You are viewing a plain text version of this content. The canonical link for it is here.

Posted to jira@kafka.apache.org by "hudeqi (Jira)" <ji...@apache.org> on 2023/05/09 13:50:00 UTC

[jira] [Updated] (KAFKA-14979) Incorrect lag was calculated when markPartitionsForTruncation in ReplicaAlterLogDirsThread

     [ https://issues.apache.org/jira/browse/KAFKA-14979?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

hudeqi updated KAFKA-14979:
---------------------------
    Description: 
When the partitions of ReplicaFetcherThread finished truncating, the ReplicaAlterLogDirsThread to which these partitions belong needs to be marked truncate. The lag value in the newState (PartitionFetchState) obtained in this process is still the original value (state.lag). If the truncationOffset is smaller than the original state.fetchOffset, then the original lag value is incorrect and needs to be updated. It should be the original lag value plus the difference between the original state.fetchOffset and truncationOffset.
If the original lag value is used incorrectly, then ReplicaAlterLogDirsThread may set the wrong lag value for fetcherLagStats when executing processFetchRequest. So it might be more reasonable to recalculate a lag value based on new state.fetchOffset.

> Incorrect lag was calculated when markPartitionsForTruncation in ReplicaAlterLogDirsThread
> ------------------------------------------------------------------------------------------
>
>                 Key: KAFKA-14979
>                 URL: https://issues.apache.org/jira/browse/KAFKA-14979
>             Project: Kafka
>          Issue Type: Improvement
>          Components: core
>    Affects Versions: 3.3.2
>            Reporter: hudeqi
>            Assignee: hudeqi
>            Priority: Major
>
> When the partitions of ReplicaFetcherThread finished truncating, the ReplicaAlterLogDirsThread to which these partitions belong needs to be marked truncate. The lag value in the newState (PartitionFetchState) obtained in this process is still the original value (state.lag). If the truncationOffset is smaller than the original state.fetchOffset, then the original lag value is incorrect and needs to be updated. It should be the original lag value plus the difference between the original state.fetchOffset and truncationOffset.
> If the original lag value is used incorrectly, then ReplicaAlterLogDirsThread may set the wrong lag value for fetcherLagStats when executing processFetchRequest. So it might be more reasonable to recalculate a lag value based on new state.fetchOffset.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)