production/docker-compose.m2.yml

version: "3.8"

# Optimized for MacBook M2 Pro Max 96GB with Gemma 4 support
# - MLX local inference runs natively on Metal (host, not in Docker)
# - Gemma 4 31B-BF16 fits comfortably in 96GB unified memory
# - Redis, Postgres, API server in lightweight containers
# - NIM → Cloudflare → Gemini → MLX fallback chain
# - Everything runs natively on Apple Silicon

services:
  api:
    build:
      context: .
      dockerfile: Dockerfile.prod
    ports:
      - "8000:8000"
    environment:
      - PORT=8000
      - WORKERS=2
      - REDIS_URL=redis://redis:6379
      - DATABASE_URL=postgresql://ml_intern:ml_intern@postgres:5432/ml_intern
      - MAX_CONCURRENT_REQUESTS=100
      - DEFAULT_RPM_LIMIT=40
      - REQUEST_TIMEOUT=120
      - CACHE_TTL_SECONDS=300
      - BUDGET_USD_PER_SESSION=10.0
      - CIRCUIT_BREAKER_FAILURE_THRESHOLD=3
      - CIRCUIT_BREAKER_RECOVERY_TIMEOUT=30
      # Fallback chain: NIM → Cloudflare → Gemini → MLX
      - FALLBACK_ENABLED=true
      - FALLBACK_PRIMARY=nim
      - FALLBACK_SECONDARY=cloudflare
      - FALLBACK_TERTIARY=gemini
      # NIM primary (cloud GPU)
      - NIM_API_BASE=https://integrate.api.nvidia.com/v1
      - NVIDIA_API_KEY=${NVIDIA_API_KEY:-}
      # Cloudflare secondary fallback
      - CLOUDFLARE_API_KEY=${CLOUDFLARE_API_KEY:-}
      - CLOUDFLARE_ACCOUNT_ID=${CLOUDFLARE_ACCOUNT_ID:-}
      # Google Gemini tertiary fallback
      - GEMINI_API_KEY=${GEMINI_API_KEY:-}
      - GEMINI_API_BASE=https://generativelanguage.googleapis.com/v1beta/openai
      # MLX local (runs on host Metal, not in container)
      - MLX_ENABLED=${MLX_ENABLED:-false}
      - MLX_API_BASE=http://host.docker.internal:8000/v1
      - LOG_LEVEL=INFO
    depends_on:
      redis:
        condition: service_healthy
      postgres:
        condition: service_healthy
    networks:
      - ml_intern_network
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    restart: unless-stopped

  worker:
    build:
      context: .
      dockerfile: Dockerfile.prod
    command: ["python", "-m", "production.worker"]
    environment:
      - REDIS_URL=redis://redis:6379
      - DATABASE_URL=postgresql://ml_intern:ml_intern@postgres:5432/ml_intern
      - LOG_LEVEL=INFO
    depends_on:
      redis:
        condition: service_healthy
      postgres:
        condition: service_healthy
    networks:
      - ml_intern_network
    restart: unless-stopped

  redis:
    image: redis:7-alpine
    ports:
      - "127.0.0.1:6379:6379"
    volumes:
      - redis_data:/data
    command: redis-server --appendonly yes --maxmemory 512mb --maxmemory-policy allkeys-lru
    networks:
      - ml_intern_network
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 10s
      timeout: 3s
      retries: 3
    restart: unless-stopped

  postgres:
    image: postgres:16-alpine
    ports:
      - "127.0.0.1:5432:5432"
    environment:
      - POSTGRES_USER=ml_intern
      - POSTGRES_PASSWORD=ml_intern
      - POSTGRES_DB=ml_intern
    volumes:
      - postgres_data:/var/lib/postgresql/data
      - ./init.sql:/docker-entrypoint-initdb.d/init.sql:ro
    networks:
      - ml_intern_network
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ml_intern"]
      interval: 10s
      timeout: 3s
      retries: 5
    restart: unless-stopped

  nginx:
    image: nginx:alpine
    ports:
      - "80:80"
    volumes:
      - ./nginx.conf:/etc/nginx/nginx.conf:ro
    depends_on:
      - api
    networks:
      - ml_intern_network
    restart: unless-stopped

volumes:
  redis_data:
  postgres_data:

networks:
  ml_intern_network:
    driver: bridge