[ 
https://issues.apache.org/jira/browse/TEZ-1075?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14003697#comment-14003697
 ] 

Bikas Saha commented on TEZ-1075:
---------------------------------

First reason is technical consistency. A 0 vertex DAG is legitimate and its 
confusing to have it error out. Secondly, it should not waste cluster resources 
to run it. It does not look like that is happening from the patch.
Essentially, automated DAG generators use their own heuristics/optimizers etc. 
It may not be possible to guarantee that what started out initially as work 
will not end up being empty. Of course, they could run a post optimization step 
to see if there is any work left or they may submit the DAG directly. We dont 
control that piece.

> 0 vertex dags should succeed
> ----------------------------
>
>                 Key: TEZ-1075
>                 URL: https://issues.apache.org/jira/browse/TEZ-1075
>             Project: Apache Tez
>          Issue Type: Bug
>            Reporter: Bikas Saha
>            Assignee: Chen He
>              Labels: newbie++
>         Attachments: TEZ-1075-v2.patch, TEZ-1075.patch
>
>
> Currently 0 vertex dags fail because of the following code in DAGImpl
> {code}    if (numVertices == 0) {
>       addDiagnostic("No vertices for dag");
>       trySetTerminationCause(DAGTerminationCause.ZERO_VERTICES);
>       if (event != null) {
>         return DAGState.FAILED;
>       }
>       return finished(DAGState.FAILED);
>     }
> {code}
> IMO a 0 vertex DAG is a legitimate construct and should just pass. Similar to 
> a 0 task vertex. Its essentially a no-op operation.



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

Reply via email to