[
https://issues.apache.org/jira/browse/TEZ-1827?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14355657#comment-14355657
]
Hitesh Shah commented on TEZ-1827:
----------------------------------
bq. Is it possible for MiniTezClusterWithTimeline to extends MiniTezCluster so
that we can reduce some code duplication ? But it is not related to this jira,
can be done in another jira.
The reason for this is an incompatibility in MiniYARNCluster. There was no
source compatible way to invoke the MiniYARNCluster in 2.2 and 2.4+ depending
on whether timeline needed to be enabled or not enabled.
bq. Would this create a race condition that allows the cluster to shutdown
because the test finished but before the AMs could unregister from the RM. This
would leave hung App master processes after the test.
Not sure I understand the concern. The current patch waits for a bit of time
before invoking a yarn app -kill. YARN will kill the apps by sending a signal
which will work in all situations where the process group can be killed
properly. This does raise an issue on Darwin though. However, I am not sure how
sending a kill and then waiting for the apps to complete will do anything.
Sending a kill to YARN implies that the app report will come back as killed.
The app itself may have its processes still running but there is no way to
verify it. Furthermore, there will always be hung AM processes if YARN itself
cannot kill the AM process properly.
Will fix the tez.test prefix property in next patch. Good catch [~zjffdu].
> MiniTezCluster takes 10 minutes to shut down
> --------------------------------------------
>
> Key: TEZ-1827
> URL: https://issues.apache.org/jira/browse/TEZ-1827
> Project: Apache Tez
> Issue Type: Bug
> Affects Versions: 0.5.2
> Reporter: Ashutosh Chauhan
> Assignee: Hitesh Shah
> Priority: Minor
> Attachments: TEZ-1827.1.patch
>
>
> Always reproducible
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
