[thxCode]

What type of PR is this?

/kind bug
/area devicemanager

What this PR does / why we need it:

Fix three edge cases in GetPreferredAllocation when honoring device.gpustack.ai/accelerator.preferred-index and device.gpustack.ai/accelerator.preferred-id annotations:

• Preserve the user-declared priority order in preferredAcceleratorIDsOf by returning an ordered, deduplicated []string instead of an unordered sets.Set[string].
• Order matched preferred accelerators by their declared preference rank (after mustInclude accelerators), cap sel.Selected to allocationSize so kubelet does not receive more tokens than requested and pick among them in random map order, and sort the final hinted tokens in ascending hardware order.
• Only fall back to sel.Unselected (or return an empty hint when len(sel.Unselected) == 0) when remainingSize > 0, preventing a satisfied preferred selection from being discarded when all available accelerators on the node are listed in preferred-index and one listed accelerator is unavailable.
• Add table-driven unit tests in pkg/deviceplugin/server_test.go covering out-of-order priority selection, over-specified preferred lists, and empty-unselected fallback.

Which issue(s) this PR links to:

NONE

Special notes for your reviewer:

NONE

Does this PR introduce a user-facing change?

Fix GetPreferredAllocation to honor preferred-index / preferred-id priority order, cap hinted devices at the requested allocation size, and keep satisfied preferred selections when no unselected fallback devices remain.

[gpustack-code-review]

🔍 OpenCodeReview found 1 issue(s) in this PR.

• ✅ Successfully posted inline: 0 comment(s)
• 📋 Routed to summary by policy: 1 comment(s)

---

📄 pkg/deviceplugin/server.go (L479-L481)

⚠️ GitHub could not post this as an inline comment: Routed to summary (severity low · category test)

Reminder (project rule, non-blocking to verify here): this is a bug fix in allocation logic — preference ordering, capping the hint to allocationSize, and eliminating the negative-index panic on the unselected fallback — so it should ship a regression test in the same PR. *_test.go files are excluded from this review, so the presence or absence of such a test cannot be confirmed here.

[thxCode]

Review Triage Summary

ID	Location	Verdict	Outcome
IC_kwDOSVaals8AAAABguvfWQ (6491463513)	pkg/deviceplugin/server.go:479-481	False positive (already covered)	No code change required. Regression tests are already included in this PR in pkg/deviceplugin/server_test.go (TestResourceServer_PreferredAllocation_OrderSizeAndFallback and TestResourceServer_PreferredAllocation_StalePreferredIDDoesNotPanic), which OpenCodeReview did not see because *_test.go files are excluded from its diff scope.