services:
  # ============================================
  # 경량 버전 (외부 API 사용 + 로컬 MongoDB)
  # ============================================
  
  # 1) 임베딩 단계 (preprocess 생략, qa_raw.jsonl 직접 사용)
  embed:
    image: "${RAG_BATCH_IMAGE:-rag-batch-slim:latest}"
    network_mode: host
    build:
      context: .
      dockerfile: docker/batch-slim.Dockerfile
    command: ["python", "/app/scripts/ingest_qa.py"]
    working_dir: /app
    depends_on:
      - qdrant
    # 모델 게이트웨이 호스트명 → IP 매핑 (DNS 미등록 대응)
    extra_hosts:
      - "llm-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
      - "embedding-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
      - "reranker-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
    environment:
      # 프록시 비활성화 (명시적)
      HTTP_PROXY: ""
      HTTPS_PROXY: ""
      http_proxy: ""
      https_proxy: ""
      # Python 출력 버퍼링 비활성화 (실시간 로그)
      PYTHONUNBUFFERED: "1"
      # 임베딩(게이트웨이) 설정
      TEI_EMBED_URL: "${TEI_EMBED_URL:-https://embedding-ai.ex.co.kr}"
      EMBED_MODEL_NAME: "${EMBED_MODEL_NAME:-Qwen/Qwen3-Embedding-8B}"
      MODEL_API_KEY: "${MODEL_API_KEY:-}"
      MODEL_VERIFY_SSL: "${MODEL_VERIFY_SSL:-false}"
      API_TIMEOUT: "${API_TIMEOUT:-60}"
      EMBED_BATCH_SIZE: "${EMBED_BATCH_SIZE:-100}"
      # Qdrant 저장 배치 (4096차원 → 32MB 요청 한도 초과 방지, 100 권장)
      VECTOR_BATCH_SIZE: "${VECTOR_BATCH_SIZE:-100}"
      # 벡터 DB 설정
      VECTOR_STORE: "${VECTOR_STORE:-qdrant}"
      QDRANT_HOST: "${QDRANT_HOST:-127.0.0.1}"
      QDRANT_PORT: "${QDRANT_PORT:-6333}"
      QDRANT_COLLECTION: "${QDRANT_COLLECTION:-qa_vectors}"
      HYBRID_SEARCH_ENABLED: "${HYBRID_SEARCH_ENABLED:-false}"
      SPARSE_SEARCH_ENABLED: "${SPARSE_SEARCH_ENABLED:-true}"
      SPARSE_TOP_K: "${SPARSE_TOP_K:-30}"
      HYBRID_MERGE_TOP_K: "${HYBRID_MERGE_TOP_K:-40}"
      # 프록시 우회 (내부 IP 직접 연결)
      NO_PROXY: "${NO_PROXY:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8,.ex.co.kr}"
      no_proxy: "${no_proxy:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8,.ex.co.kr}"
    volumes:
      - ./data:/app/data
      - ./scripts:/app/scripts
    restart: "no"

  # 2) 인덱싱 단계
  index:
    image: "${RAG_BATCH_IMAGE:-rag-batch-slim:latest}"
    network_mode: host
    command: ["python", "/app/scripts/build_index_qa.py"]
    working_dir: /app
    depends_on:
      embed:
        condition: service_completed_successfully
    environment:
      # 프록시 비활성화 (명시적)
      HTTP_PROXY: ""
      HTTPS_PROXY: ""
      http_proxy: ""
      https_proxy: ""
      # Python 출력 버퍼링 비활성화 (실시간 로그)
      PYTHONUNBUFFERED: "1"
      # 벡터 DB 설정
      VECTOR_STORE: "${VECTOR_STORE:-faiss}"
      # 프록시 우회
      NO_PROXY: "${NO_PROXY:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8}"
      no_proxy: "${no_proxy:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8}"
    volumes:
      - ./data:/app/data
      - ./scripts:/app/scripts
    restart: "no"

