DRA: ResourceClaim Support for Workloads

[nojnhuh]

Enhancement Description

DRA ResourceClaims and ResourceClaimTemplates can currently be associated only with Pods. This limits how resource requirements can be expressed for higher-level groups of Pods organized into Workloads (#4671) and PodGroups (#5832).

This enhancement will add fields to the Workload and PodGroup APIs that define references to ResourceClaims and ResourceClaimTemplates with similar semantics to the existing Pod fields. A ResourceClaimTemplate referenced by a claim in Workload or PodGroup will result in a new ResourceClaim generated from that template for that claim. A ResourceClaim (literal or generated from a ResourceClaimTemplate) will be reserved for the Workload, PodGroup, or Pod which references it, addressing the limit of 256 entries in ResourceClaim's status.reservedFor to enable sharing one claim among effectively any number of Pods under the same Workload.

• One-line enhancement description (can be used as a release note): Workload and PodGroup objects can define claims for DRA resources which can be claimed by their component Pods.
• Kubernetes Enhancement Proposal: https://github.com/kubernetes/enhancements/blob/master/keps/sig-scheduling/5729-resourceclaim-support-for-workloads/README.md
• Discussion Link:
• PRs by stage and milestone:
  • Alpha - v1.36
    • KEP (k/enhancements) update PR(s):
      • KEP-5729: DRA: ResourceClaim Support for Workloads #5736
    • Code (k/k) update PR(s):
      • KEP-5729: DRA: ResourceClaim Support for Workloads kubernetes#136989
    • Docs (k/website) update PR(s):
      • Docs update for KEP-5729: DRA: ResourceClaim Support for Workloads website#54596
  • Beta - v1.37
    • KEP (k/enhancements) update PR(s):
      • KEP-5729: Update ResourceClaim Support for Workloads for beta in 1.37 #5976
    • Code (k/k) update PR(s):
      • DRA: Fix shared in-flight allocation accounting in error paths kubernetes#139418
      • Address DRA controller feedback from DRAWorkloadResourceClaims implementation kubernetes#138363
      • DRA: scheduler_perf for DRAWorkloadResourceClaims kubernetes#138775
      • DRA: align ResourceClaim unreservation/deallocation for PodGroups with Pods kubernetes#139864
      • DRA: check for allocated claim before marking in-flight kubernetes#140269
      • resourceclaim controller: Add owner kind labels to create metric kubernetes#140422
      • KEP-5729: Promote DRAWorkloadResourceClaims to beta kubernetes#140334
    • Docs (k/website) update PR(s):
      • KEP-5729: Update ResourceClaim Support for Workloads for beta website#56184

/sig scheduling
/wg device-management

Please keep this description up to date. This will help the Enhancement Team to track the evolution of the enhancement efficiently.

[nojnhuh]

cc @johnbelamaric @wojtek-t @erictune @helayoty

/assign

[macsko]

/milestone v1.36
/label lead-opted-in