[YUNIKORN-657] Expose reason of application failure to pods

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.11
• Component/s: shim - kubernetes
• Labels: pull-request-available
• Target Version: 0.11
• Assignee: Chaoran Yu
• Reporter: Chaoran Yu
• Created: 25/Apr/21 18:46
• Updated: 21/Jan/22 21:48
• Resolved: 10/May/21 04:49

Description

An application may fail for a number of reasons. For example,

• In gang scheduling, placeholders have expired before all of them can be successfully allocated
• When no placement rules are defined (i.e. static queues are used), an application is submitted to an non-existent queue
• The total amount of resources requested by a gang-scheduled app exceeds the capacity of the queue

YK's the finite state machine has Failed as a terminal state of an app, meaning that YK won't try to bring back a failed app ever again. The consequence is that pods of such failed apps will be stuck in pending indefinitely. A better behavior is for YK to mark those pods as failed too, while also passing the reason of the failure to those pods.

Issue Links

• causes: YUNIKORN-675 Pod status update could fail due to conflicts
• links to: GitHub Pull Request #258
• links to: GitHub Pull Request #270
• links to: GitHub Pull Request #271