hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Wangda Tan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-2918) Don't fail RM if queue's configured labels are not existed in cluster-node-labels
Date Mon, 10 Aug 2015 21:42:46 GMT

    [ https://issues.apache.org/jira/browse/YARN-2918?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680812#comment-14680812
] 

Wangda Tan commented on YARN-2918:
----------------------------------

[~dheeren],
Sorry for the inconvenience. The label where the node labels persisted is specified at:
{code}
yarn.node-labels.fs-store.root-dir
{code}

If you don't specify that, it will be /tmp/node-label by default, so I guess your problem
is node label dir is cleaned up by OS.

> Don't fail RM if queue's configured labels are not existed in cluster-node-labels
> ---------------------------------------------------------------------------------
>
>                 Key: YARN-2918
>                 URL: https://issues.apache.org/jira/browse/YARN-2918
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Rohith Sharma K S
>            Assignee: Wangda Tan
>              Labels: 2.6.1-candidate
>             Fix For: 2.8.0, 2.7.1
>
>         Attachments: YARN-2918.1.patch, YARN-2918.2.patch, YARN-2918.3.patch
>
>
> Currently, if admin setup labels on queues {{<queue-path>.accessible-node-labels
= ...}}. And the label is not added to RM, queue's initialization will fail and RM will fail
too:
> {noformat}
> 2014-12-03 20:11:50,126 FATAL org.apache.hadoop.yarn.server.resourcemanager.ResourceManager:
Error starting ResourceManager
> ...
> Caused by: java.io.IOException: NodeLabelManager doesn't include label = x, please check.
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerUtils.checkIfLabelInClusterNodeLabels(SchedulerUtils.java:287)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.AbstractCSQueue.<init>(AbstractCSQueue.java:109)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueue.<init>(LeafQueue.java:120)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.parseQueue(CapacityScheduler.java:567)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.parseQueue(CapacityScheduler.java:587)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.initializeQueues(CapacityScheduler.java:462)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.initScheduler(CapacityScheduler.java:294)
> 	at org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler.serviceInit(CapacityScheduler.java:324)
> 	at org.apache.hadoop.service.AbstractService.init(AbstractService.java:163)
> {noformat}
> This is not a good user experience, we should stop fail RM so that admin can configure
queue/labels in following steps:
> - Configure queue (with label)
> - Start RM
> - Add labels to RM
> - Submit applications
> Now admin has to:
> - Configure queue (without label)
> - Start RM
> - Add labels to RM
> - Refresh queue's config (with label)
> - Submit applications



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message