Repartition errors in Hyades pods

[cartermitchellLM]

We are troubleshooting the database connectivity of our hyades k8s cluster, and in the logs of our Repo Meta Analyzer and Vulnerability Analyzer we are seeing the following error that seems to be related to our redpanda (kafka) instance.

2026-03-05 18:56:47,394 ERROR [org.apa.kaf.cli.adm.int.DeleteRecordsHandler] (kafka-admin-client-thread | hyades-repository-meta-analyzer-3131bda9-a1f8-4703-a9f5-5d1cc6a01660-admin) [AdminClient clientId=hyades-repository-meta-analyzer-3131bda9-a1f8-4703-a9f5-5d1cc6a01660-admin] DeleteRecords request for topic partition hyades-repository-meta-analyzer-command-by-purl-coordinates-repartition-1 failed due to an unexpected error OFFSET_OUT_OF_RANGE

The error repeats in groups of 3, referring to repartitions 0 through 2. What are some ways to clear out the message queue or get redpanda to play nicely again with the other services?

[nscuro]

So this happens because Kafka Streams tries to clean up after itself, and purge records from the repartition topic once they're no longer needed. It seems that either Redpanda doesn't properly support the deletion of records, or your instance lacks permissions to delete records (should not be the case unless you explicitly configured ACLs in Redpanda).

Functionality-wise, this is not a big concern. The repartition topics should have limited retention so the broker will eventually remove old records anyway. To get rid of the noise, you have at least two options:

• Silence this particular error log by configured the env var: QUARKUS_LOG_CATEGORY__ORG_APACHE_KAFKA_CLIENTS_ADMIN_INTERNALS_DELETERECORDSHANDLER=OFF
• Increase the interval at which record deletion will be attempted to a large value (e.g. max int): KAFKA_STREAMS_REPARTITION_PURGE_INTERVAL_MS=2147483647

[cartermitchellLM]

Where do you add these configuration settings?