# `forgejo_runner`

Installs and runs a Forgejo Actions runner, registers it with the Forgejo
instance, and keeps a health check on a systemd timer.

Converted from `deploy_forgejo_runner_playbook.yml` (409 lines) under Plan 6.
The playbook is now 16 lines.

## Phases

`tasks/main.yml` imports five files in order:

| | |
|---|---|
| `prerequisites.yml` | Docker must be present |
| `install.yml` | binary, system user, working directory |
| `configure.yml` | config file, registration with the instance |
| `service.yml` | systemd unit, start, assert it came up |
| `healthcheck.yml` | check script, unit, timer |

`import_tasks`, not `include_tasks` — static imports are visible to
`--list-tasks`, which is how the conversion was verified against the playbook it
replaced.

## Monitoring: one variable, no product knowledge

This role contains **nothing specific to any monitoring system**. What used to
be here — an ~80-line embedded Python script creating monitors over the Uptime
Kuma API, a `/tmp` credentials file, token extraction, a systemd `Environment=`
rewrite, and 8 `when: uptime_kuma_enabled` guards — is gone.

What remains answers the actual question, *is this service healthy*, and records
it two ways:

- **the exit code**, which systemd keeps: `systemctl is-failed
  forgejo-runner-healthcheck.service` is a complete answer with no monitoring
  system involved at all;
- **a log file** at `{{ healthcheck_log_file }}`.

To report health somewhere, set one variable:

```yaml
healthcheck_push_url: "https://example/api/push/TOKEN"
```

Any endpoint accepting an HTTP ping works. Empty (the default) means check, log,
exit honestly, report nowhere — which is also the right setting for a *pull*-based
monitor like Prometheus' textfile collector, since that reads unit state instead.

The push URL is a credential (anyone holding it can forge an "up"), so callers
pass it from the vault rather than committing it.

## One behaviour change, deliberate

`Assert runner is running` used to be guarded by `uptime_kuma_enabled`, so it
never ran. It is not a monitoring task — it is the deployment checking its own
work — and the deprecation banner swept it up by mistake. It is ungated here,
which means a runner that fails to start now fails the play instead of
deploying "successfully" in silence.