DRA: Reserved Capacity

[johnbelamaric]

Enhancement Description

Accelerators fail. Especially for multi-node devices, it can be useful to "hold back" a certain number of accelerators so that a failure can be "repaired" (by substituting in one of the hold back devices). We should consider if this warrants any special handling in DRA. For example, we could model reserve resources so they require a special device toleration. This could be used to allow them to be scheduled with low-priority or preemptible workloads unless they are needed to repair a multi-node device.

• One-line enhancement description (can be used as a release note): DRA now allows vendors to model "reserved" capacity, which can be used to manage failures in collections of accelerators or other devices.
• Kubernetes Enhancement Proposal: TBD
• Discussion Link:
• PRs by stage and milestone:
  • Alpha - v1.xx
    • KEP (k/enhancements) update PR(s):
    • Code (k/k) update PR(s):
    • Docs (k/website) update PR(s):

/wg device-management
/sig scheduling
/sig node
/cc @pohly @klueska @mortent

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