refactor(rag): 架構審查修正 — leWOOOgo 合規 + 去重 + httpx 關機

- C2: _run_index() 業務邏輯移入 KnowledgeRAGService.index_all_sources()
      Router 層只做 background_tasks.add_task(_run_index) 轉發
- C3: glob("*.md") → rglob("*.md") — 掃描巢狀子目錄
- C4: docstring 修正 Ollama 188 → 111
- I2: index_document() 先刪舊版本 (_delete_by_source_id) 避免重複累積
- I3: debug endpoint 改用 settings.OLLAMA_URL 取代硬碼 IP
- I4: main.py shutdown 加入 get_knowledge_rag_service().close()

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/api/src/api/v1/rag.py

@@ -75,8 +75,9 @@ async def rag_debug() -> dict:
     ollama_ok: bool | str = False
     try:
         async with httpx.AsyncClient(timeout=10.0) as c:
+            from src.core.config import get_settings as _gs
             r = await c.post(
-                "http://192.168.0.111:11434/api/embeddings",
+                f"{_gs().OLLAMA_URL}/api/embeddings",
                 json={"model": "nomic-embed-text", "prompt": "test"},
             )
             ollama_ok = r.status_code == 200 if r.status_code == 200 else f"http_{r.status_code}"
@@ -98,38 +99,6 @@ async def rag_stats() -> dict:
 # ============================================================
 
 async def _run_index() -> None:
-    """背景：掃描所有文件來源並向量化"""
-    import structlog
-    from pathlib import Path
-
-    logger = structlog.get_logger(__name__)
+    """背景：委派給 KnowledgeRAGService.index_all_sources() 執行"""
     svc = get_knowledge_rag_service()
-
-    sources = [
-        Path("docs/runbooks"),
-        Path("docs/adr"),
-        Path("docs"),
-        Path(".agents/skills"),
-    ]
-
-    total = 0
-    for source_dir in sources:
-        if not source_dir.exists():
-            continue
-        for md_file in source_dir.glob("*.md"):
-            try:
-                text = md_file.read_text(encoding="utf-8", errors="ignore")
-                source_type = source_dir.parts[-1]  # e.g. "runbooks", "adr", "skills"
-                ok = await svc.index_document(
-                    source=source_type,
-                    source_id=str(md_file),
-                    title=md_file.stem,
-                    text=text,
-                )
-                if ok:
-                    total += 1
-                logger.debug("rag_indexed", file=str(md_file))
-            except Exception as e:
-                logger.warning("rag_index_file_failed", file=str(md_file), error=str(e))
-
-    logger.info("rag_index_complete", total_chunks=total)
+    await svc.index_all_sources()

apps/api/src/main.py

@@ -296,6 +296,9 @@ async def lifespan(_app: FastAPI) -> AsyncGenerator[None, None]:
     # Phase 5.4: Close Telegram Gateway
     telegram_gw = get_telegram_gateway()
     await telegram_gw.close()
+    # Phase 33: Close RAG Service httpx client (ADR-067)
+    from src.services.knowledge_rag_service import get_knowledge_rag_service
+    await get_knowledge_rag_service().close()
     # Phase 5: Close HTTP Clients (統帥鐵律: 連線池回收)
     await close_all_http_clients()
     # Phase 6.1.1: Close Redis Pool (統帥鐵律: Redis 連線池回收)

apps/api/src/services/knowledge_rag_service.py

@@ -11,7 +11,7 @@ AWOOOI — Knowledge RAG Service (Phase 33, ADR-067)
   - 初期 < 100 筆: 線性搜尋
   - 超過 100 筆: 執行 CREATE INDEX ivfflat (手動觸發)
 
-向量模型: nomic-embed-text (Ollama 188, 768維)
+向量模型: nomic-embed-text (Ollama 111, 768維) — 188:11434 被 NetworkPolicy v1.3 封閉
 生成模型: qwen2.5:7b-instruct (Ollama 111)
 
 2026-04-10 Claude Sonnet 4.6 Asia/Taipei
@@ -87,6 +87,8 @@ class KnowledgeRAGService:
         將文件向量化並儲存到 pgvector
         自動分段 (每段 500 字, overlap 100)
         """
+        # 先刪舊版本，避免重複索引累積 (I2 審查修正 2026-04-10)
+        await self._delete_by_source_id(source_id)
         chunks = self._chunk_text(text, chunk_size=500, overlap=100)
         success = 0
         for chunk in chunks:
@@ -143,6 +145,18 @@ class KnowledgeRAGService:
             logger.error("rag_pgvector_search_failed", error=str(e))
             return []
 
+    async def _delete_by_source_id(self, source_id: str) -> None:
+        try:
+            from src.db.base import get_db_context
+            from sqlalchemy import text
+            async with get_db_context() as db:
+                await db.execute(
+                    text("DELETE FROM rag_chunks WHERE source_id = :source_id"),
+                    {"source_id": source_id},
+                )
+        except Exception as e:
+            logger.warning("rag_delete_old_chunks_failed", source_id=source_id, error=str(e))
+
     async def _save_chunk(
         self,
         source: str, source_id: str, title: str,
@@ -216,6 +230,41 @@ class KnowledgeRAGService:
             start += chunk_size - overlap
         return [c for c in chunks if c.strip()]
 
+    async def index_all_sources(self) -> int:
+        """
+        掃描所有知識來源並向量化（供 /rag/index 端點呼叫）
+        來源: docs/runbooks/, docs/adr/, docs/, .agents/skills/
+        回傳成功索引的文件數
+        """
+        from pathlib import Path
+
+        sources = [
+            Path("docs/runbooks"),
+            Path("docs/adr"),
+            Path("docs"),
+            Path(".agents/skills"),
+        ]
+        total = 0
+        for source_dir in sources:
+            if not source_dir.exists():
+                continue
+            for md_file in source_dir.rglob("*.md"):
+                try:
+                    text = md_file.read_text(encoding="utf-8", errors="ignore")
+                    ok = await self.index_document(
+                        source=source_dir.parts[-1],
+                        source_id=str(md_file),
+                        title=md_file.stem,
+                        text=text,
+                    )
+                    if ok:
+                        total += 1
+                    logger.debug("rag_source_indexed", file=str(md_file))
+                except Exception as e:
+                    logger.warning("rag_source_index_failed", file=str(md_file), error=str(e))
+        logger.info("rag_index_all_complete", total_docs=total)
+        return total
+
     async def get_stats(self) -> dict:
         """RAG 知識庫統計"""
         from src.db.base import get_db_context