cluster/docker-compose.yml

version: '3.8'

services:
  nginx:
    image: nginx:alpine
    container_name: jirack-nginx
    ports:
      - "7869:80"
    volumes:
      - ./nginx.conf:/etc/nginx/nginx.conf:ro
    restart: unless-stopped
    depends_on:
      - jirack1
      - jirack2
      - jirack3
      - jirack4
      - jirack5
    networks:
      - jirack_net

  jirack1:
    image: cmsmanhattan/jirack_10b_int8:latest
    container_name: jirack-1
    restart: unless-stopped
    volumes:
      - model_data:/app/model:ro
    networks:
      - jirack_net
    deploy:
      resources:
        limits:
          cpus: '2.0'      # максимум 2 ядра
          memory: 32G        
        reservations:
          cpus: '0.8'
          memory: 24G

  jirack2:
    image: cmsmanhattan/jirack_10b_int8:latest
    container_name: jirack-2
    restart: unless-stopped
    volumes:
      - model_data:/app/model:ro
    networks:
      - jirack_net
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 32G
        reservations:
          cpus: '0.8'
          memory: 24G

  jirack3:
    image: cmsmanhattan/jirack_10b_int8:latest
    container_name: jirack-3
    restart: unless-stopped
    volumes:
      - model_data:/app/model:ro
    networks:
      - jirack_net
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 32G
        reservations:
          cpus: '0.8'
          memory: 24G

  jirack4:
    image: cmsmanhattan/jirack_10b_int8:latest
    container_name: jirack-4
    restart: unless-stopped
    volumes:
      - model_data:/app/model:ro
    networks:
      - jirack_net
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 32G
        reservations:
          cpus: '0.8'
          memory: 24G

  jirack5:
    image: cmsmanhattan/jirack_10b_int8:latest
    container_name: jirack-5
    restart: unless-stopped
    volumes:
      - model_data:/app/model:ro
    networks:
      - jirack_net
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 32G
        reservations:
          cpus: '0.8'
          memory: 24G

networks:
  jirack_net:
    driver: bridge

volumes:
  model_data:
    driver: local