[
https://issues.apache.org/jira/browse/YARN-6667?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17597472#comment-17597472
]
ASF GitHub Bot commented on YARN-6667:
--------------------------------------
slfan1989 commented on code in PR #4810:
URL: https://github.com/apache/hadoop/pull/4810#discussion_r957943343
##########
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/amrmproxy/FederationInterceptor.java:
##########
@@ -1761,4 +1794,26 @@ public static <T> boolean isNullOrEmpty(Collection<T> c)
{
public static <T1, T2> boolean isNullOrEmpty(Map<T1, T2> c) {
return (c == null || c.size() == 0);
}
+
+ @VisibleForTesting
+ protected void cacheAllocatedContainersForSubClusterId(
+ List<Container> containers, SubClusterId subClusterId) {
+ cacheAllocatedContainers(containers, subClusterId);
+ }
+
+ @VisibleForTesting
+ protected Map<ContainerId, SubClusterId> getContainerIdToSubClusterIdMap() {
+ return containerIdToSubClusterIdMap;
+ }
+
+ private boolean isSCHealth(SubClusterId subClusterId) throws YarnException {
+ boolean isSCHealth = true;
+ Set<SubClusterId> timeOutScs = getTimedOutSCs(true);
+ SubClusterInfo subClusterInfo =
federationFacade.getSubCluster(subClusterId);
+ if (timeOutScs.contains(subClusterId) ||
+ subClusterInfo == null || subClusterInfo.getState().isUnusable()) {
+ isSCHealth = false;
Review Comment:
Thanks for your suggestion, I will modify the code.
> Handle containerId duplicate without failing the heartbeat in Federation
> Interceptor
> ------------------------------------------------------------------------------------
>
> Key: YARN-6667
> URL: https://issues.apache.org/jira/browse/YARN-6667
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Botong Huang
> Assignee: fanshilun
> Priority: Minor
> Labels: pull-request-available
>
> From the actual situation, the probability of this happening is very low.
> It can only be caused by the master-slave fail-hover of YARN and the wrong
> Epoch parameter configuration.
> We will try to be compatible with this situation and let the Application run
> as much as possible, using the following measures:
> 1. Select a node whose heartbeat does not time out for allocation, and at the
> same time require the node to be in the RUNNING state.
> 2. If the heartbeat of both RMs does not time out, and both are in the
> RUNNING state, select the previously allocated RM for Container processing.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
