openclaw - 💡(How to fix) Fix CI: Blacksmith-backed PR checks remain queued with no runner assigned

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Blacksmith-backed GitHub Actions jobs for multiple PRs remain queued for hours with no runner assigned, while GitHub-hosted ubuntu-24.04 jobs continue to run normally.

Steps to reproduce

1. Inspect queued Actions runs:

   gh run list --repo openclaw/openclaw --status queued --limit 100 \
     --json databaseId,workflowName,status,createdAt,headBranch,headSha,event,url

2. Inspect queued jobs for an affected run:

   gh api repos/openclaw/openclaw/actions/runs/26172608830/jobs --paginate \
     --jq '.jobs[] | {id,name,status,started_at,runner_name,runner_group_name,labels}'

3. Compare with GitHub-hosted jobs on the same PR head, such as Workflow Sanity jobs on ubuntu-24.04.

Expected behavior

Blacksmith-backed jobs with labels such as blacksmith-4vcpu-ubuntu-2404 and blacksmith-8vcpu-ubuntu-2404 should be assigned to an available runner and start, or fail with a clear infrastructure signal if no matching runner pool is available.

Actual behavior

Observed at 2026-05-20T17:59:53Z:

• #82258 had Blacksmith-backed jobs queued for about 2h30m.
• Repo-wide queued Actions showed 27 unique queued branches.
• At least 11 open PRs mapped directly to queued runs: #84418, #84621, #84626, #84636, #84641, #84645, #84648, #84649, #84652, #84658, #84661.
• A main push CI run was also queued.
• The stuck workflows are mainly CI, CodeQL, CodeQL Critical Quality, and OpenGrep — PR Diff.
• Queued jobs report labels such as blacksmith-4vcpu-ubuntu-2404 and blacksmith-8vcpu-ubuntu-2404, with empty runner_name / runner_group_name.
• GitHub-hosted jobs on ubuntu-24.04, such as Workflow Sanity jobs, were picked up and completed normally on the same PR head.

Example evidence from #82258:

preflight                         QUEUED since 2026-05-20T15:28:49Z  labels: blacksmith-4vcpu-ubuntu-2404  runner_name: empty
security-fast                     QUEUED since 2026-05-20T15:28:50Z  labels: blacksmith-4vcpu-ubuntu-2404  runner_name: empty
Security High (core-auth-secrets) QUEUED since 2026-05-20T15:28:49Z  labels: blacksmith-8vcpu-ubuntu-2404  runner_name: empty
Scan changed paths (precise)      QUEUED since 2026-05-20T15:28:49Z  labels: blacksmith-4vcpu-ubuntu-2404  runner_name: empty

OpenClaw version

N/A - repository CI infrastructure issue.

Operating system

N/A - GitHub Actions / Blacksmith runner infrastructure.

Install method

N/A - GitHub Actions / Blacksmith runner infrastructure.

Model

N/A - GitHub Actions / Blacksmith runner infrastructure.

Provider / routing chain

N/A - GitHub Actions / Blacksmith runner infrastructure.

Additional provider/model setup details

N/A - GitHub Actions / Blacksmith runner infrastructure.

Logs, screenshots, and evidence

gh pr checks 82258 --repo openclaw/openclaw --json name,state,bucket,startedAt,link

gh api repos/openclaw/openclaw/actions/runs/26172608830/jobs --paginate \
  --jq '.jobs[] | {id,name,status,started_at,runner_name,runner_group_name,labels}'

gh run list --repo openclaw/openclaw --status queued --limit 100 \
  --json databaseId,workflowName,status,createdAt,headBranch,headSha,event,url

Impact and severity

Affected: PR and main CI checks that use Blacksmith-backed runners.

Severity: High for merge velocity because required PR checks can remain pending without actionable code failures.

Frequency: Observed across dozens of queued workflow runs at the snapshot time.

Consequence: Contributors and maintainers cannot distinguish code failures from runner unavailability, and PRs can remain blocked despite local proof or passing GitHub-hosted checks.

Additional information

This looks like Blacksmith runner availability/routing rather than a PR-code failure because GitHub-hosted ubuntu-24.04 jobs completed normally, while Blacksmith-labeled jobs stayed queued with no assigned runner.