[jira] [Commented] (REEF-539) Tasks cannot be restarted when there are multiple communication groups

["Markus Weimer (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/REEF-539?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14649707#comment-14649707 ] 

Markus Weimer commented on REEF-539:
------------------------------------

[~nachoacano] and I spoke offline about the issue to help me understand it:

When a Task fails, it gets routed to restart logic of the group communication layer. As it stands today, that logic assumes that all Tasks belong to the same, one communication group. This assumption is true in all of our current examples, but not intended: In fact, the API supports multiple (potentially overlapping) communication groups. In which case a Task can belong to zero or more of these groups.

The fix suggested in [#322|https://github.com/apache/incubator-reef/pull/322] addresses one of these issues, the case where there are multiple, non-overlapping communication groups.

> Tasks cannot be restarted when there are multiple communication groups
> ----------------------------------------------------------------------
>
>                 Key: REEF-539
>                 URL: https://issues.apache.org/jira/browse/REEF-539
>             Project: REEF
>          Issue Type: Bug
>            Reporter: Ignacio Cano
>            Assignee: Ignacio Cano
>            Priority: Minor
>
> Slave tasks cannot be restarted when there are multiple communication groups.
> A handler iterates over all CommunicationGroupDriverImpl and tries to fail the task. If that task didn't belong to that particular communication group, a NullPointerException is thrown.



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