# Containerized startup (C2 + worker)

One command brings up the cs.money **C2** (control plane) and a **worker**. Postgres
runs independently on the host; the C2 connects to it and auto-applies EF migrations
on boot.

```powershell
docker-compose up --build
```

- **C2** → http://localhost:5080 (`/health`, `/jobs/*`, `/market/*`)
- **Worker noVNC** → http://localhost:6080/vnc.html — watch the browser, and solve a
  Cloudflare challenge by hand if one appears.

## Prerequisites

1. **Host Postgres reachable from containers.** The C2 connects via
   `host.docker.internal`. Postgres must (a) listen on the Docker-facing interface
   (`listen_addresses = '*'` in `postgresql.conf`) and (b) allow the container subnet
   in `pg_hba.conf`. The DB (`skintracker`) should already have the schema, but the
   C2 also runs `Database.Migrate()` at startup as a safety net.
2. **A real exit IP for the worker.** A bare datacenter/container IP gets challenged
   hard by Cloudflare. Set `PROXY` to a residential exit (see below).

## Configuration (env vars / a `.env` file next to docker-compose.yml)

| Var | Default | Purpose |
|-----|---------|---------|
| `SKINTRACKER_CONN` | `Host=host.docker.internal;Port=5432;Database=skintracker;Username=postgres` | C2 → Postgres connection string |
| `WORKER_TOKEN` | `dev-worker-token` | Shared secret; C2 and worker must match |
| `PROXY` | _(none)_ | Worker proxy `host:port` (auth-free) |
| `SOLVE_SECONDS` | `45` | Time the worker waits for you to clear Cloudflare |
| `MAX_PAGES_PER_JOB` | `20` | Cap on offset pages per skin+wear job |
| `LOAD_IMAGES` | _(off)_ | `1` re-enables image loading (debugging) |

## Scaling workers

```powershell
docker-compose up --build --scale worker=3
```

Remove the worker `ports:` mapping first — multiple workers can't share host port 6080
for noVNC. (Each gets the display internally; expose per-worker only if you need to
watch a specific one.)

## Notes / known gaps

- **IPRoyal auth:** `PROXY` is passed to Chromium as `--proxy-server`, which ignores
  `user:pass`. For credentialed IPRoyal either IP-whitelist the worker's egress IP, or
  add a small forwarding-proxy sidecar that injects the auth (the .NET
  `LocalForwardingProxy` does this for the CSFloat path; a worker-side equivalent is a
  follow-up).
- **Unattended Cloudflare:** the worker leans on nodriver + a residential IP clearing
  CF automatically. When it can't, use the noVNC tab to solve it once; the warmed
  profile then carries the clearance.