S1: Stop Ashlar's inherited workflows from flooding the shared runner #34

Closed
opened 2026-11-09 17:05:55 +00:00 by claude · 3 comments
Member

Gaming/ashlar inherited OpenMW's .github/workflows (push, release, macOS, Windows). Gitea runs them because the repo has no .gitea/workflows. On 2026-11-09 that made 41 runs in a day, mostly queued, including jobs that can never run here. Every other repo's CI waits behind them.

Work

  • Check how Gitea 1.27 picks a workflow directory. Prefer adding .gitea/workflows/ so upstream's files stay untouched for merges; if that doesn't override .github, disable them another way and explain why.
  • Add a minimal Linux-only workflow, or none, as agreed on the Ashlar milestones. Limit runs with concurrency and path filters.
  • Cancel Ashlar's queued and running backlog that the old workflows created.

Acceptance

  • A push to Ashlar starts at most one run, and it can finish on this runner.
  • No Ashlar macOS or Windows jobs are queued.
  • Comment with run links.
Gaming/ashlar inherited OpenMW's `.github/workflows` (push, release, macOS, Windows). Gitea runs them because the repo has no `.gitea/workflows`. On 2026-11-09 that made 41 runs in a day, mostly queued, including jobs that can never run here. Every other repo's CI waits behind them. **Work** - Check how Gitea 1.27 picks a workflow directory. Prefer adding `.gitea/workflows/` so upstream's files stay untouched for merges; if that doesn't override `.github`, disable them another way and explain why. - Add a minimal Linux-only workflow, or none, as agreed on the Ashlar milestones. Limit runs with `concurrency` and path filters. - Cancel Ashlar's queued and running backlog that the old workflows created. **Acceptance** - A push to Ashlar starts at most one run, and it can finish on this runner. - No Ashlar macOS or Windows jobs are queued. - Comment with run links.
claude added this to the CI/CD 0: Stabilize the runner milestone 2026-11-09 17:05:55 +00:00
claude added the cicd
priority
critical
labels 2026-11-09 17:05:55 +00:00
claude added this to the CI/CD Roadmap project 2026-11-09 17:07:33 +00:00
claude moved this to To Do in CI/CD Roadmap on 2026-11-09 17:07:59 +00:00
Author
Member

Done (except the backlog cancel): Gaming/ashlar@50ab15cbe9 adds .gitea/workflows/ashlar-checks.yaml.

How Gitea picks the directory: Gitea reads workflows from .gitea/workflows in the pushed commit and falls back to .github/workflows only if .gitea/workflows is missing. Adding our own directory therefore turns off upstream's four workflows here. Their files stay unchanged, so upstream merges stay clean.

The workflow: it runs on main only, when docker/** or .gitea/workflows/** changes, plus a manual trigger. It has a concurrency group with cancel-in-progress and a 10-minute timeout, and it runs bash -n on the Docker build scripts. A full engine build can't run on this runner yet. That stays in Gaming/ashlar#4 and G10 (#58).

Tested

Caveat, documented in ASHLAR.md: a ref whose commit has no .gitea/workflows, such as a newly pushed upstream openmw-* tag, still falls back to .github. ASHLAR.md now says not to push upstream tags or branches to origin.

Still open: 41 old runs remain queued (runs 1 to 41, https://gitea.eom.dev/Gaming/ashlar/actions). Each holds macOS and Windows jobs that can never be scheduled here. The scheduled task's attempt to cancel them was blocked by the session's permission policy, so they're still there. Either of these would finish it:

  1. Recommended: allow mcp__Gitea__actions_run_write for the scheduled task, and the next run cancels runs 19682 to 19722 (Ashlar's only).
  2. Cancel them yourself from the Actions page.

Leaving this open until the queue is clear, because that's part of the acceptance criteria.

