[
https://issues.apache.org/jira/browse/TEZ-2759?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14724644#comment-14724644
]
Jeff Zhang edited comment on TEZ-2759 at 9/1/15 3:07 AM:
---------------------------------------------------------
I suspect some containers might be allocated but not launched yet. min-held
containers should be keep alive ?
Need to check the logs to confirm that.
was (Author: zjffdu):
I suspect some containers might be allocated but not launched yet. Need to
check the logs to confirm that.
> Containers not getting released if app is killed in the middle of execution
> ---------------------------------------------------------------------------
>
> Key: TEZ-2759
> URL: https://issues.apache.org/jira/browse/TEZ-2759
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Rajesh Balamohan
>
> Steps to reproduce (taking example of hive as it is easier to reproduce):
> 1. Launch tez session and run a hive query.
> 2. "ctrl+c" in the middle of query execution
> 3. This kills the query. However, the appmaster does not release all the
> containers it had. It randomly retains some containers and they are released
> only after AM timeouts (i.e after tez.session.am.dag.submit.timeout.secs)
> Impact is that, there can be unused containers and this can cause potential
> cluster underutilization.
> Will attach the logs soon.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
