-sync replica set (ISR) is smaller than its replica list. A follower drops out of the ISR when it has not caught up with the leader within replica.lag.time.max.ms. The alert fires when a topic has stayed under-replicated for several minutes, which rules out the short blips seen during a normal broker restart.

Data is still being served, but durability is reduced. If the ISR shrinks below min.insync.replicas, producers using acks=all get NotEnoughReplicas errors.

Common causes

A broker is down or restarting, so all its follower replicas are out of the ISR.
Slow or saturated disk on one broker, so its followers cannot fetch fast enough.
Network problems between brokers or throttled replication traffic.
Reassignment in progress, where new replicas are still copying data.
Traffic spike on a single large topic that followers cannot keep up with.

First checks

See which topics and how many partitions are affected:

sum by (topic) (kafka_topic_partition_under_replicated_partition) > 0

List the partitions and compare Replicas with Isr. The broker id missing from Isr again and again is the suspect:
```
kafka-topics.sh --bootstrap-server <broker>:9092 --describe --under-replicated-partitions
```
Confirm all brokers are up:
```
kafka_brokers
```

On the suspect broker, look at disk and replication errors:

kubectl -n <kafka-namespace> exec <broker-pod> -- df -h
kubectl -n <kafka-namespace> logs <broker-pod> --since=30m | grep -iE 'ReplicaFetcher|Shrinking ISR|error'

Check for a running reassignment:

kafka-reassign-partitions.sh --bootstrap-server <broker>:9092 --list

Fixing it

Restore the missing or slow broker first; followers rejoin the ISR on their own once they catch up. Fix disk saturation or move heavy partitions to less loaded brokers. If a reassignment is throttled too tightly, raise the replication throttle so it finishes. Avoid restarting more brokers until the count returns to zero.

KafkaBrokersDown: the most common root cause.
KafkaTopicPartitionNoLeader: what happens if the ISR shrinks to nothing.