You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Rohith Sharma K S (JIRA)" <ji...@apache.org> on 2016/04/04 07:51:25 UTC
[jira] [Commented] (YARN-4898) Avoid sending NODE_LABELS_UPDATE
event to scheduler when node label is not configured
[ https://issues.apache.org/jira/browse/YARN-4898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15223691#comment-15223691 ]
Rohith Sharma K S commented on YARN-4898:
-----------------------------------------
Couple of things observing from code is
# RMNodelableManager service is started even though node label is not configured. I am not sure why it was done.
# About the impact, not much major but we can optimize it so that AsyncDispatcher load can be reduced.
cc :/ [~leftnoteasy] [~Naganarasimha]
> Avoid sending NODE_LABELS_UPDATE event to scheduler when node label is not configured
> -------------------------------------------------------------------------------------
>
> Key: YARN-4898
> URL: https://issues.apache.org/jira/browse/YARN-4898
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Reporter: Rohith Sharma K S
>
> It is observed that when ever new NodeManager is registered or expired with ResourceManager, a new additional event NODE_LABELS_UPDATE has been triggered even though *Node Label is not enabled*. This makes dispatcher to process one event without any real use.
> logs : where all 100 of node-labels-update events processed by dispatcher.
> {noformat}
> 2016-03-30 15:42:24,461 DEBUG [AsyncDispatcher event handler] event.AsyncDispatcher (AsyncDispatcher.java:dispatch(175)) - Dispatching the event org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeLabelsUpdateSchedulerEvent.EventType: NODE_LABELS_UPDATE
> 2016-03-30 15:42:24,461 DEBUG [AsyncDispatcher event handler] event.AsyncDispatcher (AsyncDispatcher.java:dispatch(175)) - Dispatching the event org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeLabelsUpdateSchedulerEvent.EventType: NODE_LABELS_UPDATE
> 2016-03-30 15:42:24,462 DEBUG [AsyncDispatcher event handler] event.AsyncDispatcher (AsyncDispatcher.java:dispatch(175)) - Dispatching the event org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeLabelsUpdateSchedulerEvent.EventType: NODE_LABELS_UPDATE
> 2016-03-30 15:42:24,462 DEBUG [AsyncDispatcher event handler] event.AsyncDispatcher (AsyncDispatcher.java:dispatch(175)) - Dispatching the event org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeLabelsUpdateSchedulerEvent.EventType: NODE_LABELS_UPDATE
> {noformat}
> Point of concern is In large cluster, it is seen that registering thousands of nodes altogether to running cluster might/would cause other events processing to get delayed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)