DRA: Resource Availability Visibility

[johnbelamaric]

Enhancement Description

DRA stores raw capacity data in ResourceSlices. The consumption of that capacity is recorded in ResourceClaims. However, there is no simple way for a user to see how much capacity is still available on a node or in a resource pool (a part of which is represented in a ResourceSlice). It would be useful to be able to summarize this information for the user, particularly because ResourceClaims are namespaced and so most users would not be able to even see all the resource claims that are consuming a given slice.

/wg device-management
/sig cli
/sig node
/sig scheduling

cc @mortent @pohly @klueska @ArangoGutierrez

• One-line enhancement description (can be used as a release note): You can now use kubectl describe resourceslice and kubectl describe node to see DRA resource pool consumption information.
• Kubernetes Enhancement Proposal: https://github.com/kubernetes/enhancements/tree/master/keps/sig-node/5677-dra-resource-availability-visibility
• Discussion Links:
  • https://kubernetes.slack.com/archives/C0409NGC1TK/p1761676222405659
  • https://docs.google.com/document/d/16dKvcTf35I0uUxeJ0UTF6g_dgktKh70UQnJ6b4Rf8nU/edit?tab=t.0
• PRs by stage and milestone:
  • Alpha - v1.36
    • KEP (k/enhancements) update PR(s):
      • KEP-5677: DRA Resource Availability Visibility #5749
    • Code (k/k) update PR(s):
      • KEP-5677: Add ResourcePoolStatusRequest API for DRA resource availability visibility kubernetes#137028
    • Docs (k/website) update PR(s): Add docs for ResourcePoolStatusRequest (KEP-5677) website#54456

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

[johnbelamaric]

From @erictune : https://docs.google.com/document/d/16dKvcTf35I0uUxeJ0UTF6g_dgktKh70UQnJ6b4Rf8nU/edit?tab=t.0

[johnbelamaric]

cc @seans3 who I believe has given some thought to this sort of thing

[johnbelamaric]

It's likely to accomplish this we would have ResourceSlice controllers that summarize information into a ResourceSlice.Status field. Or, perhaps we need another way since pool information is spread across many slices.

Along those lines, we also have a number of cases where we are unable to fully validate the resource pools represented by the ResourceSlices, because normal API validation only sees the individual ResourceSlice, not the entire set of slices that represent the pool. Today we detect these issues during scheduling; if we have a controller summarizing information about a slice and/or pool, checking this validity would also be useful.