[
https://issues.apache.org/jira/browse/YARN-7450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16249816#comment-16249816
]
Ravi Prakash commented on YARN-7450:
------------------------------------
This is more complicated than I thought.
[UserGroupInformation.reloginFromKeyTab()|https://github.com/apache/hadoop/blob/975a57a6886e81e412bea35bf597beccc807a66f/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/security/UserGroupInformation.java#L1321]
*clears* away existing credentials, so a temporary Kerberos failure which may
have been resolved by trying to retry the login in some time, will be exhibited
immediately.
Hi [~daryn] ! Am I reading this right? Is it practical to fix this behavior?
> ATS Client should retry on intermittent Kerberos issues.
> --------------------------------------------------------
>
> Key: YARN-7450
> URL: https://issues.apache.org/jira/browse/YARN-7450
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: ATSv2
> Affects Versions: 2.7.3
> Environment: Hadoop-2.7.3
> Reporter: Ravi Prakash
>
> We saw a stack trace (posted in the first comment) in the ResourceManager
> logs for the TimelineClientImpl not being able to relogin from keytab.
> I'm guessing there was an intermittent network issue that failed the kerberos
> relogin from keytab. However, I'm assuming this was *not* retried because I
> only saw one instance of this stack trace. I propose that this operation
> should have been retried.
> It seems, this caused events at the ResourceManager to queue up and
> eventually stop responding to even basic {{yarn application -list}} commands.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
