Define and enforce cross-team GPU access policies

[acalhounRH]

Context

From the May 28 demo meeting, Aanya asked about the policy for using GPUs on clusters designated for other teams. Ashish advised maintaining the current model of sticking to team-designated clusters until broader consensus is reached.

Current State

• No code-level enforcement of team-to-cluster assignments
• Reservations have a team field but it is optional and purely informational
• Any authenticated user can reserve any cluster
• The current policy is organizational/verbal: teams should use their designated clusters

Proposed Changes

• Define a formal cross-team access policy and document it
• Consider implementing policy in the system:
  • Cluster ownership: Associate clusters with teams in the registry
  • Soft warnings: Alert users when reserving a cluster outside their team assignment
  • Approval workflow: Require lead approval for cross-team reservations
  • Visibility: Show team designations prominently on cluster cards and reservation forms
• Evaluate whether technical enforcement is needed or if improved visibility is sufficient

Decision Needed

The team needs to align on:

1. Whether cross-team access should remain informal or be codified
2. Level of enforcement (advisory, approval-gated, or blocked)
3. Process for requesting cross-team access when needed (e.g., urgent customer requirements)