[
https://issues.apache.org/jira/browse/YARN-5889?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Eric Payne updated YARN-5889:
-----------------------------
Attachment: YARN-5889.0001.suggested.patchnotes
Thanks [~sunilg] for providing this work.
I am concerned about the way in which {{LeafQueue#getComputedUserLimit}} and
{{LeafQueue#getComputedActiveUserLimit}} are deciding when to recompute user
limits.
- In the case of {{getComputedActiveUserLimit}}, if number of active users
has increased or decreased, all active users in {{preComputedActiveUserLimit}}
are invalided, and not just the one that was activated/deactivated. This
requires recalculation for other users when it is not necessary.
- In the case of {{getComputedUserLimit}}, comparing a cached
{{lastUserCount}} with {{users.size()}} could miss occasions when recomputation
would be necessary.
My understanding of [~jlowe]'s [comment
(above)|https://issues.apache.org/jira/browse/YARN-5889?focusedCommentId=15745552&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15745552]
is that the {{LeafQueue}} would increment a change counter which would be
cached in each {{User}}'s object and then would be used to determine when
recalculation of user limit should occur.
Rather than document each occurrence, I have uploaded a patch with my
suggestions. {{YARN-5889.0001.suggested.patchnotes}} is the diff between
{{YARN-5889.0001.patch}} and my suggestions.
I haven't tested all of the scenarios yet (I'll complete that soon), but I
think this covers my concerns about the current patch, and I think it is more
efficient. Please have a look and let me know your thoughts.
> Improve user-limit calculation in capacity scheduler
> ----------------------------------------------------
>
> Key: YARN-5889
> URL: https://issues.apache.org/jira/browse/YARN-5889
> Project: Hadoop YARN
> Issue Type: Bug
> Components: capacity scheduler
> Reporter: Sunil G
> Assignee: Sunil G
> Attachments: YARN-5889.0001.patch,
> YARN-5889.0001.suggested.patchnotes, YARN-5889.v0.patch, YARN-5889.v1.patch,
> YARN-5889.v2.patch
>
>
> Currently user-limit is computed during every heartbeat allocation cycle with
> a write lock. To improve performance, this tickets is focussing on moving
> user-limit calculation out of heartbeat allocation flow.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
