[YUNIKORN-1908] Support GPU in placeholders

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.4.0
• Component/s: shim - kubernetes
• Labels: pull-request-available
• Target Version: 1.4.0
• Assignee: Wilfred Spiegelenburg
• Reporter: Wilfred Spiegelenburg
• Created: 09/Aug/23 03:25
• Updated: 20/Nov/23 16:42
• Resolved: 16/Aug/23 19:15

Description

When we create a placeholder pod we read the list of resources defined in the task group and add that as a list of resource Requests to the new pod.

We do not check the content and just use the key as the resource type. This goes horribly wrong for taskgroups that contain extended resources like GPU requests.

We need to change the placeholder creation to add a limit for all extended resources to prevent this error message: 

2023-08-08T22:53:34.863Z        ERROR   cache/placeholder_manager.go:99 failed to create placeholder pod        {"error": "Pod \"tg-test-yunikorn-0\" is invalid: spec.containers[0].resources.limits: Required value: Limit must be set for non overcommitable resources"} 

Issue Links

• Blocked: YUNIKORN-2009 Placeholder pods should be created with limits
• is related to: YUNIKORN-1909 e2e test for taskgroups with limited resources
• links to: GitHub Pull Request #651