[
https://issues.apache.org/jira/browse/TEZ-1620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14152727#comment-14152727
]
Jeff Zhang commented on TEZ-1620:
---------------------------------
Attach the new patch.
bq. How much the test runtime increase after this change?
No significant changes. Take TestTezJobs as example, it save around 5 seconds.
Because even we don't have sleeping 10 seconds, still have to wait for several
seconds until apps finished.
bq. Were we able to verify that after multiple test runs, we dont have any
orphaned DAGAppMaster processes.
run TestTezJobs 10 times, no orphaned DAGAppMaster processes
bq. waitForAppsFinish() -> waitForAppsToFinish() ?
Fix it
bq. We should be using YARNClient instead of directly accessing
getResourceManager().getClientService(). Theortetically, thats YARN internal
code and can be changed anytime.
Fix it.
bq. Is Collections2 available within our guava version?
Yes
bq. Can you please check if this builds with Hadoop 2.2.
It can builds with hadoop 2.2 except tez-yarn-timeline-history which is not
caused by this jira. (hadoop 2.2 don't have yarn-timeline api)
> Wait for application finish before stopping MiniTezCluster
> ----------------------------------------------------------
>
> Key: TEZ-1620
> URL: https://issues.apache.org/jira/browse/TEZ-1620
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Jeff Zhang
> Assignee: Jeff Zhang
> Attachments: Tez-1620-2.patch, Tez-1620.patch
>
>
> Currently, we sleep 10 seconds to wait for DAGAppMaster to finish, otherwise
> DAGAppMaster will hang there for connecting RM to unregister.
> We should wait for all the applications finish before stopping
> MiniTezCluster.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
