You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Benedikt Linse (Jira)" <ji...@apache.org> on 2020/11/19 15:37:00 UTC

[jira] [Created] (KAFKA-10746) Consumer poll timeout Expiration should be logged as WARNING not INFO.

Benedikt Linse created KAFKA-10746:
--------------------------------------

             Summary: Consumer poll timeout Expiration should be logged as WARNING not INFO. 
                 Key: KAFKA-10746
                 URL: https://issues.apache.org/jira/browse/KAFKA-10746
             Project: Kafka
          Issue Type: Improvement
          Components: clients
    Affects Versions: 2.5.1, 2.6.0, 2.5.0
            Reporter: Benedikt Linse


When a consumer does not poll regularly, and the `max.poll.interval.ms` threshold is reached, the consumer leaves the consumer group, and the reason is logged as an INFO message:

[https://github.com/a0x8o/kafka/blob/e032a4ad9bac2392b4406d0a3b245d6011edd15b/clients/src/main/java/org/apache/kafka/clients/consumer/internals/AbstractCoordinator.java#L1356]

[https://github.com/a0x8o/kafka/blob/e032a4ad9bac2392b4406d0a3b245d6011edd15b/clients/src/main/java/org/apache/kafka/clients/consumer/internals/AbstractCoordinator.java#L1016]

Most Kafka users ignore INFO messages or have the log level set to WARN. Still many users run into this issue, since their applications take too long to process the polled records, and then the consumer fails to commit the offsets, which leads to duplicate message processing. Not seeing the error message in the first place means that users loose a lot of time debugging and searching for the reason for duplicate message processing.

Therefore it seems like the log level of this message should be increased to WARN. 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)