  # 3) API 서비스
  api:
    image: "${RAG_API_IMAGE:-rag-api-slim:latest}"
    network_mode: host   # embed/index/admin과 통일: 127.0.0.1로 qdrant/mongo 접근
    build:
      context: .
      dockerfile: docker/service-slim.Dockerfile
    depends_on:
      - index
      - qdrant
    # 모델 게이트웨이 호스트명 → IP 매핑 (DNS 미등록 대응)
    extra_hosts:
      - "llm-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
      - "embedding-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
      - "reranker-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
    environment:
      # 프록시 비활성화 (명시적)
      HTTP_PROXY: ""
      HTTPS_PROXY: ""
      http_proxy: ""
      https_proxy: ""
      # Python 출력 버퍼링 비활성화 (실시간 로그)
      PYTHONUNBUFFERED: "1"
      # 모델 게이트웨이 공통 인증/TLS
      MODEL_API_KEY: "${MODEL_API_KEY:-}"
      MODEL_VERIFY_SSL: "${MODEL_VERIFY_SSL:-false}"
      # LLM API (게이트웨이)
      LLM_BASE_URL: "${LLM_BASE_URL:-https://llm-ai.ex.co.kr}"
      LLM_MODEL_NAME: "${LLM_MODEL_NAME:-Qwen/Qwen3.6-27B-FP8}"
      LLM_API_KEY: "${LLM_API_KEY:-}"
      
      # 임베딩 / 리랭커 (게이트웨이)
      TEI_EMBED_URL: "${TEI_EMBED_URL:-https://embedding-ai.ex.co.kr}"
      EMBED_MODEL_NAME: "${EMBED_MODEL_NAME:-Qwen/Qwen3-Embedding-8B}"
      TEI_RERANK_URL: "${TEI_RERANK_URL:-https://reranker-ai.ex.co.kr}"
      TEI_RERANK_MODEL: "${TEI_RERANK_MODEL:-Qwen/Qwen3-Reranker-8B}"
      EMBED_API_STYLE: "${EMBED_API_STYLE:-openai}"
      RERANK_API_STYLE: "${RERANK_API_STYLE:-gateway}"
      API_TIMEOUT: "${API_TIMEOUT:-60}"

      # Agent tool loop → chatbotApi (WAS 동일 서버)
      AGENT_MODE: "${AGENT_MODE:-legacy}"
      AGENT_MAX_ROUNDS: "${AGENT_MAX_ROUNDS:-6}"
      CHATBOT_API_BASE_URL: "${CHATBOT_API_BASE_URL:-http://127.0.0.1:8086/api}"
      INTERNAL_TOOL_API_KEY: "${INTERNAL_TOOL_API_KEY:-}"
      CHATBOT_TOOL_TIMEOUT: "${CHATBOT_TOOL_TIMEOUT:-30}"
      
      # MongoDB (로컬 호스트 - 데이터베이스별 인증)
      MONGO_HOST: "${MONGO_HOST:-127.0.0.1}"
      MONGO_PORT: "${MONGO_PORT:-27017}"
      MONGO_USER: "${MONGO_USER:-exlink}"
      MONGO_PASSWORD: "${MONGO_PASSWORD:-!wkcproqkf1}"
      MONGO_DATABASE: "${MONGO_DATABASE:-chat_history}"
      MONGO_COLLECTION: "${MONGO_COLLECTION:-rag_conversations}"
      
      # 벡터 DB
      VECTOR_STORE: "${VECTOR_STORE:-qdrant}"
      QDRANT_HOST: "${QDRANT_HOST:-127.0.0.1}"
      QDRANT_PORT: "${QDRANT_PORT:-6333}"
      QDRANT_COLLECTION: "${QDRANT_COLLECTION:-qa_vectors}"
      HYBRID_SEARCH_ENABLED: "${HYBRID_SEARCH_ENABLED:-false}"
      SPARSE_SEARCH_ENABLED: "${SPARSE_SEARCH_ENABLED:-true}"
      SPARSE_TOP_K: "${SPARSE_TOP_K:-30}"
      HYBRID_MERGE_TOP_K: "${HYBRID_MERGE_TOP_K:-40}"
      
      # 성능 튜닝
      FAISS_TOP_K: "${FAISS_TOP_K:-30}"
      FAISS_THRESHOLD: "${FAISS_THRESHOLD:-0.55}"
      FAISS_THRESHOLD_REWRITE: "${FAISS_THRESHOLD_REWRITE:-0.50}"
      RERANK_CANDIDATES: "${RERANK_CANDIDATES:-20}"
      RERANK_BATCH_SIZE: "${RERANK_BATCH_SIZE:-16}"
      LOW_CONFIDENCE_THRESHOLD: "${LOW_CONFIDENCE_THRESHOLD:-0.65}"
      HIGH_CONFIDENCE_THRESHOLD: "${HIGH_CONFIDENCE_THRESHOLD:-0.75}"
      TOP_N_FOR_LLM: "${TOP_N_FOR_LLM:-5}"
      LLM_MAX_TOKENS: "${LLM_MAX_TOKENS:-2048}"
      QUERY_REWRITE_ENABLED: "${QUERY_REWRITE_ENABLED:-true}"
      CHAT_HISTORY_LIMIT: "${CHAT_HISTORY_LIMIT:-10}"
      CHAT_HISTORY_HOURS: "${CHAT_HISTORY_HOURS:-24}"
      CHAT_HISTORY_ALWAYS_INCLUDE: "${CHAT_HISTORY_ALWAYS_INCLUDE:-true}"
      
