[YUNIKORN-1395] Account for preempted placeholder in the placeholder data

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.1.0
• Fix Version/s: 1.2.0
• Component/s: core - scheduler
• Labels: pull-request-available
• Target Version: 1.2.0
• Assignee: Wilfred Spiegelenburg
• Reporter: Wilfred Spiegelenburg
• Created: 10/Nov/22 22:10
• Updated: 27/Feb/23 04:35
• Resolved: 10/Nov/22 22:30

Description

If we pre-empt a placeholder to make room for a daemon set pod we do not update the counter inside the application placeholder data object.

After the pre-emption it shows as a placeholder that should be there but is no longer on the system.

When we replace or timeout a placeholder we do account for it correctly.

We directly call app.RemoveAllocation(uuid) from the partition code. We do not take into account the account the termination type there which we should to make sure we account correctly for the preemption in the data.

Issue Links

• is caused by: YUNIKORN-1085 DaemonSet pods may fail to be scheduled on new nodes added during autoscaling
• split to: YUNIKORN-1605 Add unit tests for preempted placeholders
• links to: GitHub Pull Request #447