You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Guozhang Wang (Jira)" <ji...@apache.org> on 2020/05/16 15:21:00 UTC

[jira] [Resolved] (KAFKA-7042) Fall back to the old behavior when the broker is too old to recognize LIST_OFFSET versions

     [ https://issues.apache.org/jira/browse/KAFKA-7042?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Guozhang Wang resolved KAFKA-7042.
----------------------------------
    Resolution: Not A Problem

> Fall back to the old behavior when the broker is too old to recognize LIST_OFFSET versions
> ------------------------------------------------------------------------------------------
>
>                 Key: KAFKA-7042
>                 URL: https://issues.apache.org/jira/browse/KAFKA-7042
>             Project: Kafka
>          Issue Type: Improvement
>          Components: consumer
>            Reporter: Guozhang Wang
>            Priority: Major
>
> When READ_COMMITTED is turned on (since 0.11.0), LIST_OFFSET requires min_version to be 2 on the consumer client side. On the other hand, if broker is no newer than 0.10.2 it can only recognize the version of LIST_OFFSET up to 1. In this case the consumer talking to such an old broker will throw an exception right away.
> What we can improve though, is that when the consumer realized broker does not recognize LIST_OFFSET of at least 2, it can fall back to the old behavior of READ_UNCOMMITTED since the data on that broker should not have any txn markers anyways. By doing this we would lift the hard restriction that consumers with READ_COMMITTED cannot work with an older version of broker (remember we are trying to achieve broker compatibility since 0.10.0).



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