# Running Dropwatch in Docker (Unraid)

One container is one miner. Never run two against the same Twitch account (a PC copy and a container,
or two containers): they fight over watch progress.

## Build

The image needs the commit it was built from. Without it every build looks the same, so the page's
"Dropwatch was updated" notice and the copy of the database before an upgrade both stop working, and
the build refuses to run:

```
docker build --build-arg DROPWATCH_COMMIT=main@$(git rev-parse --short HEAD) -t dropwatch:latest .
```

## Run

```
docker run -d --name dropwatch --restart unless-stopped \
  -p 5800:5800 -v /mnt/user/appdata/dropwatch:/data \
  -e TZ=America/New_York \
  --log-opt max-size=10m --log-opt max-file=3 \
  dropwatch:latest
```

| Setting | What it does |
|---|---|
| `--restart unless-stopped` | **Required.** When mining crashes or stops ticking for two checks in a row, Dropwatch exits on purpose and the restart policy starts a fresh one. Without a restart policy it stays stopped. |
| `/data` | The database (`drops.db`: Twitch sign-in, lists, history), the campaign catalogue and kept images. Owned by `PUID:PGID`. |
| `PUID` / `PGID` | The user the miner runs as; default `99:100` (Unraid's nobody:users), so the host can read `/data`. |
| `TZ` | The server's time zone (the weekly summary's default zone and log times). Default UTC. |
| `DROPWATCH_HOSTS` | Only for a public domain name, such as a reverse proxy's (`dropwatch.example.com`), comma-separated. IP addresses, `localhost`, bare server names (`tower`) and private suffixes (`.local`, `.lan`, `.home`, `.home.arpa`, `.internal`, `.localdomain`) always work. Other names are refused so a web page can't reach Dropwatch through DNS rebinding. `*` turns the check off. |
| `FORWARDED_ALLOW_IPS` | Behind a reverse proxy, the proxy's address, so the login lockout sees each visitor's own address instead of the proxy's (otherwise five wrong guesses through the proxy lock everyone out). |
| `--log-opt` | Docker keeps the miner's log (it writes to standard output); this caps it at 3 files of 10 MB. |

## Health

The image's health check calls `GET /api/health/miner`, which answers 503 when the mining loop crashed or
hasn't ticked for 15 minutes (longer than any back-off). Signed out, paused or idle are healthy.
Unraid shows the status; the restart itself comes from the watchdog plus the restart policy above.

## Upgrades

On start, a new build copies `drops.db` to `drops.db.before-<build>` before touching it (the two newest
copies are kept). To roll back: stop the container, copy the file back over `drops.db`, run the old image.