Yao Guangdong created MAPREDUCE-7365:
----------------------------------------
Summary: AppMaster register UAM can lost application priority in
subCluster
Key: MAPREDUCE-7365
URL: https://issues.apache.org/jira/browse/MAPREDUCE-7365
Project: Hadoop Map/Reduce
Issue Type: Bug
Components: applicationmaster, nodemanager, resourcemanager
Affects Versions: 3.3.0, 3.1.0, 3.0.0, 2.9.0
Reporter: Yao Guangdong
Fix For: 3.3.2
AppMaster register uam to subCluster can lost application priority in yarn
federation cluster.Which make the subCluster's RM allocate resouce to
application use default priority regardless of the application's real priority.
By analyzing the code, I found that the appMaster submitApplication to
subCluster did't set the priority to ApplicationSubmissionContext. As follows:
{code:java}
private void submitUnmanagedApp(ApplicationId appId)
throws YarnException, IOException {
SubmitApplicationRequest submitRequest =
this.recordFactory.newRecordInstance(SubmitApplicationRequest.class);
ApplicationSubmissionContext context = this.recordFactory
.newRecordInstance(ApplicationSubmissionContext.class);
context.setApplicationId(appId);
context.setApplicationName(APP_NAME + "-" + appNameSuffix);
if (StringUtils.isBlank(this.queueName)) {
context.setQueue(this.conf.get(DEFAULT_QUEUE_CONFIG,
YarnConfiguration.DEFAULT_QUEUE_NAME));
} else {
context.setQueue(this.queueName);
}
ContainerLaunchContext amContainer =
this.recordFactory.newRecordInstance(ContainerLaunchContext.class);
Resource resource = BuilderUtils.newResource(1024, 1);
context.setResource(resource);
context.setAMContainerSpec(amContainer);
submitRequest.setApplicationSubmissionContext(context);
context.setUnmanagedAM(true);
context.setKeepContainersAcrossApplicationAttempts(
this.keepContainersAcrossApplicationAttempts);
LOG.info("Submitting unmanaged application {}", appId);
this.rmClient.submitApplication(submitRequest);
}
{code}
Finnally, I fixed this by set application's priority to the
ApplicationSubmissionContext. The priority which from the homeCluster's
response when register appMaster to homeCluster's RM.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: mapreduce-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: mapreduce-issues-h...@hadoop.apache.org
