docker-compose.yml

# docker-compose.yml - Chandra OCR 2 전체 스택
# 위치: ~/projects/chandra/docker/docker-compose.yml

# --mm-processor-kwargs '{"max_pixels": 1003520, "min_pixels": 100352}'
services:
  # ============================================
  # Chandra vLLM 추론 서버 (GPU)
  # ============================================
  vllm-server:
    image: vllm/vllm-openai:v0.17.0
    container_name: chandra-vllm
    restart: unless-stopped
    ipc: host
    # 외부 포트 노출 불필요 (FastAPI가 같은 compose 네트워크에서 접근)
    # 디버깅 필요 시 아래 주석 해제
    # ports:
    #   - "8000:8000"
    volumes:
      - ~/projects/models/chandra-ocr-2:/models/chandra-ocr-2:ro
    environment:
      - HF_HUB_OFFLINE=1
      - TRANSFORMERS_OFFLINE=1
    command: >
      --model /models/chandra-ocr-2
      --served-model-name chandra
      --dtype bfloat16
      --max-model-len 12384
      --max-num-seqs 64
      --max-num-batched-tokens 8192
      --enable-prefix-caching
      --mm-processor-kwargs '{"max_pixels": 2007040, "min_pixels": 100352}'
      --gpu-memory-utilization 0.85
      --trust-remote-code
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 180s    # 모델 로딩 대기 (3분)

  # ============================================
  # FastAPI 서비스 (기존 인터페이스 유지)
  # ============================================
  fastapi:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: chandra-fastapi
    restart: unless-stopped
    ports:
      - "10001:10001"
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock    # LibreOffice 컨테이너 접근용
      - ./uploads:/app/uploads
    environment:
      - CHANDRA_VLLM_IP=vllm-server
      - CHANDRA_VLLM_PORT=8000
    depends_on:
      vllm-server:
        condition: service_healthy