Scalable Docker Swarm compose file for Immich, workloads split between GPU and CPU nodes, each container running as non-root

immich.yml

#
# WARNING: To install Immich, follow our guide: https://immich.app/docs/install/docker-compose
#
# Make sure to use the docker-compose.yml of the current release:
#
# https://github.com/immich-app/immich/releases/latest/download/docker-compose.yml
#
# The compose file on main may not be compatible with the latest release.

services:
  immich-server:
    image: ghcr.io/immich-app/immich-server:release
    user:  2000001:2000001
    volumes:
      - immich-upload:/data    #Main Immich data
      - archives:/archives:ro  #External Library
      - /etc/localtime:/etc/localtime:ro
    env_file:
      - immich.env
    environment:
      IMMICH_WORKERS_INCLUDE: 'api'
#Exposed via traefik, no need to expose directly
#    ports:
#      - '2283:2283'
    depends_on:
      - redis
    healthcheck:
      disable: false
    networks:
      - traefik_proxy
      - immich
    deploy:
      mode: global
#      placement:
#        constraints:
#          - node.role == manager
        # - node.labels.gpu == true
      resources:
        reservations:
          memory: 2G
          cpus: '4'
      labels:
        - "traefik.enable=true"
        - "traefik.http.routers.immich-ui-https.tls.certresolver=letsencrypt"
        - "traefik.http.routers.immich-ui-https.rule=Host(`externalhostname.example.com`) || Host(`internalhostname.example.com`)"
        - "traefik.http.routers.immich-ui-https.entrypoints=websecure"
        - "traefik.http.routers.immich-ui-https.service=immich-ui"
        - "traefik.http.services.immich-ui.loadbalancer.server.port=2283"

  immich-server-microservices:
    image: ghcr.io/immich-app/immich-server:release
    user:  2000001:2000001
    volumes:
      - immich-upload:/data
      - archives:/archives:ro
      - /etc/localtime:/etc/localtime:ro
    env_file:
      - immich.env
    environment:
      IMMICH_WORKERS_EXCLUDE: 'api'
    depends_on:
      - redis
    # restart: always
    healthcheck:
      disable: false
    networks:
      - traefik_proxy
      - immich
    deploy:
      mode: global
      placement:
        constraints:
          - node.role == worker
          - node.labels.gpu == true
      resources:
        reservations:
          memory: 4G
          cpus: '4'
            generic_resources:
              - discrete_resource_spec:
                 kind: 'NVIDIA-GPU-ImmichServer'
                 value: 0

#Cuda and CPU machine learning containers both aliased to "immich-ml"
#Configure ML endpoint in Immich to http://immich-ml:3003 and it will loadbalance to both
  immich-machine-learning-cuda:
    image: ghcr.io/immich-app/immich-machine-learning:release-cuda
    user:  2000001:2000001
    volumes:
      - model-cache:/cache
    env_file:
      - immich.env
    healthcheck:
      disable: false
    networks:
      immich:
        aliases:
          - immich-ml
    deploy:
      mode: replicated
      replicas: 1
      placement:
        constraints:
        - node.labels.gpu == true
      resources:
        reservations:
          generic_resources:
            - discrete_resource_spec:
                kind: 'NVIDIA-GPU-ImmichML'
                value: 0

  immich-machine-learning-cpu:
    # container_name: immich_machine_learning
    # For hardware acceleration, add one of -[armnn, cuda, rocm, openvino, rknn] to the image tag.
    # Example tag: release-cuda
    image: ghcr.io/immich-app/immich-machine-learning:release
    user:  2000001:2000001
    volumes:
      - model-cache:/cache
    env_file:
      - immich.env
    healthcheck:
      disable: false
    networks:
      immich:
        aliases:
          - immich-ml
    deploy:
      mode: replicated
      replicas: 1
      placement:
        constraints:
          - node.role == worker

  redis:
    image: docker.io/valkey/valkey:8-alpine
    user:  2000001:2000001
    command:
      - "valkey-server"
      - "--io-threads 6"
    healthcheck:
      test: valkey-cli ping || exit 1
    networks:
      - immich
    volumes:
      - immich-redis:/data
    deploy:
      mode: replicated
      replicas: 1
      placement:
        constraints:
        - node.role == worker

volumes:
 immich-upload:
  driver: local
  driver_opts:
     type: "nfs"
     o: ""
     device: ""
 immich-redis:
  driver: local
  driver_opts:
     type: "nfs"
     o: ""
     device: ""
 archives:
  driver: local
  driver_opts:
     type: "nfs"
     o: ""
     device: ""
 model-cache:
  driver: local
  driver_opts:
     type: "nfs"
     o: ""
     device: ""

networks:
  immich:
  traefik_proxy:
    external: true