apex-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (APEXCORE-439) After dynamic repartition application appears blocked
Date Sun, 24 Apr 2016 15:47:12 GMT

    [ https://issues.apache.org/jira/browse/APEXCORE-439?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15255627#comment-15255627
] 

ASF GitHub Bot commented on APEXCORE-439:
-----------------------------------------

Github user PramodSSImmaneni commented on a diff in the pull request:

    https://github.com/apache/incubator-apex-core/pull/314#discussion_r60846619
  
    --- Diff: engine/src/main/java/com/datatorrent/stram/plan/physical/PhysicalPlan.java ---
    @@ -979,6 +998,14 @@ public void deployChanges() {
         this.undeployOpers.removeAll(newOpers.keySet());
    --- End diff --
    
    @tweise The getDependents also includes the operators in the argument set, since they
are new we don't want to undeploy them. So it should be this
    
    Set<PTOperator> ndeps = getDependents(newOpers.keySet());
    ndeps.removeAll(newOpers.keySet());
    this.undeployOpers.addAll(ndeps);
    
    As you pointed out ndeps doesn't have to be added to deployOpers as it is already being
done in the next line.
    
    I had a line in there 
    ndeps.removeAll(this.undeployOpers);
    to ensure that operators that are already scheduled to be undeployed (removed completely)
don't get re-deployed because of this but on further thought this probably is not required
because if some operators are already going to be undeployed they will not be in the dependency
chain for newly deployable operators, is that assumption valid?


> After dynamic repartition application appears blocked
> -----------------------------------------------------
>
>                 Key: APEXCORE-439
>                 URL: https://issues.apache.org/jira/browse/APEXCORE-439
>             Project: Apache Apex Core
>          Issue Type: Bug
>    Affects Versions: 3.3.0
>            Reporter: Munagala V. Ramanath
>            Assignee: Munagala V. Ramanath
>
> When a file input operator (subclass of AbstractFileInputOperator) is dynamically repartitioned,
some downstream unifiers are not identified as needing to be redeployed causing the window
id to not advance and all downstream operators to be killed and restarted.



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

Mime
View raw message