[
https://issues.apache.org/jira/browse/YARN-6130?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16098138#comment-16098138
]
Rohith Sharma K S commented on YARN-6130:
-----------------------------------------
bq. The intention here is to avoid updating the token in AM UGI on every
allocate response. We can potentially cache the RMID and version to ensure that
the version of token coming from RM is same as the one already updated by AM in
its UGI. Thoughts?
Even without rmIdentifies, if token is updated with same rm_identifiers then
AM has to update it right? Am I missing any particular scenario?
[~jianhe]
bq. what is the existing AppCollectorData#rmIdentifier used for ?
this is used to handle race condition between 2 NM sending collector address to
RM. Lets say, because of split brain one NM is out of sync and application is
relaunched in different NodeManager. After NM is reconnected from split brain,
both the NMs will keep sending collector data to RM and updates wrong collector
address in RM which intern AM will update wrong collector address.
> [ATSv2 Security] Generate a delegation token for AM when app collector is
> created and pass it to AM via NM and RM
> -----------------------------------------------------------------------------------------------------------------
>
> Key: YARN-6130
> URL: https://issues.apache.org/jira/browse/YARN-6130
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Varun Saxena
> Assignee: Varun Saxena
> Labels: yarn-5355-merge-blocker
> Attachments: YARN-6130-YARN-5355.01.patch,
> YARN-6130-YARN-5355.02.patch, YARN-6130-YARN-5355.03.patch,
> YARN-6130-YARN-5355.04.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
