[
https://issues.apache.org/jira/browse/YARN-5788?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15620535#comment-15620535
]
Varun Saxena edited comment on YARN-5788 at 10/30/16 8:52 PM:
--------------------------------------------------------------
Thanks [~bibinchundatt] for the patch.
Although updateClusterResource would handle just about everything and easiest
way to handle this but as cluster resource will remain the same on replace
labels on node, do we need to carry out each step in updateClusterResource ?
For instance, queue usage related ratios and capacities for instance need to be
updated only for 2 sets of relevant partitions for each node i.e. the old and
new/replaced one. This can be useful to bypass queues where both relevant
partitions are not applicable.
We should activate apps as this some apps can be activated with replace label
on node. This is a heavy operation and we will hold a write lock on scheduler
for this but this should be fine as replace labels on node should not be a
frequent operation.
cc [~leftnoteasy]. Your thoughts on this ?
was (Author: varun_saxena):
Thanks [~bibinchundatt] for the patch.
Although updateClusterResource would handle just about everything and easiest
way to handle this but as cluster resource will remain the same on replace
labels on node, do we need to carry out each step in updateClusterResource ?
For instance, queue usage related ratios and capacities for instance need to be
updated only for 2 relevant partitions i.e. the old and new/replaced one. This
can be useful to bypass queues where both relevant partitions are not
applicable. However, in node label update event sent to scheduler, we do not
send the old node to label mapping. Should we ?
We should activate apps as this some apps can be activated with replace label
on node. This is a heavy operation and we will hold a write lock on scheduler
for this but this should be fine as replace labels on node should not be a
frequent operation.
cc [~leftnoteasy]. Your thoughts on this ?
> AM limit resource in UI and REST not updated after -replaceLabelsOnNode
> -----------------------------------------------------------------------
>
> Key: YARN-5788
> URL: https://issues.apache.org/jira/browse/YARN-5788
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Bibin A Chundatt
> Assignee: Bibin A Chundatt
> Priority: Minor
> Attachments: Screenshot from 2016-10-27 22-31-46.png,
> YARN-5788.0001.patch, YARN-5788.0002.patch
>
>
> Steps to reproduce
> ==============
> # Enable node labels
> # Configure capacity scheduler xml with label capacity configuration
> # Add labelx to cluster
> # Replace Node labels on node1
> # Check scheduler and Rest API
> *Actual*
> Am limit in scheduler UI is still based on old resource
> *Expected*
> AM limit to be updated based new partition resource
--
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
