[jira] [Commented] (OOZIE-1863) Bundle submit should fail without submitting any coord if one of coord has issue

["Shwetha G S (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/OOZIE-1863?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14037066#comment-14037066 ] 

Shwetha G S commented on OOZIE-1863:
------------------------------------

Currently, bundle change API succeeds even if change of some coords fail(OOZIE-1807). There can be an option to enable atomic bundle change where bundle change API fails if some coord fail to change (all or nothing). This can be done only after commands start sharing the DB transaction(See OOZIE-1807 for more details)

> Bundle submit should fail without submitting any coord if one of coord has issue
> --------------------------------------------------------------------------------
>
>                 Key: OOZIE-1863
>                 URL: https://issues.apache.org/jira/browse/OOZIE-1863
>             Project: Oozie
>          Issue Type: Bug
>            Reporter: Purshotam Shah
>
> Currently, bundle submit command queues coord submit for all coords.
> StatusTransitService, which normally runs after 5 min check the status of each submitted coord and if any of them has failed, it kills running bundle.
> This approach has two issue
> a. Bundle is status is shown as "killed" ( few submitted coord is also killed). It will be difficult for user to find out the root cause.
> b. Few of the coord will be in running state for sometime and might corrupt or produce stale data.
> Bundle submit should dryrun coord before queuing submit command.
> We can further improve the bundle job submission logic.
> We can call CoordSubmitXCommand synchronously and if any coord submit fails ( bcz of DB or other evn issues), kill other submitted coord jobs.



--
This message was sent by Atlassian JIRA
(v6.2#6252)