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

Posted to commits@samza.apache.org by "Jake Maes (JIRA)" <ji...@apache.org> on 2017/10/18 01:13:00 UTC

[jira] [Updated] (SAMZA-1465) Performance regression for KafkaCheckpointManager

     [ https://issues.apache.org/jira/browse/SAMZA-1465?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jake Maes updated SAMZA-1465:
-----------------------------
    Summary: Performance regression for KafkaCheckpointManager  (was: Restore old 'startingOffset' behavior for KafkaCheckpointManager)

> Performance regression for KafkaCheckpointManager
> -------------------------------------------------
>
>                 Key: SAMZA-1465
>                 URL: https://issues.apache.org/jira/browse/SAMZA-1465
>             Project: Samza
>          Issue Type: Bug
>            Reporter: Jake Maes
>            Assignee: Jake Maes
>
> SAMZA-1427 introduced a performance regression with KafkaCheckpointManager. 
> We invoke the checkpoint manager once for each task and the old code would only read the delta for each subsequent call. The SAMZA-1427 code would reread the entire checkpoint topic for each call. 
> This effect is even worse when log compaction is not running on the broker. 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)