chore: 元数据并发安全增强与重排序配置更新

- knowledge: 写操作方法增加元数据重载，manager 增加 fcntl 文件锁
- config/engine: 重排序默认后端切换为云端，新增云端配置项

config.example.py

@@ -106,6 +106,11 @@ RECALL_MULTIPLIER = 3            # 候选池最小倍数 = top_k * 此值
 
 # ----- 重排序 -----
 USE_RERANK = True
+RERANK_BACKEND = os.getenv("RERANK_BACKEND", "cloud")   # cloud / local / fallback
+RERANK_CLOUD_MODEL = os.getenv("RERANK_CLOUD_MODEL", "xop3qwen8breranker")
+RERANK_CLOUD_API_KEY = os.getenv("RERANK_CLOUD_API_KEY", "")
+RERANK_CLOUD_BASE_URL = os.getenv("RERANK_CLOUD_BASE_URL", "https://maas-api.cn-huabei-1.xf-yun.com/v2/rerank")
+RERANK_CLOUD_TIMEOUT = int(os.getenv("RERANK_CLOUD_TIMEOUT", "15"))
 RERANK_CANDIDATES = 20           # 送入重排序的候选数
 RERANK_TOP_K = 15                # 重排序后保留数
 RERANK_USE_ONNX = os.getenv("RERANK_USE_ONNX", "true").lower() == "true"

core/engine.py

@@ -112,10 +112,10 @@ except ImportError:
     EMBEDDING_DEVICE = "auto"
     RERANK_DEVICE = "auto"
     RERANK_USE_ONNX = False
-    RERANK_BACKEND = "local"
+    RERANK_BACKEND = "cloud"
     RERANK_CLOUD_MODEL = "xop3qwen8breranker"
     RERANK_CLOUD_API_KEY = ""
-    RERANK_CLOUD_BASE_URL = "https://dashscope.aliyuncs.com/compatible-api/v1/reranks"
+    RERANK_CLOUD_BASE_URL = "https://maas-api.cn-huabei-1.xf-yun.com/v2/rerank"
     RERANK_CLOUD_TIMEOUT = 15
 
 

knowledge/collection.py

@@ -134,6 +134,9 @@ class CollectionMixin:
         """
         from .base import BM25Index
 
+        # 从磁盘重新加载元数据，确保多 worker 进程间状态一致
+        self._metadata = self._load_metadata()
+
         if not kb_name or not kb_name.replace('_', '').isalnum():
             return False, "向量库名称只能包含字母、数字和下划线"
 
@@ -190,6 +193,9 @@ class CollectionMixin:
         Returns:
             更新成功返回 True，向量库不存在返回 False
         """
+        # 从磁盘重新加载元数据，确保多 worker 进程间状态一致
+        self._metadata = self._load_metadata()
+
         collections = self._metadata.get("collections", {})
         if kb_name not in collections:
             return False
@@ -228,6 +234,9 @@ class CollectionMixin:
         """
         import shutil
 
+        # 从磁盘重新加载元数据，确保多 worker 进程间状态一致
+        self._metadata = self._load_metadata()
+
         if kb_name == PUBLIC_KB_NAME:
             return False, "公开知识库不能删除"
 
@@ -422,4 +431,6 @@ class CollectionMixin:
         Returns:
             存在返回 True，不存在返回 False
         """
+        # 从磁盘重新加载元数据，确保多 worker 进程间状态一致
+        self._metadata = self._load_metadata()
         return kb_name in self._metadata.get("collections", {})

knowledge/manager.py

@@ -29,6 +29,11 @@
 import os
 import json
 import threading
+try:
+    import fcntl
+    _HAS_FCNTL = True
+except ImportError:
+    _HAS_FCNTL = False  # Windows 环境无 fcntl
 from typing import List, Dict, Optional, Tuple
 from pathlib import Path
 import logging
@@ -134,22 +139,36 @@ class KnowledgeBaseManager(
         logger.info(f"知识库管理器初始化完成，路径: {self.base_path}，发现 {len(existing_kbs)} 个向量库: {existing_kbs}")
 
     def _load_metadata(self) -> dict:
-        """加载元数据"""
+        """加载元数据（带文件锁，确保多 worker 进程间一致）"""
         metadata_path = os.path.join(self.base_path, KB_METADATA_FILE)
         if os.path.exists(metadata_path):
             try:
                 with open(metadata_path, 'r', encoding='utf-8') as f:
-                    return json.load(f)
+                    if _HAS_FCNTL:
+                        fcntl.flock(f.fileno(), fcntl.LOCK_SH)
+                    try:
+                        return json.load(f)
+                    finally:
+                        if _HAS_FCNTL:
+                            fcntl.flock(f.fileno(), fcntl.LOCK_UN)
             except Exception as e:
                 logger.error(f"加载元数据失败: {e}")
         return {"collections": {}}
 
     def _save_metadata(self):
-        """保存元数据"""
+        """保存元数据（带文件锁，防止并发写入数据覆盖）"""
         metadata_path = os.path.join(self.base_path, KB_METADATA_FILE)
         try:
             with open(metadata_path, 'w', encoding='utf-8') as f:
-                json.dump(self._metadata, f, ensure_ascii=False, indent=2)
+                if _HAS_FCNTL:
+                    fcntl.flock(f.fileno(), fcntl.LOCK_EX)
+                try:
+                    json.dump(self._metadata, f, ensure_ascii=False, indent=2)
+                    f.flush()
+                    os.fsync(f.fileno())
+                finally:
+                    if _HAS_FCNTL:
+                        fcntl.flock(f.fileno(), fcntl.LOCK_UN)
         except Exception as e:
             logger.error(f"保存元数据失败: {e}")