KEP-5729: DRA: ResourceClaim Support for Workloads

[nojnhuh]

• One-line PR description: Add initial draft for KEP-5729: DRA: ResourceClaim Support for Workloads

• Issue link: DRA: ResourceClaim Support for Workloads #5729

• Other comments:

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: nojnhuh
Once this PR has been reviewed and has the lgtm label, please assign sanposhiho for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• keps/sig-scheduling/OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[nojnhuh]

I'm ready for feedback on the Summary and Motivation sections. Mostly want to make sure this is appropriately scoped w.r.t. #5732.

/cc @wojtek-t @erictune @44past4 @helayoty @johnbelamaric

I'm still working on the meat of the proposal and hope to have a draft of that ready to share by the middle of next week before I go on vacation through the end of the year.

[nojnhuh]

I've added the first draft of the API. Still working through the rest of the KEP, but I think that section is ready to start getting some feedback.

/label api-review

[k8s-ci-robot]

@nojnhuh: The following tests failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
pull-enhancements-test	c09de1f	link	true	/test pull-enhancements-test
pull-enhancements-verify	c09de1f	link	true	/test pull-enhancements-verify

Full PR test history. Your PR dashboard. Please help us cut down on flakes by linking to an open issue when you hit one in your PR.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here.

[44past4]

An important problem to solve is defining the exact moment in time when deallocation of PodGroups' ResourceClaims will occur. This will be challenging because PodGroups do not have a well define lifecycle. Should this be put as a separate goal here?

[44past4]

An alternative/complementary solution could include an option to define an optional ResourceClaim constraints (similar to ResourceClaim.spec.devices.constraints) on the level of PodGroup which would need to apply to all ResourceClaims used by PodGroup's Pods which come from a given ResourceClaimTemplate.

This way each pod could have its own ResourceClaim created from a given template but when scheduling those Pods as part of PodGroup kube-scheduler would make sure that an additional PodGroup level ResourceClaim constraint is fulfilled for all devices allocated to all Pods which are part of this PodGroup. This would provide for instance an easy way to specify that all GPUs allocated to Pods in a given PodGroup needs to come from a single block (they need to have the same value for a given attribute of the Device - like gpu/block-name in the example below):

apiVersion: scheduling.k8s.io/v1alpha1
kind: Workload
metadata:
  name: my-workload
  namespace: default
spec:
  podGroups:
  - name: group-1
    policy:
      basic: {}
    resourceClaims:
    - name: wl-claim
      resourceClaimTemplateName: gpu-claim-template
      constraints:
      - matchAttribute: gpu/block-name
---
apiVersion: resource.k8s.io/v1
kind: ResourceClaimTemplate
metadata:
  name: gpu-claim-template
  namespace: default
spec:
  spec:
    devices:
      requests:
      - name: my-device
        exactly:
          deviceClassName: gpu
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: gpu-claim-example-1
  namespace: default
spec:
  replicas: 2
  selector:
    matchLabels:
      app: wl-claim-example-1
  template:
    metadata:
      labels:
        app: wl-claim-example-1
    spec:
      containers:
      - name: pause
        image: "registry.k8s.io/pause:3.6"
        resources:
          claims:
          - name: resource-1
      resourceClaims:
      - name: resource-1
        resourceClaimTemplateName: gpu-claim
      workloadRef:
        name: my-workload
        podGroup: group-1
        podGroupReplicaKey: "1"

[44past4]

This alternative/complementary solution is discussed in https://docs.google.com/document/d/11rC_qDtArIOx_ZQfM-G4H5qIYFm0rX_GUz1yPGPrv3k/edit?usp=sharing document.