[
https://issues.apache.org/jira/browse/YARN-6820?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16095298#comment-16095298
]
Jason Lowe commented on YARN-6820:
----------------------------------
A simple whitelist of users that are allowed to use the readers and thus access
all ATSv2 data would probably let us deploy ATSv2 in a limited way. We could
deploy ATSv2 alongside our existing ATSv1.5 and have the producers write to
_both_ ATSv1.5 and ATSv2. We would whitelist our admin users for ATSv2 but
still direct all our regular users to UIs based on ATSv1.5. So it would allow
us to deploy ATSv2, but only admins would see any benefits. Our normal users
cannot have unlimited access to the data within ATSv2, since one user could
access metadata (counters, configs, etc.) for another user's job. Therefore a
whitelist option isn't something we can live with if we want normal users to be
able to access ATSv2. Until there's a "real" ACL solution for ATSv2 we would
be forced to continue deploying ATSv1.5 so our users can still access the Tez
UI, YARN history server, etc.
> Restrict read access to timelineservice v2 data
> ------------------------------------------------
>
> Key: YARN-6820
> URL: https://issues.apache.org/jira/browse/YARN-6820
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Vrushali C
> Labels: yarn-5355-merge-blocker
>
> Need to provide a way to restrict read access in ATSv2. Not all users should
> be able to read all entities. On the flip side, some folks may not need any
> read restrictions, so we need to provide a way to disable this access
> restriction as well.
> Initially this access restriction could be done in a simple way via a
> whitelist of users allowed to read data. That set of users can read all data,
> no other user can read any data. Can be turned off for all users to read all
> data.
> Could be stored in a "domain" table in hbase perhaps. Or a configuration
> setting for the cluster. Or something else that's simple enough. ATSv1 has a
> concept of domain for isolating users for reading. Would be good to keep that
> in consideration.
> In ATSv1, domain offers a namespace for Timeline server allowing users to
> host multiple entities, isolating them from other users and applications. A
> “Domain” in ATSV1 primarily stores owner info, read and& write ACL
> information, created and modified time stamp information. Each Domain is
> identified by an ID which must be unique across all users in the YARN cluster.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
