services:
  server:
    restart: always
    build:
      context: .
    volumes:
      - ${MODELS_PATH}:/models:ro
    environment:
      DEVICE: cuda
      ENABLE_TABLE_RECOG: "true"
      MAX_TIME: 400
    shm_size: 10gb
    # TODO
    # healthcheck:
    #   test: ["CMD", "curl", "-f", "http://localhost:8000/v1/health"]
    #   interval: 1m30s
    #   timeout: 500s # may block for a long time
    #   retries: 5
    #   start_period: 30s
    deploy:
      mode: replicated
      replicas: 4
      endpoint_mode: dnsrr
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    depends_on:
      - caddy

  caddy:
    restart: always
    image: public.ecr.aws/docker/library/caddy
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile:ro
    ports:
      - "127.0.0.1:8001:8000"