Zhengbo Li created YARN-10831:
---------------------------------
Summary: Allow checking over-commit after reconnect event
Key: YARN-10831
URL: https://issues.apache.org/jira/browse/YARN-10831
Project: Hadoop YARN
Issue Type: Improvement
Reporter: Zhengbo Li
Currently the container over-commit check is skipped after a node re-connect
event, because the "timeout" period is always default to -1, which makes the
`signalContainersIfOvercommitted` method skip the check:
[line
link]([https://github.com/apache/hadoop/blob/03cfc852791c14fad39db4e5b14104a276c08e59/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/AbstractYarnScheduler.java#L1260)]
However in our case, because of the type of VM we use a node's resource could
change after a re-connect event. That means its CPU core / memory could be less
then causing container overcommit. Therefore we should allow configuring the
timeout period for reconnect event to be non -1 value to perform suck
overcommit check.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
