Summary:

• Depending on CNV team plans for using SNO in RDU4 GPU enabled HW, consider sharing SNO instances used in both (or just VAF) teams CI jobs with both VAF and CNV teams needs

Goal:

• When requesting a host for GPU related CI jobs by VAF or CNV teams, reduce the count from 2 to 1.5, for vaf, and from 1 to 0.5, for cnv (assuming that SNO host is shared by two teams, we count it as 0.5)
• When requesting a host for GPU related CI jobs by VAF or CNV teams, do not reconfigure networking if the host was previously leased by either of the teams, which speeds up administrative process and gets the lease available faster.

TimeBox:

• 1 sprint

Deliverables/Outcomes:

• We know what plans CNV team has for testing GPU features of OCP in RDU4 HW, and if we can fit it with our needs as well
• We know how to configure zuul and/or prow CI jobs of VAF/CNV teams for such a shared SNO setup.

Limitations and caveats:

• If testing things for VAF on OCP nightly builds works for us (likely it does but for rhoso 19 only?)
• In zuul, we can provide a shared SNO instance for multiple dependent CI jobs, but what about prow?
• For vGPU / MIG testing in rhoso18 scope, we could use brq2 lab, unless there is NVLink/NVSwitch requirement, which makes it wanting RDU4 hw then