You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@reef.apache.org by "Sergey Dudoladov (JIRA)" <ji...@apache.org> on 2016/08/03 22:23:20 UTC

[jira] [Resolved] (REEF-1503) Do not shut down driver on attempt to send close message to a non-running task

     [ https://issues.apache.org/jira/browse/REEF-1503?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Sergey Dudoladov resolved REEF-1503.
------------------------------------
    Resolution: Fixed

> Do not shut down driver on attempt to send close message to a non-running task
> ------------------------------------------------------------------------------
>
>                 Key: REEF-1503
>                 URL: https://issues.apache.org/jira/browse/REEF-1503
>             Project: REEF
>          Issue Type: Bug
>            Reporter: Mariia Mykhailova
>            Assignee: Mariia Mykhailova
>            Priority: Minor
>
> In IMRU work we've found a corner case which makes driver send close message to a task which has already failed. When several tasks fail in close succession, IMRU driver IFailedTask handler for first task acquires a lock and tries to shut down all tasks which it remembers as running in TaskManager. Second failed task arrives soon after that, but can't mark itself as failed in TaskManager because corresponding IFailedTask handler waits for the lock. So driver tries to close this task, throws exception and shuts down.
> I propose to modify code at {{org/apache/reef/runtime/common/driver/task/RunningTaskImpl.java}} so that {{close(final byte[] message)}} just logs a warning about attempt to close task which is not running (like {{close()}} method already does), instead of throwing exception which kills driver.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)