[YARN-4361] Total resource count mistake:NodeRemovedSchedulerEvent in ReconnectNodeTransition will reduce the newNode.getTotalCapability() in Multi-thread model

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.6.2
• Component/s: resourcemanager
• Labels: patch
• Target Version/s: 2.6.2
• Assignee: Unassigned
• Reporter: jialei weng
• Created: 17/Nov/15 04:14
• Updated: 03/Dec/15 04:23
• Resolved: 03/Dec/15 04:23

Description

Total resource count mistake:
NodeRemovedSchedulerEvent in ReconnectNodeTransition will reduce the newNode.getTotalCapability() in Multi-thread model. Since the RMNode and scheduler in different queue. So it cannot guarantee the remove-update-add operation in sequence. Sometimes the total resource will reduce the newNode.getTotalCapability() when handling NodeRemovedSchedulerEvent.

Issue Links

• is related to: YARN-2561 MR job client cannot reconnect to AM after NM restart.