[YARN-5262] Optimize sending RMNodeFinishedContainersPulledByAMEvent for every AM heartbeat

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.8.0, 2.6.5, 2.7.4, 3.0.0-alpha1
• Component/s: resourcemanager
• Target Version/s: 2.9.0
• Hadoop Flags: Reviewed
• Assignee: Rohith Sharma K S
• Reporter: Rohith Sharma K S
• Created: 16/Jun/16 09:39
• Updated: 04/Aug/22 08:22
• Resolved: 29/Jun/16 04:45

Description

It is observed that RM triggers an one event for every ApplicationMaster#allocate request in the following trace. This is not necessarily required and it can be optimized such that send only if any containers are there to acknowledge to NodeManager.

	RMAppAttemptImpl.sendFinishedContainersToNM() line: 1871	
	RMAppAttemptImpl.pullJustFinishedContainers() line: 805	
	ApplicationMasterService.allocate(AllocateRequest) line: 567

Issue Links

• fixes: YARN-4741 RM is flooded with RMNodeFinishedContainersPulledByAMEvents in the async dispatcher event queue