[
https://issues.apache.org/jira/browse/KYLIN-2857?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16160332#comment-16160332
]
liyang commented on KYLIN-2857:
-------------------------------
A few comments on the patch.
* How about calling {{Configuration.addResource(new Path(confValue))}} instead
of parsing XML by ourselves?
* The Configuration object returned from
{{HadoopUtil.getCurrentConfiguration()}} is a shared object and must not be
modified for one job.
* Will be most ideal if new jobs can launch with the {{getJobConf()}} returned
configuration, to make sure resume and launch share exactly the same config.
Requires refactoring of the {{MRUtil.runMRJob()}} however...
> MR configuration should be overwritten by user specified parameters when
> resuming MR jobs
> -----------------------------------------------------------------------------------------
>
> Key: KYLIN-2857
> URL: https://issues.apache.org/jira/browse/KYLIN-2857
> Project: Kylin
> Issue Type: Improvement
> Components: Job Engine
> Reporter: zhengdong
> Assignee: Dong Li
> Attachments:
> KYLIN-2857-MR-configuration-should-be-overwritten-by.patch
>
>
> For now, MR job conf could be overwritten by kylin_job_conf.xml file, and
> also further overwritten at project/cube level.
> Nonetheless, when resuming a MR job after a job server restarting or other
> kind of interruption, we only use the default conf to trace the job status.
> That's usually not a problem until user hope to overwrite job status tracing
> related parameters (job history server address e.g.) to support multi history
> servers.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
