[YARN-1073] NM to recognise when it can't spawn process and stop accepting containers

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 2.1.0-beta
• Component/s: nodemanager

Environment

OS/X with not enough file handles

• Assignee: Unassigned
• Reporter: Steve Loughran
• Created: 16/Aug/13 20:18
• Updated: 30/Apr/14 14:28

Description

when creating too many containers with a claimed resource use of 0 RAM or vCores, the NM got to the state where exec() was continually failing -but nothing seemed to recognise this and blacklist the node.

Something should be noting that all container launches for an app/container are failing and do something. While AMs can/should code this, NM failure is something at the YARN-level

Issue Links

• relates to: YARN-2005 Blacklisting support for scheduling AMs