[YUNIKORN-1658] autoscaling is not triggered for requests with unlimited resources

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.0.0, 1.1.0, 1.2.0
• Fix Version/s: 1.3.0
• Component/s: core - scheduler
• Labels: pull-request-available
• Target Version: 1.3.0
• Assignee: Wilfred Spiegelenburg
• Reporter: Wilfred Spiegelenburg
• Created: 28/Mar/23 03:51
• Updated: 20/Jun/23 14:41
• Resolved: 28/Mar/23 13:44

Description

Auto scaling is triggered by marking pods as unschedulable. The core triggers the checks to mark pods unschedulable. The checks is based on the maximum resource quotas for the queues and the requested resources.

The number of requests to mark outstanding can never exceed the resources amount available within the quota for the queues.

If a request has unlimited resource types in a limited queue scale up will not be triggered due to failures in the headroom calculation. The headroom calculation for the resource type that is not limited will become negative. This then breaks the "fit in" check as the unlimited resource is now defined as a negative value.

Issue Links

• is depended upon by: YUNIKORN-1632 Yunikorn fails to account for the max number of pods on a node
• relates to: YUNIKORN-1559 rethink QOS BestEffort pod scheduling
• links to: GitHub Pull Request #523