[YARN-10841] Fix token reset synchronization for UAM response token

• Type: Sub-task
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.4.0
• Fix Version/s: 3.4.0
• Component/s: federation
• Labels: pull-request-available
• Target Version/s: 3.4.0
• Hadoop Flags: Reviewed
• Assignee: Minni Mittal
• Reporter: Minni Mittal
• Created: 05/Jul/21 13:50
• Updated: 12/Feb/24 06:10
• Resolved: 29/Jul/21 09:26

Description

2021-06-24T10:11:39,465 [ERROR] [AMRM Heartbeater thread] |impl.AMRMClientAsyncImpl|: Exception on heartbeat
org.apache.hadoop.yarn.exceptions.YarnException: org.apache.hadoop.yarn.exceptions.YarnRuntimeException: amrmToken from UAM cluster-0 should be null here
at org.apache.hadoop.yarn.server.nodemanager.amrmproxy.FederationInterceptor.allocate(FederationInterceptor.java:782)
 

 

2021-06-24T10:10:12,608 INFO  [616916] FederationInterceptor: Received new UAM amrmToken with keyId 843616604 

Hearbeatcallback sets token to null. But because of synchroniztion issue, it happened after mergeAllocate is called. So, while allocate merge is happening the value should get set to null and should have happened Inside lock

Issue Links

• links to: GitHub Pull Request #3194
• links to: https://github.com/apache/hadoop/pull/3194