Retention limits how long log data remain available. A consumer that falls behind that boundary may lose the ability to replay from its old offset. Kafka normally deletes eligible closed log segments asynchronously, so the retention setting is not a promise to remove each individual record at an exact instant. The local cleaner simplifies that mechanism by filtering record timestamps when you request cleanup. The lesson uses that visible approximation to teach the replay boundary. Set the requested horizon and run the cleaner, then verify the old record is gone and the current record remains. Do not confuse a smaller retained log with a consumer that finished its work.
Inspect the topics, partition logs and group offsets before changing the cluster.