[ 
https://issues.apache.org/jira/browse/TEZ-2410?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14529778#comment-14529778
 ] 

Jeff Zhang commented on TEZ-2410:
---------------------------------

bq. The new VertexGroupStatus enum is duplicating the logic that is being 
encoded in successfulCommits because all the enums can be represented as 
methods on VertexGroupInfo which use a combination of "committed" and 
"successfulCommits". So it may be better to use new methods on VertexGroupInfo 
instead of having 2 ways of representing the same thing which may go out of 
sync as code evolves.
The original purpose of VertexGroupStatus is to track the status of commits in 
VertexGroup (successfulCommits=0 can mean both COMMIT_NOT_STARTED or 
COMMITTING), and the status will be used in DAGImpl#vertexReRunning. 
But after TEZ-2413, this VertexGroupStatus can be removed. Maybe I can include 
TEZ-2413 in this patch. [~bikassaha] Thoughts ?

bq. 1) For normal vertices we do not write a group commit finished event. (the 
patch does not seem to cover any of these)  
Do you mean to add statement like this ?
{code}
    historyEventHandler.verifyVertexGroupCommitStartedEvent("v1", 0);           
   
    historyEventHandler.verifyVertexGroupCommitFinishedEvent("v1", 0);       // 
use vertex name to verify vertex does not cause VertexGroupCommitFinished
{code}

bq. 2) For groups we only write the group commit finished event once. (the 
patch covers some of these)
3 cases are covered in TestCommit (any other case I missed ?)
* If the commit is onDAGSuccess, then VertexGroupFinished should not be logged
* If the commit is onVertexSuccess,  then VertexGroupFinished should be logged 
once
* If the commit is onVertexSuccess and vertex group has multiple outputs, then 
VertexGroupCommitFinished should still be logged once

> VertexGroupCommitFinishedEvent is not logged correctly
> ------------------------------------------------------
>
>                 Key: TEZ-2410
>                 URL: https://issues.apache.org/jira/browse/TEZ-2410
>             Project: Apache Tez
>          Issue Type: Bug
>    Affects Versions: 0.7.0
>            Reporter: Jeff Zhang
>            Assignee: Jeff Zhang
>            Priority: Blocker
>             Fix For: 0.7.0
>
>         Attachments: TEZ-2410-1.patch
>
>
> VertexGroupCommitFinishedEvent may be logged for non-vertex group commits.
> VertexGroupCommitFinishedEvent may be logged for each member vertex of the 
> group instead of once per group.



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

Reply via email to