[
https://issues.apache.org/jira/browse/YARN-11957?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18079650#comment-18079650
]
ASF GitHub Bot commented on YARN-11957:
---------------------------------------
zhangxiping1 opened a new pull request, #8489:
URL: https://github.com/apache/hadoop/pull/8489
### Description of PR
In our scenario, the labels of a batch of nodes are updated periodically.
After the active RM had been running for a long time, an active-standby
switchover was triggered, and the switchover took an unusually long time. By
analyzing the logs and jstack information related to the standby RM's
transition to active, we found that a core thread,
"SchedulerEventDispatcher:Event Processor", had been continuously processing
updateNodeLabelsAndQueueResource. The service logs also showed a large number
of "REPLACE labels on nodes" messages.
> Frequent NM Label Updates Cause Longer RM Failover Time
> -------------------------------------------------------
>
> Key: YARN-11957
> URL: https://issues.apache.org/jira/browse/YARN-11957
> Project: Hadoop YARN
> Issue Type: Bug
> Affects Versions: 3.3.0
> Reporter: Xiping Zhang
> Assignee: Xiping Zhang
> Priority: Major
>
> In our scenario, the labels of a batch of nodes are updated periodically.
> After the active RM had been running for a long time, an active-standby
> switchover was triggered, and the switchover took an unusually long time. By
> analyzing the logs and jstack information related to the standby RM's
> transition to active, we found that a core thread,
> "SchedulerEventDispatcher:Event Processor", had been continuously processing
> updateNodeLabelsAndQueueResource. The service logs also showed a large number
> of "REPLACE labels on nodes" messages.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
