fix: use default pgpu alias when an override is not available

[rajatchopra]

The GFD pod needs a GPU and it needs to know which 'device name' is used for publishing the devices. When P_GPU_ALIAS is present, it is that one, but otherwise pick one GPU device's native name.

[coveralls]

Pull Request Test Coverage Report for Build 21923530054

Details

• 0 of 31 (0.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.9%) to 28.412%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
cmd/main.go	0	9	0.0%
pkg/device_plugin/gfd.go	0	22	0.0%

Totals
Change from base Build 21921289502:	-0.9%
Covered Lines:	229
Relevant Lines:	806

---

💛 - Coveralls

[jojimt]

@rajatchopra can we add the defaults at source (https://github.com/NVIDIA/sandbox-device-plugin/blob/main/cmd/main.go#L37-L38)

[Copilot]

Pull request overview

Updates the NVIDIA sandbox device plugin to pick a usable GPU resource name for the GFD pod when P_GPU_ALIAS isn’t set, and refines CDI spec generation and logging around IOMMUFD/VFIO behavior.

Changes:

• Use a discovered GPU device name as the default resource name for the GFD pod when P_GPU_ALIAS is not provided.
• Rework CDI spec generation to support alias vs per-device-type output for both GPUs and NVSwitches.
• Adjust IOMMUFD-related logging and allocation paths (and make cdiRoot configurable for tests).

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
pkg/device_plugin/gfd.go	Builds GFD pod resource name from a derived GPU device name; adds helper to select that name.
pkg/device_plugin/generic_device_plugin.go	Refactors allocation branching/logging between IOMMUFD and legacy VFIO modes.
pkg/device_plugin/device_plugin.go	Updates logs and changes how IOMMU keys are derived in IOMMUFD mode.
pkg/device_plugin/constants.go	Makes cdiRoot mutable for test redirection.
pkg/device_plugin/cdi.go	Reworks CDI generation for alias vs heterogeneous modes and changes device naming in CDI specs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rajatchopra]

@rajatchopra can we add the defaults at source (https://github.com/NVIDIA/sandbox-device-plugin/blob/main/cmd/main.go#L37-L38)

No I guess. For two reasons:

1. We will occlude the original device name if that was the intent
2. heterogeneous case will never work

[jojimt]

We will occlude the original device name if that was the intent

1. This is something we inherited from kubevirt. I don't think a use case exists.
2. We can make a distinction between not set vs set to "" by using os.LookupEnv.

[rajatchopra]

2. We can make a distinction between not set vs set to "" by using os.LookupEnv.

Yes, that is a good way to get around. Fixed.

[jojimt]

lgtm