fix(cli): resources show suppresses the phantom GPU on GPU-less machines

[saadqbal]

What

Found during a proactive pre-bugbot review of the #266 promote. On a CPU-only host, tracebloc resources (show) reports a phantom GPU:

• The chart stamps GPU_REQUESTS/GPU_LIMITS = nvidia.com/gpu=1 as literal container env on every install, so ParseTraining returns HasGPU=true.
• renderResources then prints per training run: up to N CPU · M GiB · 1 GPU even though the machine has no GPU — and under --verbose it also prints gpu: none detected, contradicting itself.

The set path already normalizes exactly this (phantomGPU := current.HasGPU && !machineHasGPU, Bugbot #241); the show path didn't.

Fix

Drop train.HasGPU when the node read succeeded and the node exposes no GPU — the same nodeErr == nil && len(machine.GPU) == 0 guard already used for the "none detected" detail. A node-read failure leaves it untouched (can't confirm absence; the capacity line already says "unavailable").

Verification

• gofmt/vet clean; go test ./internal/cli/ -run TestRenderResources green.
• New TestRenderResources_PhantomGPUSuppressed (no-GPU node + phantom GPU_LIMITS env) — fails before the fix (prints · 1 GPU), passes after. TestRenderResources_GPUSurfaced (real GPU node) still shows the GPU.

Once merged to develop, promote PR #266 picks it up.

🤖 Generated with Claude Code

---

Note

Low Risk
Read-only display fix in the CLI resources view with a focused regression test; no cluster mutation or auth changes.

Overview
On CPU-only hosts, tracebloc resources could still show 1 GPU on the per-run ceiling because chart-default GPU_LIMITS env makes ParseTraining set HasGPU, which clashed with verbose gpu: none detected.

renderResources now clears train.HasGPU when node capacity was read successfully and the machine has no GPU devices—the same guard as the verbose detail and aligned with resources set phantom-GPU handling (Bugbot #241). If node listing fails, behavior is unchanged.

Adds TestRenderResources_PhantomGPUSuppressed for a no-GPU node with phantom GPU_LIMITS env.

^{Reviewed by Cursor Bugbot for commit 7331df1. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

👋 Heads-up — Code review queue is at 36 / 30

Above the WIP limit. The team convention is to review existing PRs before opening new work.

Open PRs currently in Code review (oldest first):

• averaging-service#181 — feat(weights): normalize-on-read — mixed cycles average instead of rejecting (SEC-03 §8 step 2) · author: @shujaatTracebloc · no reviewer assigned
• averaging-service#182 — feat(weights): averaging writes SafeTensors (TF + PyTorch) — SEC-03 §8 step 4 · author: @shujaatTracebloc · no reviewer assigned
• backend#1079 — feat(global_meta): edge dataset_meta exposure + attributes contract at ingest (#924 G4a) · author: @divyasinghds · no reviewer assigned
• backend#1086 — docs(rfc): SafeTensors weight-format migration — SEC-03 Phase 1 (RFC 0004) · author: @shujaatTracebloc · no reviewer assigned
• backend#1093 — chore(deps): bump django from 5.2.14 to 5.2.15 · author: @dependabot · no reviewer assigned
• backend#1095 — feat(experiment): configurable preprocessing knobs incl. tabular scaler — RFC 0003 L1 + L1b (#1094) · author: @LukasWodka · no reviewer assigned
• backend#1100 — fix(boot): pin SDK install to tracebloc==0.11.2, drop 404 dev line (#1098) · author: @LukasWodka · reviewer: @saqlainsyed007
• backend#1105 — perf(api): query micro-fixes — notifications N+1, cached data_scientist, composite index, sampling (#975) · author: @aptracebloc · no reviewer assigned
• cli#266 — main - > enhance CLI features and tests · author: @saadqbal · no reviewer assigned
• client-runtime#179 — Staging -> Resync ingest.v1 schema for time_series_classification and improve pod watch · author: @saadqbal · no reviewer assigned

Pull from review before opening new work. (This is a nudge from the kanban WIP check, not a block.)

[LukasWodka]

Reviewed — correct and well-scoped. ✅

The guard only drops HasGPU when the node read succeeded and confirmed no GPU (nodeErr == nil && len(machine.GPU) == 0), so a node-read failure correctly leaves it untouched (can't prove absence — the capacity line already says "unavailable"). Mirrors the set path's #241 normalization, and the new test genuinely fails-before/passes-after on the · 1 GPU assertion.

Two tiny test nits (non-blocking):

• In TestRenderResources_PhantomGPUSuppressed, the first assertion strings.Contains(out, "GPU") && !strings.Contains(out, "none detected") is effectively a no-op here: under WithVerbose(true) the raw-env dump prints GPU_LIMITS (so Contains("GPU") is already true) andnone detected is present, so the && never fires. The · 1 GPU check is the one actually carrying the regression — worth leaning on that (and maybe dropping the first).
• Optional: also assert the per-run line's positive shape (ends at GiB), so a future refactor that drops the GPU clause differently is still pinned.

LGTM to merge.

—
Review assisted by Claude Code.