[
https://issues.apache.org/jira/browse/YARN-6448?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15957606#comment-15957606
]
Yufei Gu commented on YARN-6448:
--------------------------------
According to my tests, any changes to {{unallocatedResource}} of a node may
break the method {{ClusterNodeTracker#sortedNodeList}}, so it's not enough to
add a write lock in it since {{unallocatedResource}} could be changed outside
of class {{ClusterNodeTracker}}. We should mark
{{ClusterNodeTracker#sortedNodeList}} not thread safe, and solve the issue in
continuous scheduling by adding a lock out side of {{List<FSSchedulerNode>
nodeIdList =  nodeTracker.sortedNodeList(nodeAvailableResourceComparator);}}
> Add back the lock in continuous scheduling while sorting nodes
> --------------------------------------------------------------
>
> Key: YARN-6448
> URL: https://issues.apache.org/jira/browse/YARN-6448
> Project: Hadoop YARN
> Issue Type: Bug
> Affects Versions: 2.9.0, 3.0.0-alpha2
> Reporter: Yufei Gu
> Assignee: Yufei Gu
>
> YARN-4719 remove the lock in continuous scheduling while sorting nodes. It
> breaks the order in comparison if nodes changes while sorting.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
