KEP: add non-preempting option to PriorityClasses

keps/sig-scheduling/20190317-non-preempting-priorityclass

@@ -0,0 +1,104 @@
+---
+title: Add NonPreempting Option For PriorityClasses
+authors:
+  - "@vllry"
+owning-sig: sig-scheduling
+participating-sigs:
+  - sig-scheduling
+reviewers:
+  - "k82cn"
+  - "wgliang"
+approvers:
+  - "bsalamat"
+editor: Vallery Lancey
+creation-date: 2019-03-17
+last-updated: 2019-03-17
+status: implementable
+see-also:
+replaces:
+superseded-by:
+---
+
+# Allow PriorityClasses To Be Non-Preempting
+
+## Table of Contents
+
+* [Table of Contents](#table-of-contents)
+* [Summary](#summary)
+* [Motivation](#motivation)
+    * [Goals](#goals)
+    * [Non-Goals](#non-goals)
+* [Proposal](#proposal)
+    * [Risks and Mitigations](#risks-and-mitigations)
+* [Graduation Criteria](#graduation-criteria)
+* [Implementation History](#implementation-history)
+
+
+## Summary
+
+[PriorityClasses](https://kubernetes.io/docs/concepts/configuration/pod-priority-preemption/) are a beta feature,
+which impact the scheduling and eviction of pods.
+Pods will be scheduled according to descending priority.
+If a pod cannot be scheduled due to insufficient resources,
+lower-priority pods will be descheduled to make room.
+
+This proposal makes the pre-empting (descheduling) behavior optional,
+by adding a new field to PriorityClasses.
+If a PriorityClass does not have preemption enabled,
+the scheduler will not preempt pods in order to schedule a pod of that priority.
+
+## Motivation
+
+High-priority, non-preempting workloads are a common data science use case.
+Preempting batch workloads is a waste, as the work unit must be repeated.
+
+### Goals
+
+Add a boolean to PriorityClasses,
+to enable or disable preemption for pods of that PriorityClass.
+
+### Non-Goals
+
+## Proposal
+
+Add a NonPreempting field to PriorityClasses.
+This field will default to false,
+for backwards compatibility.
+
+If NonPreempting is false,
+the scheduler will preempt lower priority pods to schedule this pod,
+as is current behavior.
+
+If NonPreempting is true,
+a pod of that class will not preempt other pods if it cannot be scheduled.
+
+Update our documentation to reflect this new feature.
+
+### Risks and Mitigations
+
+The new feature may malfuction,
+or preemption may be accidentally impaired.
+New tests (covering both nonpreepting workloads and mixed workloads),
+and the existing preempting PriorityClass tests should be used to prove stability.
+
+## Graduation Criteria
+
+* Users are reporting that this resolves their workload priority use-cases
+(if not, additional enhancements would be tightly linked to this one).
+* The feature has been stable and reliable in at least 2 releases.
+* Adequate documentation exists for preemption and the optional field.
+* Test coverage includes non-preempting use cases.
+* Conformance requirements for non-preempting PriorityClasses are agreed upon.
+
+## Testing Plan
+Add unit and e2e tests for nonpreempting PriorityClasses to the existing scheduler tests.
+
+Ensure existing tests (for preempting PriorityClasses) do not break.
+
+## Implementation History
+
+[Original Github issue](https://github.com/kubernetes/kubernetes/issues/67671)
+
+Pod Priority and Preemption are tracked as part of [enhancement#564](https://github.com/kubernetes/enhancements/issues/564).
+The proposal for Pod Priority can be [found here](https://github.com/kubernetes/community/blob/master/contributors/design-proposals/scheduling/pod-priority-api.md)
+and Preemption proposal is [here](https://github.com/kubernetes/community/blob/master/contributors/design-proposals/scheduling/pod-preemption.md).