Agent 2.0 exdev 서버 배포 스택

- server-dev start/stop/deploy 및 Gitea push 자동 배포 - local-dev 로컬 개발 환경 Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-21 22:57:30 +09:00
commit 4b86b2a660
344 changed files with 45787 additions and 0 deletions
@@ -0,0 +1,231 @@
+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
+