auto-stock/docker-compose.yml

services:
  # PostgreSQL Database
  postgres:
    image: postgres:15-alpine
    container_name: autostock-postgres
    environment:
      TZ: Asia/Seoul
      POSTGRES_USER: ${POSTGRES_USER}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
      POSTGRES_DB: ${POSTGRES_DB}
    volumes:
      - postgres_data:/var/lib/postgresql/data
      - ./docker/init-db.sql:/docker-entrypoint-initdb.d/init.sql
    ports:
      - "127.0.0.1:15433:5432"
    networks:
      - autostock-network
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER}"]
      interval: 10s
      timeout: 5s
      retries: 5
    deploy:
      resources:
        limits:
          memory: 1G

  # Redis Cache
  redis:
    image: redis:7-alpine
    container_name: autostock-redis
    # [L-7 2026-10-02] allkeys-lru → volatile-lru: TTL 없는 상태 키(mdd:state·adaptive 이력)가 축출되면
    # MDD peak/플래그가 재시작 없이 사라진다. 캐시 키·주문 락은 전부 TTL 이 있다(setex_json·_cache_set·set(ex=) 전수 확인).
    # 주의: TTL 키가 없는 상태로 200mb 에 닿으면 쓰기가 OOM 오류가 된다(현재 데이터 규모상 비현실적).
    command: redis-server --requirepass ${REDIS_PASSWORD:-autostock_redis_change_me} --maxmemory 200mb --maxmemory-policy volatile-lru
    environment:
      TZ: Asia/Seoul
    ports:
      - "127.0.0.1:6379:6379"
    volumes:
      - redis_data:/data
    networks:
      - autostock-network
    healthcheck:
      test: ["CMD", "redis-cli", "-a", "${REDIS_PASSWORD:-autostock_redis_change_me}", "ping"]
      interval: 10s
      timeout: 3s
      retries: 5
    deploy:
      resources:
        limits:
          memory: 256M

  # RabbitMQ Message Queue
  rabbitmq:
    image: rabbitmq:3-management
    container_name: autostock-rabbitmq
    environment:
      TZ: Asia/Seoul
      RABBITMQ_DEFAULT_USER: ${RABBITMQ_DEFAULT_USER}
      RABBITMQ_DEFAULT_PASS: ${RABBITMQ_DEFAULT_PASS}
    ports:
      - "127.0.0.1:5672:5672"
      - "127.0.0.1:15672:15672"
    volumes:
      - rabbitmq_data:/var/lib/rabbitmq
    # - rabbitmq.conf:/etc/rabbitmq/rabbitmq.conf
    networks:
      - autostock-network
    healthcheck:
      test: rabbitmq-diagnostics -q ping
      interval: 30s
      timeout: 10s
      retries: 5
    deploy:
      resources:
        limits:
          memory: 512M

  # MCP Server
  mcp-server:
    build:
      context: ./mcp-server
      dockerfile: Dockerfile
    container_name: autostock-mcp-server
    env_file:
      - .env
    environment:
      - TZ=Asia/Seoul
    ports:
      - "127.0.0.1:8001:8001"
    networks:
      - autostock-network
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
    restart: unless-stopped
    deploy:
      resources:
        limits:
          memory: 256M

  # Backend API Server
  backend:
    build:
      context: ./backend
      dockerfile: Dockerfile
    container_name: autostock-backend
    # [H-5 2026-10-02] 기본 10초 < db_worker 종료 drain 예산(DRAIN_DEADLINE_SECONDS=20).
    # SIGTERM 후 10초에 SIGKILL 이 와서 drain 중인 DB 쓰기가 유실됐다.
    # [M2] 예산 = uvicorn graceful 5 + 백그라운드 대기 8 + drain 20 = 33초 (+여유) → 40s. backend/Dockerfile CMD 주석 참조.
    stop_grace_period: 40s
    env_file:
      - .env
    environment:
      - TZ=Asia/Seoul
      - MCP_SERVER_URL=http://mcp-server:8001
    ports:
      - "127.0.0.1:8000:8000"
    volumes:
      - ./logs:/logs
      # KIS 액세스 토큰 캐시 — 컨테이너 재생성 때마다 사라지면 매 배포마다 토큰을 새로
      # 발급받게 되고, KIS가 잦은 재발급을 거부한다(EGW00103 "유효하지 않은 AppKey").
      - ./token_cache:/root/.cache/auto-stock
    networks:
      - autostock-network
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
      mcp-server:
        condition: service_started
    # [R2 M] 프로세스 밖 생존 감시: 이벤트 루프 hang / uvicorn 무응답을 컨테이너 상태(unhealthy)로 노출.
    # 판정은 "10초 내 HTTP 200 응답"만 본다 — /health 의 status=degraded(DB 순간 장애·장외 stale)로
    # unhealthy 가 되면 오탐이 잦다. degraded/schedulers.stale 감시는 앱 밖 cron 의 몫:
    # scripts/health_watch.py (평일 장중 5분 cron, 연속 이상 시 텔레그램) + .claude/scripts/prod_smoke.sh 의 /health 확인.
    # 위 cron 대신 compose 사이드카 health-watch 가 같은 스크립트를 돌린다(NAS cron 등록 불필요).
    # (plain compose 는 unhealthy 컨테이너를 재시작하지 않는다 — 경보·수동 확인용 표식)
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request,sys; sys.exit(0 if urllib.request.urlopen('http://127.0.0.1:8000/health', timeout=10).status == 200 else 1)"]
      interval: 60s
      timeout: 15s
      retries: 3
      start_period: 120s
    restart: unless-stopped
    deploy:
      resources:
        limits:
          memory: 768M

