# Disaster Recovery: Git + Docs stack

Goal: restore `git.frostlabs.me` and the docs sites after losing the VPS.

## What state actually matters

| Thing | Where | Rebuildable? |
|---|---|---|
| Gitea data (repos, DB, config) | `/opt/gitea/gitea/` (SQLite + repos) | **No — back this up** |
| Runner registration | `/opt/gitea/runner/.runner` | No (but re-registerable in minutes) |
| Built sites | `/srv/www/` | Yes — rebuilt on next push |
| Tunnel / nginx / compose config | `/opt/gitea/*.yml`, `.env`, `nginx.conf` | Keep a copy off-box |

The docs repos themselves are safe as long as one clone survives (VPS, homelab,
laptop). This runbook is about restoring the **server**.

## Backup (do this on a schedule)

```bash
# Off-box snapshot of everything stateful on the VPS
rsync -aP /opt/gitea/ backup-target:/backups/gitea/
```

!!! tip
    A `gitea dump` is the "proper" way, but for a single-user SQLite instance an
    rsync of `/opt/gitea` while the stack is stopped is simplest and complete.

## Restore onto a fresh VPS

1. Provision the box, install Docker + compose.
2. Recreate `/opt/gitea/` and restore the backup into it.
3. Recreate `.env` with `TUNNEL_TOKEN` and (optionally) `RUNNER_TOKEN`.
4. `mkdir -p /srv/www/docs-public /srv/www/docs-private`
5. `docker compose up -d`
6. In Cloudflare Zero Trust, point the tunnel at the new box (token is the same
   tunnel, so hostnames just reattach). Confirm `git.frostlabs.me` loads.
7. If the runner didn't reconnect, re-register it (Site Admin → Actions → Runners).
8. Push any repo (or hit **Run workflow**) to rebuild both sites into `/srv/www`.

## Verify

- [ ] `git.frostlabs.me` reachable, repos present
- [ ] `runbook.frostlabs.me` rebuilt and gated by CF Access
- [ ] `docs.frostlabs.me` rebuilt and public
- [ ] a test push triggers a green Actions run