forgejo-runner: commit /opt/forgejo-runner/{docker-compose.yml,data/config.yaml} to git (infra-as-code) #531

New issue

Closed

opened 2026-05-28 00:05:52 +02:00 by claude · 0 comments

claude commented

2026-05-28 00:05:52 +02:00

Collaborator

Context

During pdurlej/platform#527 smoke validation I had to hand-edit two files on rs2000:

/opt/forgejo-runner/docker-compose.yml — added -c /data/config.yaml to the runner command so the daemon picks up the config file explicitly (default search behaviour was inconsistent).
/opt/forgejo-runner/data/config.yaml — new file containing:
```
log:
  level: info
runner:
  capacity: 1
container:
  options: --volume /opt/forgejo-runner/data/platformctl:/data/platformctl:ro
  valid_volumes:
    - /opt/forgejo-runner/data/platformctl
```
(The valid_volumes entry is required — forgejo-runner v7.0.0 silently drops any --volume in container.options that isn't whitelisted there. This is documented behaviour, not a bug; see the v7.0.0 config example. We lost ~2 hours rediscovering it.)

These changes survive container restarts but don't survive a re-provisioning of the runner host. There's no git record of what state the runner expects, and there's a docker-compose.yml.bak next to it now.

Proposed fix

Commit both files into the platform repo under, say, infra/forgejo-runner/:

infra/forgejo-runner/
├── docker-compose.yml
├── config.yaml        # the runner config (no secrets)
└── README.md          # 5-line operator runbook: where to copy, how to restart

The README should make explicit:

container.options and container.valid_volumes must agree, or volumes are silently dropped (link to v7.0.0 config example).
Restarting the runner requires cd /opt/forgejo-runner && docker compose restart runner.
canary.env is generated separately by write_canary_env.py (cross-link to the sister issue about OLLAMA specs).

Then a runbooks/forgejo-runner-rebuild.md or similar that says "scp these files into place on a fresh host, then run docker compose up -d".

Out of scope

Automating the runner host bootstrap (Ansible / Nix etc.) — that's a bigger discussion. This issue is just "stop the bus factor of these files living only on rs2000".
Adding forgejo-canary-runner (the systemd unit) — that's a separate runner using a different runtime, not affected by this fix.

Validation

After the change, verify git diff between the committed files and /opt/forgejo-runner/... shows no drift.

Refs: pdurlej/platform#524 #527, https://code.forgejo.org/forgejo/runner/raw/tag/v7.0.0/internal/pkg/config/config.example.yaml

## Context During `pdurlej/platform#527` smoke validation I had to hand-edit two files on `rs2000`: 1. `/opt/forgejo-runner/docker-compose.yml` — added `-c /data/config.yaml` to the runner command so the daemon picks up the config file explicitly (default search behaviour was inconsistent). 2. `/opt/forgejo-runner/data/config.yaml` — new file containing: ```yaml log: level: info runner: capacity: 1 container: options: --volume /opt/forgejo-runner/data/platformctl:/data/platformctl:ro valid_volumes: - /opt/forgejo-runner/data/platformctl ``` (The `valid_volumes` entry is **required** — `forgejo-runner` v7.0.0 silently drops any `--volume` in `container.options` that isn't whitelisted there. This is documented behaviour, not a bug; see the v7.0.0 config example. We lost ~2 hours rediscovering it.) These changes survive container restarts but **don't survive a re-provisioning of the runner host**. There's no git record of what state the runner expects, and there's a `docker-compose.yml.bak` next to it now. ## Proposed fix Commit both files into the platform repo under, say, `infra/forgejo-runner/`: ``` infra/forgejo-runner/ ├── docker-compose.yml ├── config.yaml # the runner config (no secrets) └── README.md # 5-line operator runbook: where to copy, how to restart ``` The README should make explicit: - `container.options` and `container.valid_volumes` must agree, or volumes are silently dropped (link to v7.0.0 config example). - Restarting the runner requires `cd /opt/forgejo-runner && docker compose restart runner`. - `canary.env` is generated separately by `write_canary_env.py` (cross-link to the sister issue about OLLAMA specs). Then a `runbooks/forgejo-runner-rebuild.md` or similar that says "scp these files into place on a fresh host, then run docker compose up -d". ## Out of scope - Automating the runner host bootstrap (Ansible / Nix etc.) — that's a bigger discussion. This issue is just "stop the bus factor of these files living only on rs2000". - Adding `forgejo-canary-runner` (the systemd unit) — that's a separate runner using a different runtime, not affected by this fix. ## Validation After the change, verify `git diff` between the committed files and `/opt/forgejo-runner/...` shows no drift. Refs: `pdurlej/platform#524 #527`, https://code.forgejo.org/forgejo/runner/raw/tag/v7.0.0/internal/pkg/config/config.example.yaml

claude added the

ready-for-agent

tier/lite

labels

2026-05-28 00:05:52 +02:00