Problem

GPU CI tests (inference, conformance, training) only trigger on a narrow set of path patterns. Recent PRs that affected GPU test behavior were not covered:

• feat: wire NFDHardwareDetector into production snapshot pipeline #502 changed pkg/collector/gpu/ and pkg/snapshotter/ — broke GPU model detection in snapshot validation, but GPU tests didn't run on the PR
• feat(validator): add --node-selector and --toleration flags for validation workload scheduling #444 changed pkg/validator/job/deployer.go — broke imagePullPolicy for local images, causing 100% GPU test failure for a week, but the regression wasn't caught because the PR only triggered GPU tests via kind-inference.yaml path match (coincidental)

Current trigger paths in GPU workflows (e.g., gpu-h100-inference-test.yaml):

paths:
  - '.github/workflows/gpu-h100-inference-test.yaml'
  - '.github/actions/gpu-cluster-setup/**'
  - '.github/actions/gpu-operator-install/**'
  - '.github/actions/aicr-build/**'
  - 'validators/*/Dockerfile'
  - 'pkg/evidence/**'
  - 'pkg/validator/checks/conformance/**'
  - 'pkg/validator/checks/deployment/**'
  - '.github/actions/gpu-test-cleanup/**'
  - '.github/actions/load-versions/**'
  - 'tests/manifests/**'
  - 'tests/chainsaw/ai-conformance/**'
  - 'recipes/overlays/kind*.yaml'   # (varies per workflow)
  - 'kwok/manifests/karpenter/**'
  - 'kwok/scripts/install-karpenter-kwok.sh'

Missing paths that affect GPU test behavior:

Proposal

1. Expand path triggers (catches known gaps)

Add the missing paths to all GPU test workflows:

paths:
  # Existing triggers...
  
  # Collector/snapshotter — affects GPU detection and snapshot content
  - 'pkg/collector/**'
  - 'pkg/snapshotter/**'
  - '.github/actions/gpu-snapshot-validate/**'
  
  # Validator infrastructure — affects how validator Jobs are deployed and run
  - 'pkg/validator/job/**'
  - 'pkg/validator/catalog/**'
  - 'pkg/defaults/timeouts.go'
  
  # Conformance validator source — the actual conformance checks
  - 'validators/conformance/**'

2. Add label-based trigger (safety net for unforeseen cases)

Add a run-gpu-tests label trigger to all GPU workflows:

on:
  pull_request:
    types: [labeled]

With a workflow condition:

if: >
  github.event_name \!= 'pull_request' ||
  github.event.action \!= 'labeled' ||
  github.event.label.name == 'run-gpu-tests'

This allows reviewers to manually trigger GPU tests on any PR by adding the run-gpu-tests label — catching cases where the path list is incomplete.

Affected workflows

• .github/workflows/gpu-h100-inference-test.yaml
• .github/workflows/gpu-h100-conformance-test.yaml
• .github/workflows/gpu-h100-training-test.yaml
• .github/workflows/gpu-t4-smoke-test.yaml

Impact

With these changes:

• feat: wire NFDHardwareDetector into production snapshot pipeline #502 would have triggered GPU tests (matches pkg/collector/**)
• feat(validator): add --node-selector and --toleration flags for validation workload scheduling #444 would have triggered GPU tests (matches pkg/validator/job/**)
• Any PR can be manually tested on GPU runners via the run-gpu-tests label