KafkaConsumerGroupInactive

A consumer group still has committed offsets and a backlog, but nobody is consuming for it.


Severity	warning
Source	danielqsj/kafka_exporter 1.7+
Key metrics	`kafka_consumergroup_members`, `kafka_consumergroup_lag`

What it means

Kafka keeps a group’s committed offsets after all its members leave (until offsets.retention.minutes passes). kafka_exporter reports the member count and the lag for that group. The alert fires when a group has had zero members and non-zero lag for a while: messages are arriving for a consumer that no longer exists or is not running.

Either an application is down and its work is piling up, or a group was retired and its offsets are just leftovers.

Common causes

Consumer deployment scaled to zero or stuck: CrashLoopBackOff, ImagePullBackOff, or failed scheduling.
Authentication or ACL failure: the app runs but cannot join the group after a credential rotation.
Group id changed in a new release, leaving the old group orphaned with lag.
Batch or cron consumers that run only periodically.
Decommissioned service whose group was never deleted.

First checks

Confirm the group is empty:

kafka-consumer-groups.sh --bootstrap-server <broker>:9092 --describe --group <group> --state
kafka-consumer-groups.sh --bootstrap-server <broker>:9092 --describe --group <group>

State Empty and no CONSUMER-ID confirm it.

Find the owning application and check it is running:

kubectl get deploy,statefulset -A | grep -i <app>
kubectl -n <namespace> get pods -l <consumer-selector>

Check its logs for join or auth errors:

kubectl -n <namespace> logs deploy/<consumer> --since=1h | grep -iE 'authoriz|authenticat|group|coordinator'

Check whether a new group id replaced it:

sum by (consumergroup) (kafka_consumergroup_members{consumergroup=~"<app-prefix>.*"})

Fixing it

Restore the consumer if it should be running; it resumes from the committed offsets. If the group id changed intentionally, decide whether the new group needs the old offsets and reset it with --reset-offsets before starting. For retired groups, delete them so the alert clears:

kafka-consumer-groups.sh --bootstrap-server <broker>:9092 --delete --group <group>

For periodic batch consumers, exclude the group from this alert.

KafkaConsumerGroupLagHigh: backlog size for the same group.
KafkaConsumerGroupLagGrowing: lag rising while consumers exist but are too slow.