Jellyfin 10.8.13 running as docker swarm service WITH hardware acceleration

gistfile1.txt

# Step 1: device nodes
Use [this project](https://github.com/allfro/device-mapping-manager) to brute force device nodes into running containers.

Do this by running it as a docker swarm service like so:

`docker-compose.yml`:

```
services:
  device-mapping-manager:
    image: alpinelinux/docker-cli
    entrypoint: docker
    command: |
      run
      --rm
      -i
      --name device-manager
      --privileged
      --cgroupns=host
      --pid=host
      --userns=host
      -v /sys:/host/sys
      -v /var/run/docker.sock:/var/run/docker.sock
      ghcr.io/allfro/allfro/device-mapping-manager:sha-0651661
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro
    deploy:
      mode: global
      restart_policy:
        condition: any
      placement:
        constraints:
          - node.labels.dmm == true
```
The placement constraint is optional, I wanted to only have it on specific workder nodes, maybe you want it somewhere else.

# Step 2: Jellyfin as a swarm service

Note this is just a cut-down sample to illustrate a bare-bones working sample.
You will/might need to adjust some values, like:
- timezone
- JELLYFIN_PublishedServerUrl
- media bind mount volumes
- jellyfin data volumes
- port mappings
- resource contraints (yes, jellyfin CAN run with a limit of 768Mb, but YMMV, adjust as needed and your hardware allows).

The important part for HW accel is `- /dev/dri:/dev/dri/`, under the volumes section.
If using plain docker-compose, you can achieve the same with the `devices` key, BUT with swarm, that capability has never matured.

So we have this work-a-round.

```
services:
  jellyfin:
    image: jellyfin/jellyfin:10.8.13
    healthcheck:
      interval: 30s
      timeout: 10s
      retries: 60
      start_period: 30m
    ports:
      - 7359:7359/udp
      - 8096:8096
    deploy:
      placement:
        constraints:
          - node.labels.dmm == true
      restart_policy:
        condition: on-failure
        delay: 60s
        window: 120s
      resources:
        limits:
          cpus: '3.0'
          memory: 768M
    environment:
      - TZ=YourCountry/YourCity
      - PUID=1000
      - PGID=1000
      - JELLYFIN_PublishedServerUrl=your.hostname.here
    volumes:
      - /media/mymedia:/mymedia:ro
      - /myjellyfin/cache:/cache
      - /myjellyfin/config:/config
      - /dev/dri:/dev/dri/ #This is the important part for hw accel!
```

# Notes
You can possibly opt to combine the two services into a single stack, if you have NOTHING else that requires device access.
I have not tested this however, there might be a race condition situation (?)