      # 프록시 우회 (내부 IP 직접 연결 + 게이트웨이 도메인)
      NO_PROXY: "${NO_PROXY:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8,.ex.co.kr}"
      no_proxy: "${no_proxy:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8,.ex.co.kr}"
    volumes:
      - ./data:/app/data
      - ./scripts:/app/scripts
    restart: unless-stopped

  # Qdrant 벡터 DB (기본 벡터 스토어)
  # ※ profiles 제거 → 별도 플래그 없이 항상 기동
  qdrant:
    image: qdrant/qdrant:v1.18.2   # 폐쇄망에 docker load 한 이미지 태그와 반드시 일치시킬 것
    ports:
      - "6333:6333"   # REST / gRPC
      - "6334:6334"   # gRPC (선택)
    volumes:
      - ./qdrant_data:/qdrant/storage   # 벡터 영속 저장 (컨테이너 재시작에도 유지)
    restart: unless-stopped

  # 4) 큐레이션 어드민 (벡터DB 검색/조회/삭제/추가 웹 UI)
  #    실시간 api와 같은 이미지/모듈을 재사용하되 별도 프로세스로 분리
  admin:
    image: "${RAG_API_IMAGE:-rag-api-slim:latest}"
    network_mode: host
    command:
      - python
      - -m
      - uvicorn
      - scripts.admin_service:app
      - --host
      - 0.0.0.0
      - --port
      - "${ADMIN_PORT:-28013}"
    working_dir: /app
    depends_on:
      - qdrant
    # 모델 게이트웨이 호스트명 → IP 매핑 (어드민도 임베딩 호출)
    extra_hosts:
      - "embedding-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
      - "reranker-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
      - "llm-ai.ex.co.kr:${GATEWAY_IP:-172.16.163.96}"
    environment:
      HTTP_PROXY: ""
      HTTPS_PROXY: ""
      http_proxy: ""
      https_proxy: ""
      PYTHONUNBUFFERED: "1"
      # 임베딩(게이트웨이) — 어드민 검색/추가 시 사용
      TEI_EMBED_URL: "${TEI_EMBED_URL:-https://embedding-ai.ex.co.kr}"
      EMBED_MODEL_NAME: "${EMBED_MODEL_NAME:-Qwen/Qwen3-Embedding-8B}"
      MODEL_API_KEY: "${MODEL_API_KEY:-}"
      MODEL_VERIFY_SSL: "${MODEL_VERIFY_SSL:-false}"
      API_TIMEOUT: "${API_TIMEOUT:-60}"
      VECTOR_STORE: "${VECTOR_STORE:-qdrant}"
      QDRANT_HOST: "${QDRANT_HOST:-127.0.0.1}"
      QDRANT_PORT: "${QDRANT_PORT:-6333}"
      QDRANT_COLLECTION: "${QDRANT_COLLECTION:-qa_vectors}"
      HYBRID_SEARCH_ENABLED: "${HYBRID_SEARCH_ENABLED:-false}"
      SPARSE_SEARCH_ENABLED: "${SPARSE_SEARCH_ENABLED:-true}"
      SPARSE_TOP_K: "${SPARSE_TOP_K:-30}"
      HYBRID_MERGE_TOP_K: "${HYBRID_MERGE_TOP_K:-40}"
      ADMIN_CORS_ORIGINS: "${ADMIN_CORS_ORIGINS:-*}"   # 별도 프론트 origin 제한 시 콤마 구분으로 지정
      NO_PROXY: "${NO_PROXY:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8,.ex.co.kr}"
      no_proxy: "${no_proxy:-localhost,127.0.0.1,172.16.0.0/12,192.168.0.0/16,10.0.0.0/8,.ex.co.kr}"
    volumes:
      - ./data:/app/data
      - ./scripts:/app/scripts
    restart: unless-stopped