NVIDIA / gpu-operator / 29760022501

31%

main: 35%

push

github

feat(validator): re-enable SR-IOV vGPU VFs before waiting for them

On the vGPU (sandbox) workload path, SR-IOV Virtual Functions are
runtime state that does not survive a node reboot. The validator already
waits for the VFs to appear before letting the vGPU device-manager apply
its config, but nothing re-creates them, so after a reboot validation
blocks until the VFs are re-established out-of-band.

Re-enable the VFs from the vGPU Manager validator via 'sriov-manage -e
ALL' inside the resolved driver root (driver container or host) right
before the existing wait. The call is idempotent (skipped when every
SR-IOV-capable GPU already has its full VF count, so it never disturbs
VFs already assigned to running VMs) and best-effort (on failure the
existing wait still runs, preserving prior behavior on setups where the
VFs are created out-of-band).

Scope is limited to VF re-enablement; no GPU reset and no MIG
reconfiguration are performed. Refs #2600.

Assisted-By: Claude <noreply@anthropic.com>
Signed-off-by: Aleksei Sviridkin <f@lex.la>

9 of 33 new or added lines in 1 file covered. (27.27%)

4125 of 13205 relevant lines covered (31.24%)

0.36 hits per line