[
https://issues.apache.org/jira/browse/TEZ-3166?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15197741#comment-15197741
]
Siddharth Seth commented on TEZ-3166:
-------------------------------------
[~aplusplus] - there's a couple more cases which need to be handled.
In TaskRunner2Callable - it's possible that the task ends before the close.
Look for cases where TaskRunner2Result is being returned. The counters should
be updated in these cases as well.
This will mostly occur when the AM asks the task to die - preempted for
instance. That needs to be handled as well. TezTaskRunner2: case
CONTAINER_STOP_REQUESTED:. I don't think we should try fixing this in this
jira, since there's a bunch of AM side changes involved (The AM has essentially
stopped tracking this task at this point, and will not accept another heartbeat
request).
The cases where the task fails itself will be covered by this jira.
For this one, could you please update TaskRunner2Result with the counters being
updated for the additional cases, and file a new jira to handle external events
to the task.
> Counters aren't fully updated and sent for failed tasks
> -------------------------------------------------------
>
> Key: TEZ-3166
> URL: https://issues.apache.org/jira/browse/TEZ-3166
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Zhiyuan Yang
> Assignee: Zhiyuan Yang
> Attachments: TEZ-3166.1.patch, TEZ-3166.2.patch
>
>
> On master branch, if an application fails very soon, counters aren't fully
> updated so on tez-ui we can only see part of counters. We should publish all
> counters even task attempt fails.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
