Point Forgejo runner substrate at in-cluster railiance01 (ADR-004)

Status probe checks k8s runner first; documents retirement of coulombcore
interim host runner.
This commit is contained in:
2026-07-03 22:29:28 +02:00
parent 7aae00e933
commit dc341d7cc6
2 changed files with 58 additions and 49 deletions

View File

@@ -2,75 +2,73 @@
Last reviewed: 2026-07-03
Status: first supported runner on coulombcore (interim). Forgejo instance runs on
railiance01; runner host is intentionally separate per Forgejo security guidance.
Status: **in-cluster runner on railiance01** (ADR-004). Supersedes interim coulombcore
host runner.
## Purpose
Provide a forge-owned Actions runner for `https://forgejo.coulomb.social` so
repos in the `coulomb` organization can build and publish without workstation
involvement. Mirrors the Gitea runner contract in
`docs/gitea-actions-runner-substrate.md`.
Forge-owned Actions runner for `https://forgejo.coulomb.social` so `coulomb`
organization repos build and publish without workstation or legacy-host Docker.
Label contract aligns with `docs/gitea-actions-runner-substrate.md` and
`docs/ci-runner-actions-gitops-ownership.md`.
## First Supported Runner
## Production Runner (ADR-004)
| Field | Value |
| --- | --- |
| Runner name | `railiance-coulombcore-build-01` |
| Runner scope | `coulomb` organization |
| Host | coulombcore (`92.205.130.254`) |
| Instance | `https://forgejo.coulomb.social/` |
| Binary | `forgejo-runner` v6.3.1 |
| Runtime | Host mode (Docker not installed on host yet) |
| Runner name | `railiance01-build-01` |
| Scope | `coulomb` organization |
| Placement | `forgejo` namespace, railiance01 k3s |
| Manifest | `railiance-apps/manifests/forgejo-runner.yaml` |
| Images | `code.forgejo.org/forgejo/runner:6.3.1`, `docker:27-dind` |
| Job runtime | DinD sidecar (`DOCKER_HOST=tcp://127.0.0.1:2375`) |
| State PVC | `forgejo-runner-data` |
| Cluster deploy authority | Not included |
### Labels
```text
self-hosted:host,linux:host,linux_amd64:host,container-build:host,registry-publish:host,railiance01:host
self-hosted:host,linux:host,linux_amd64:host,container-build:host,registry-publish:host,railiance01:host,ubuntu-latest:docker://node:20-bookworm,docker:docker://node:20-bookworm
```
`self-hosted` preserves compatibility with existing Gitea workflow `runs-on`
values during migration. Container image builds require Docker on the runner
host — install before cutover of image-publishing workflows.
## Service Layout
| Path | Purpose |
| --- | --- |
| `/usr/local/bin/forgejo-runner` | Runner binary |
| `/var/lib/forgejo-runner/config.yaml` | Generated config |
| `/var/lib/forgejo-runner/.runner` | Registration state (secret) |
| `/etc/systemd/system/forgejo-runner.service` | systemd unit |
## Probe Evidence (2026-07-03)
- Org `coulomb` created on Forgejo.
- Runner registered and `systemctl is-active forgejo-runner``active`.
- Repo `coulomb/forgejo-actions-probe` workflow `probe.yaml` (`runs-on: self-hosted`)
completed with `status: success` (API: `/api/v1/repos/coulomb/forgejo-actions-probe/actions/tasks`).
## Operator Commands
```bash
# Runner health (from railiance-forge)
RUNNER_HOST=coulombcore make forgejo-runner-status
# S5 deploy (railiance-apps)
cd ~/railiance-apps
make forgejo-runner-registration-deploy # after SOPS encrypt
make forgejo-runner-deploy
make forgejo-runner-status
# Forgejo app smoke (from railiance-apps)
make -C ~/railiance-apps forgejo-smoke
# Endpoint smoke
make forgejo-smoke
# Generate a new org registration token (in cluster, no token in logs)
# Generate registration token (in cluster; do not log token)
KUBECONFIG=~/.kube/config-hosteurope kubectl exec -n forgejo deploy/forgejo-gitea -- \
forgejo actions generate-runner-token --scope coulomb
```
## Secret Boundaries
Same as Gitea runner substrate: registration tokens and `.runner` files stay
off Git. Use approved secret paths on the runner host only.
- `forgejo-runner-registration` Kubernetes Secret — registration token only.
- Repo/org registry credentials — Forgejo Actions secrets, not in Git.
- No kubeconfig or OpenBao tokens on the runner pod by default.
## Follow-On
## Retired: Interim coulombcore Host Runner
1. Install Docker on the approved runner host for `container-build` jobs.
2. Move runner to a non-legacy host before coulombcore decommission (T03 drain).
3. Add repo-scoped registry credentials in Forgejo org secrets for image push.
`railiance-coulombcore-build-01` on coulombcore was a scheduling proof only.
Disable after in-cluster runner is healthy:
```bash
ssh coulombcore 'sudo systemctl disable --now forgejo-runner'
```
## Evidence
- Host probe (2026-07-03): `coulomb/forgejo-actions-probe` workflow `success`.
- In-cluster: verify with `make forgejo-runner-status` and a container `runs-on: ubuntu-latest` job.
## References
- `railiance-infra/docs/adr/ADR-004-forgejo-in-cluster-actions-runner.md`
- [Forgejo runner installation](https://forgejo.org/docs/v11.0/admin/actions/runner-installation/)