[
https://issues.apache.org/jira/browse/YARN-998?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15205193#comment-15205193
]
Jian He commented on YARN-998:
------------------------------
looks good to me overall,
- should ResourceTrackerService#drConf be volatile ?
- also, we may add a log in the resourceTrackerService where the resource is
override to assist debugging.
{code}
if (nodes != null && Arrays.asList(nodes).contains(nid)) {
capability.setMemory(this.drConf.getMemoryPerNode(nid));
capability.setVirtualCores(this.drConf.getVcoresPerNode(nid));
}
{code}
> Persistent resource change during NM/RM restart
> -----------------------------------------------
>
> Key: YARN-998
> URL: https://issues.apache.org/jira/browse/YARN-998
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: graceful, nodemanager, scheduler
> Reporter: Junping Du
> Assignee: Junping Du
> Attachments: YARN-998-sample.patch, YARN-998-v1.patch,
> YARN-998-v2.1.patch, YARN-998-v2.patch
>
>
> When NM is restarted by plan or from a failure, previous dynamic resource
> setting should be kept for consistency.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