— eom-cicd-roadmap

**Done (except the backlog cancel):** Gaming/ashlar@50ab15cbe9 adds `.gitea/workflows/ashlar-checks.yaml`. **How Gitea picks the directory:** Gitea reads workflows from `.gitea/workflows` in the pushed commit and falls back to `.github/workflows` only if `.gitea/workflows` is missing. Adding our own directory therefore turns off upstream's four workflows here. Their files stay unchanged, so upstream merges stay clean. **The workflow:** it runs on `main` only, when `docker/**` or `.gitea/workflows/**` changes, plus a manual trigger. It has a `concurrency` group with cancel-in-progress and a 10-minute timeout, and it runs `bash -n` on the Docker build scripts. A full engine build can't run on this runner yet. That stays in Gaming/ashlar#4 and G10 (#58). **Tested** - `actionlint` passes (in the `rhysd/actionlint` container). - The push of 50ab15c started exactly one run, with one Linux job and no macOS or Windows jobs. It passed in 40 seconds: https://gitea.eom.dev/Gaming/ashlar/actions/runs/19749 **Caveat, documented in ASHLAR.md:** a ref whose commit has no `.gitea/workflows`, such as a newly pushed upstream `openmw-*` tag, still falls back to `.github`. ASHLAR.md now says not to push upstream tags or branches to `origin`. **Still open: 41 old runs remain queued** (runs 1 to 41, https://gitea.eom.dev/Gaming/ashlar/actions). Each holds macOS and Windows jobs that can never be scheduled here. The scheduled task's attempt to cancel them was blocked by the session's permission policy, so they're still there. Either of these would finish it: 1. **Recommended:** allow `mcp__Gitea__actions_run_write` for the scheduled task, and the next run cancels runs 19682 to 19722 (Ashlar's only). 2. Cancel them yourself from the Actions page. Leaving this open until the queue is clear, because that's part of the acceptance criteria. — eom-cicd-roadmap
claude added the needs-eric label 2026-11-10 01:04:19 +00:00
Author
Member

Update, 2026-11-10: the scheduled task's attempt to cancel run 19682 was denied by its permission policy again, so all 41 old runs are still queued (their Ubuntu jobs finished, but the macOS and Windows jobs are still waiting).

These should clear on their own. Gitea cancels any job that no runner has picked up within ABANDONED_JOB_TIMEOUT, which defaults to 24h in 1.27. The oldest of these runs started about 24 hours ago. S6 (#39) lowers the timeout to 12h. The next run of this task will check whether Gitea has cancelled them. If it has, this issue can close with no action from you. Cancelling them from the Actions page is still the fastest route if you'd rather not wait.

— eom-cicd-roadmap

**Update, 2026-11-10:** the scheduled task's attempt to cancel run 19682 was denied by its permission policy again, so all 41 old runs are still queued (their Ubuntu jobs finished, but the macOS and Windows jobs are still waiting). These should clear on their own. Gitea cancels any job that no runner has picked up within `ABANDONED_JOB_TIMEOUT`, which defaults to 24h in 1.27. The oldest of these runs started about 24 hours ago. S6 (#39) lowers the timeout to 12h. The next run of this task will check whether Gitea has cancelled them. If it has, this issue can close with no action from you. Cancelling them from the Actions page is still the fastest route if you'd rather not wait. — eom-cicd-roadmap
Author
Member

Backlog cleared, 2026-11-10. At Eric's request I cancelled the 23 old Ashlar runs that were still queued (runs 8–29 and 41, IDs 19689–19710 and 19722). Ashlar now has 0 queued, 0 waiting and 0 in-progress runs, and new pushes start one Linux run that passes (runs 42–44). Every acceptance criterion is met, so I'm closing this.

For future runs: Gitea 1.27's REST API has no endpoint for cancelling a run. mcp__Gitea__actions_run_write cancel_run returns 404, which is likely why the earlier attempts failed and not only because of permission policy. Cancelling works only through the web UI: a signed-in, same-origin POST /<owner>/<repo>/actions/runs/<run id>/cancel.

— Claude, interactive session with Eric

**Backlog cleared, 2026-11-10.** At Eric's request I cancelled the 23 old Ashlar runs that were still queued (runs 8–29 and 41, IDs 19689–19710 and 19722). Ashlar now has 0 queued, 0 waiting and 0 in-progress runs, and new pushes start one Linux run that passes (runs 42–44). Every acceptance criterion is met, so I'm closing this. **For future runs:** Gitea 1.27's REST API has no endpoint for cancelling a run. `mcp__Gitea__actions_run_write cancel_run` returns 404, which is likely why the earlier attempts failed and not only because of permission policy. Cancelling works only through the web UI: a signed-in, same-origin `POST /<owner>/<repo>/actions/runs/<run id>/cancel`. — Claude, interactive session with Eric
claude removed the needs-eric label 2026-11-10 19:13:01 +00:00
eric moved this to Done in CI/CD Roadmap on 2026-11-10 19:13:06 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: DevOps/software-infrastructure#34