[Story]: Create migration for GPU users to accelerator profiles

[Gkrumbach07]

Goal

Create a migration so that users using gpus are automatically migrated to accelerator profiles

Dependency issue

• HBU/GPU detection in the backend #1441

Itemized goals

• Backend — When we startup — in kube.ts we can handle a migration (catch ALL promise errors — exit gracefully and use fastify logging)
  • Check for migration-gpu-status configmap in dashboard namespace — if found, exit early
  • Read existing AcceleratorProfiles
  • If > 0 — do nothing, exit early
  • If 404 shows up — CRD may not be installed, exit early
  • If not 404, but 0 returned:
    • if — create our default migrated-gpu with nvidia.com/gpu identifier, Nivida GPU display name, enabled, toleration equal to our existing setting (before Habana)
    • If bad detection, error or whatever — exit early
    • If can’t find GPUs, create the configmap anyways (next step)
  • Create configmap to flag operation as successful — migration-gpu-status in the dashboard namespace, simple stringData of migrated-completed: true
• UI does nothing special for upgrade, reacts purely on data
  • Don’t show any dropdown if no AcceleratorProfile CRs exist
  • Don’t error out if CRD doesn’t exist
  • Show dropdown normally no matter how many of those or origin of those AcceleratorProfile CRs

Aspects continued elsewhere

No response

Mocks

No mocks