[
https://issues.apache.org/jira/browse/FALCON-1835?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15153961#comment-15153961
]
Pallavi Rao commented on FALCON-1835:
-------------------------------------
Investigated moving to coord rerun and found the following issues/requirements:
1. Need to move to Oozie 4.2 (currently we use 4.1) as the coord rerun API has
been enhanced to support only failed node reruns.
2. The post-processing notification does not send the parent coord action id,
so retry uses workflow rerun. However, with oozie notifications, the
corresponding coord action id can be obtained and coord rerun can be performed.
So, will modify the code do coord rerun when action id is available and
workflow rerun when it is not.
3. Oozie coord rerun does not honor oozie.wf.rerun.skip.nodes property that
workflow rerun honors. This is a slight compatibility break as Falcon used to
support this option, although hardly used. Will file a bug for the same.
> Falcon should do coord rerun rather than workflow rerun to avoid parallel
> runs of same instance:
> ------------------------------------------------------------------------------------------------
>
> Key: FALCON-1835
> URL: https://issues.apache.org/jira/browse/FALCON-1835
> Project: Falcon
> Issue Type: Improvement
> Reporter: Pallavi Rao
> Assignee: Pallavi Rao
> Labels: backward-incompatible
>
> As of today, falcon does a workflow rerun, which can cause parallel instances
> to run.
> 1) Rerun via falcon retry.
> 2) Coord rerun via direct oozie command line.
> Falcon needs to move to coord rerun to rerun for same workflow which is
> already present.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