  # PostgreSQL 자동 백업 (매일 02:00 KST, 7일치 보관)
  postgres-backup:
    image: prodrigestivill/postgres-backup-local:15
    container_name: autostock-postgres-backup
    environment:
      - TZ=Asia/Seoul
      - POSTGRES_HOST=postgres
      - POSTGRES_DB=${POSTGRES_DB}
      - POSTGRES_USER=${POSTGRES_USER}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
      - SCHEDULE=0 8,17 * * *
      - BACKUP_KEEP_DAYS=7
      - BACKUP_KEEP_WEEKS=4
      - BACKUP_KEEP_MONTHS=3
      - HEALTHCHECK_PORT=8080
    volumes:
      - ./backups:/backups
    networks:
      - autostock-network
    depends_on:
      postgres:
        condition: service_healthy
    restart: unless-stopped

  # /health 능동 감시 사이드카 — NAS cron 등록 여부에 의존하지 않도록 compose 안에서 직접 돈다.
  # 평일(KST) 09~15시 5분 주기로 scripts/health_watch.py 실행(상태 전이 시 텔레그램). 장외는 stale 오탐 방지로 건너뜀.
  # 상태 파일은 named volume 에 유지(재기동해도 연속 이상 카운트 보존).
  # 마지막 실행 시각: `docker exec autostock-health-watch cat /state/last_run` (배포 스모크에서 mtime 확인용).
  health-watch:
    image: python:3.11-slim
    container_name: autostock-health-watch
    environment:
      - TZ=Asia/Seoul
      - HEALTH_URL=http://backend:8000/health
      - HEALTH_WATCH_STATE=/state/health_watch.json
      - TELEGRAM_BOT_TOKEN=${TELEGRAM_BOT_TOKEN:-}
      - TELEGRAM_CHAT_ID=${TELEGRAM_CHAT_ID:-}
    volumes:
      - ./scripts/health_watch.py:/app/health_watch.py:ro
      - health_watch_state:/state
    command:
      - sh
      - -c
      - |
        while true; do
          k=$$(date -u -d '+9 hours' '+%u %H')
          dow=$${k% *}; hh=$${k#* }
          if [ "$$dow" -le 5 ] && [ "$$((10#$$hh))" -ge 9 ] && [ "$$((10#$$hh))" -le 15 ]; then
            python /app/health_watch.py || true
            date -u '+%Y-%m-%dT%H:%M:%SZ' > /state/last_run
          fi
          sleep 300
        done
    networks:
      - autostock-network
    depends_on:
      - backend
    restart: unless-stopped
    deploy:
      resources:
        limits:
          memory: 64M

  # Frontend
  frontend:
    build:
      context: ./frontend
      dockerfile: Dockerfile
      args:
        - VITE_API_URL=${VITE_API_URL}
        - VITE_WS_URL=${VITE_WS_URL}
    container_name: autostock-frontend
    environment:
      - TZ=Asia/Seoul
    ports:
      - "127.0.0.1:3001:80"
    networks:
      - autostock-network
    depends_on:
      - backend
    restart: unless-stopped
    deploy:
      resources:
        limits:
          memory: 64M

volumes:
  postgres_data:
  redis_data:
  rabbitmq_data:
  backup_data:
  health_watch_state:

networks:
  autostock-network:
    driver: bridge