Bump container hooks to v0.8.8 and harden smoke tests

[jeanschmidt]

Impact: CI runners, integration tests, smoke tests
Risk: High

See jeanschmidt/runner-container-hooks#1 for details on what v0.8.7 introduces.

What

Bumps the patched runner-container-hooks from v0.8.5 to v0.8.7 (from jeanschmidt/runner-container-hooks fork), standardizes all integration test containers on ghcr.io/actions/actions-runner:latest, and hardens smoke tests against node churn edge cases.

Why

The container hooks v0.8.7 release includes fixes from jeanschmidt/runner-container-hooks#1. Smoke tests were flaking when pods remained scheduled on nodes that had already been terminated (disappeared from the API server) or on cordoned nodes.

How

• Switched all test containers to actions-runner image — GPU tests rely on host-level NVIDIA drivers exposed into the container, not on CUDA libraries in the container image
• Added "disappeared node" tracking in smoke tests — pods referencing nodes no longer in the API server are excluded from failure counts rather than causing false positives
• Added cordoned node (spec.unschedulable) detection to the unstable node classifier

Changes

Container hooks bump:

• hooks-warmer.yaml: HOOKS_VERSION 0.8.5 → 0.8.7

Integration test container standardization:

• build-image.yaml: Replaced privileged moby/buildkit:v0.29.0 container with non-privileged actions-runner:latest + runtime buildctl install step
• integration-test.yaml.tpl: GPU test jobs (test-gpu-t4, test-gpu-t4-multi, test-gpu-b200-2) switched from nvidia/cuda:12.6.3-base-ubuntu22.04 to actions-runner:latest
• workflow_generator.py: GPU_CONTAINER constant changed from nvidia/cuda:12.6.3-runtime-ubuntu22.04 to actions-runner:latest

Smoke test hardening:

• helpers.py: _is_node_unstable() now treats cordoned nodes (spec.unschedulable) as unstable; new get_all_node_names() helper
• test_logging.py: test_alloy_pods_running now excludes pods on "disappeared" nodes (node no longer in API server) from failure assertions; reports count in error messages

Testing

$  just integration-test arc-staging
Updating kubeconfig for pytorch-arc-staging (us-west-1)...
Updated context pytorch-arc-staging in /Users/jschmidt/.kube/config
17:36:04 [INFO] Integration test for cluster: arc-staging (pytorch-arc-staging)
17:36:04 [INFO]   Runner prefix: 'c-mt-'
17:36:04 [INFO]   B200 enabled: False
17:36:04 [INFO]   Release runners: True
17:36:04 [INFO]   Cache enforcer: True
17:36:04 [INFO]   PyPI cache slugs: cpu cu126 cu128 cu130
17:36:04 [INFO]   Smoke tests: skip
17:36:04 [INFO]   Compactor tests: skip
17:36:04 [INFO]   Branch: osdc-integration-test-arc-staging
17:36:04 [INFO] Phase 0: Cleaning up stale PRs...
17:36:09 [INFO] Phase 1: Checking for active runner pods (arc-staging only)...
17:36:16 [INFO]   No runner pods active. Skipping pool clear.
17:36:16 [INFO]   Canary repo already cloned at /Users/jschmidt/meta/ciforge/osdc/upstream/osdc/.scratch/pytorch-canary, fetching...
17:36:17 [INFO] Phase 2: Preparing PR...
17:36:33 [INFO]   PR #411 created: https://github.com/pytorch/pytorch-canary/pull/411
17:36:33 [INFO] Phase 3: Running parallel validation...
17:36:33 [INFO] Phase 4: Waiting for PR workflow runs (timeout: 50 min, buffer: 10 min)...
17:36:33 [INFO]   Filtering to runs created after 2026-04-11T00:36:17.795084+00:00
17:36:40 [INFO]   Run: OSDC Integration Test — https://github.com/pytorch/pytorch-canary/actions/runs/24270460678
17:36:40 [INFO]   1/1 runs still in progress...
17:37:13 [INFO]   1/1 runs still in progress...
17:37:45 [INFO]   1/1 runs still in progress...
17:38:18 [INFO]   1/1 runs still in progress...
17:38:52 [INFO]   1/1 runs still in progress...
17:39:24 [INFO]   1/1 runs still in progress...
17:40:00 [INFO]   1/1 runs still in progress...
17:40:31 [INFO]   1/1 runs still in progress...
17:41:05 [INFO]   1/1 runs still in progress...
17:41:40 [INFO]   1/1 runs still in progress...
17:42:13 [INFO]   1/1 runs still in progress...
17:42:49 [INFO]   All 1 run(s) completed.


============================================================
  OSDC Integration Test Results
============================================================
  Cluster: arc-staging (pytorch-arc-staging)
  Date:    2026-04-11 00:42 UTC

  PR Workflow Jobs:
    ✓ test-gpu-t4                    success
    ✓ test-pypi-cache-defaults       success
    ✓ test-pypi-cache-action-cuda    success
    ✓ test-git-cache                 success
    ✓ test-pypi-cache-action-cpu     success
    ✓ test-cpu-x86-amx               success
    ✓ test-cpu-arm64                 success
    ✓ test-cpu-x86-avx512            success
    ✓ test-cache-enforcer            success
    ✓ test-release-arm64             success
    ✓ test-gpu-t4-multi              success
    ✓ test-harbor                    success
    ✓ build-amd64 / build            success
    ✓ build-arm64 / build            success

  Smoke            ⊘ SKIPPED
  Compactor        ⊘ SKIPPED

  Overall: PASSED
============================================================

17:42:53 [INFO] Phase 5: Closing PR #411...
17:42:56 [INFO] Total integration test time: 6m52s

[huydhn]

LGTM!