[YARN-10184] NPE happens in NMClient when reinitializeContainer

• Type: Bug
• Status: Patch Available
• Priority: Major
• Resolution: Unresolved
• Assignee: kyungwan nam
• Reporter: kyungwan nam
• Created: 09/Mar/20 09:03
• Updated: 18/Mar/20 13:54

Description

NPE happens in NMClient when upgrading a yarn-service app which AM has been restarted.
Here is AM’s log at the time of the NPE.

2020-02-20 16:43:35,962 [Container  Event Dispatcher] ERROR yarn.YarnUncaughtExceptionHandler - Thread Thread[Container  Event Dispatcher,5,main] threw an Exception.
java.lang.NullPointerException
	at org.apache.hadoop.yarn.client.api.async.impl.NMClientAsyncImpl$1.run(NMClientAsyncImpl.java:172)
2020-02-20 16:43:36,398 [AMRM Callback Handler Thread] WARN  service.ServiceScheduler - Container container_e58_1581930783345_1954_01_000006 Completed. No component instance exists. exitStatus=-100. diagnostics=Container released by application 

NMClient keeps containers since the container has been started.
But, when restarting AM, NMClient is initialized and previous containers are lost.
Since then, NPE will happen when reinitializeContainer is requested.