Merge remote-tracking branch 'origin/develop' into refactor/memory-config-management

2025-12-24 16:07:27 +08:00
parent 0fff23d2a2 3afe547559
commit 292fef0fa7
24 changed files with 365 additions and 202 deletions
--- a/api/app/core/memory/agent/mcp_server/server.py
+++ b/api/app/core/memory/agent/mcp_server/server.py
@@ -131,6 +131,18 @@ def main():
        
        # Get MCP port from environment (default: 8081)
        mcp_port = int(os.getenv("MCP_PORT", "8081"))
+        logger.info(f"Starting MCP server on {settings.SERVER_IP}:{mcp_port} with SSE transport")
+        
+        # Configure DNS rebinding protection for Docker container compatibility
+        from mcp.server.fastmcp.server import TransportSecuritySettings
+        
+        # Disable DNS rebinding protection to allow Docker container hostnames
+        # This allows containers to connect using service names like 'mcp-server'
+        mcp.settings.transport_security = TransportSecuritySettings(
+            enable_dns_rebinding_protection=False,
+        )
+        logger.info("DNS rebinding protection: disabled for Docker container compatibility")
+        
        # logger.info(f"Starting MCP server on {settings.SERVER_IP}:{mcp_port} with SSE transport")
        
        # Run the server with SSE transport for HTTP connections
--- a/api/app/core/memory/storage_services/reflection_engine/example/example.json
+++ b/api/app/core/memory/storage_services/reflection_engine/example/example.json
@@ -50,9 +50,7 @@
        "entity2_name": "用户",
        "entity2": {
          "description": "叙述者，讲述个人工作与生活经历的个体",
-          "statement_id": "62beac695b1346f4871740a45db88782",
-          "name": "用户",
-          "id": "3d3896797b334572a80d57590026063d"
+          "name": "用户"
        }
      },
      {
@@ -62,9 +60,7 @@
        "entity2_name": "身份信息",
        "entity2": {
          "description": "用于个人身份识别的数据",
-          "statement_id": "030afd362e9b4110b139e68e5d3e7143",
-          "name": "身份信息",
-          "id": "aa766a517e82490599a9b3af54cfd933"
+          "name": "身份信息"
        }
      },
      {
@@ -74,9 +70,7 @@
        "entity2_name": "6222023847595898",
        "entity2": {
          "description": "用户的银行卡号码",
-          "statement_id": "6c7567cd1f3c478bb42d1b65383e6f2f",
-          "name": "6222023847595898",
-          "id": "610ba361918f4e68a65ce6ad06e5c7a0"
+          "name": "6222023847595898"
        }
      },
      {
@@ -88,9 +82,7 @@
          "entity_idx": 1,
          "aliases": ["上海办"],
          "description": "位于上海的工作办公场所",
-          "statement_id": "8b1b12e23b844b8088dfeb67da6ad669",
-          "name": "上海办公室",
-          "id": "fb702ef695c14e14af3e56786bc8815b"
+          "name": "上海办公室"
        }
      },
      {
@@ -101,9 +93,7 @@
        "entity2": {
          "aliases": ["京", "京城", "北平"],
          "description": "中国的首都城市，用户主要工作和生活所在地",
-          "statement_id": "62beac695b1346f4871740a45db88782",
-          "name": "北京",
-          "id": "81b2d1a571bb46a08a2d7a1e87efb945"
+          "name": "北京"
        }
      },
      {
@@ -113,9 +103,7 @@
        "entity2_name": "身份证号",
        "entity2": {
          "description": "中华人民共和国公民的身份号码",
-          "statement_id": "030afd362e9b4110b139e68e5d3e7143",
-          "name": "身份证号",
-          "id": "3e5f920645b2404fadb0e9ff60d1306e"
+          "name": "身份证号"
        }
      }
    ]
--- a/api/app/core/memory/storage_services/reflection_engine/self_reflexion.py
+++ b/api/app/core/memory/storage_services/reflection_engine/self_reflexion.py
@@ -269,8 +269,6 @@ class ReflectionEngine:
            # # 检查是否真的有冲突
            conflicts_found=''

-            # 记录冲突数据
-            await self._log_data("conflict", conflict_data)
            conflicts_found=''
            # 3. 解决冲突
            solved_data = await self._resolve_conflicts(conflict_data, statement_databasets)
@@ -288,8 +286,6 @@ class ReflectionEngine:
            conflicts_resolved = len(solved_data)
            logging.info(f"解决了 {conflicts_resolved} 个冲突")

-            # 记录解决方案
-            await self._log_data("solved_data", solved_data)

            # 4. 应用反思结果（更新记忆库）
            memories_updated=await self._apply_reflection_results(solved_data)
@@ -390,14 +386,7 @@ class ReflectionEngine:
            memory_verifies.append(item['memory_verify'])
        result_data['memory_verifies'] = memory_verifies
        result_data['quality_assessments'] = quality_assessments
-
-        # 检查是否真的有冲突
-        has_conflict = conflict_data[0].get('conflict', False)
-        conflicts_found = len(conflict_data[0]['data']) if has_conflict else 0
-        logging.info(f"冲突状态: {has_conflict}, 发现 {conflicts_found} 个冲突")
-
-        # 记录冲突数据
-        await self._log_data("conflict", conflict_data)
+        conflicts_found=''

        # Clearn conflict_data，And memory_verify和quality_assessment
        cleaned_conflict_data = []
@@ -407,6 +396,7 @@ class ReflectionEngine:
                'conflict': item['conflict']
            }
            cleaned_conflict_data.append(cleaned_item)
+
        # 3. 解决冲突
        solved_data = await self._resolve_conflicts(cleaned_conflict_data, source_data)
        if not solved_data:
@@ -645,26 +635,7 @@ class ReflectionEngine:
        success_count = await neo4j_data(changes)
        return success_count

-    async def _log_data(self, label: str, data: Any) -> None:
-        """
-        记录数据到文件

-        Args:
-            label: 数据标签
-            data: 要记录的数据
-        """
-
-        def _write():
-            try:
-                with open("reflexion_data.json", "a", encoding="utf-8") as f:
-                    f.write(f"### {label} ###\n")
-                    json.dump(data, f, ensure_ascii=False, indent=4)
-                    f.write("\n\n")
-            except Exception as e:
-                logging.warning(f"记录数据失败: {e}")
-
-        # 在后台线程中执行写入，避免阻塞事件循环
-        await asyncio.to_thread(_write)

    # 基于时间的反思方法
    async def time_based_reflection(
@@ -753,4 +724,3 @@ class ReflectionEngine:
            raise ValueError(f"未知的反思基线: {self.config.baseline}")


-
--- a/api/app/core/memory/utils/prompt/prompts/evaluate.jinja2
+++ b/api/app/core/memory/utils/prompt/prompts/evaluate.jinja2
@@ -17,10 +17,12 @@
 - **日期属性冲突**: 同一人的生日等单值属性出现多值
 - **先后约束违反**: 存在A→B约束但t(A)>t(B)（如入学>毕业）
 - **互斥重叠**: 同一时间出现在不同地点等互斥事件
+- **隐私审核**: 存在隐私信息也作为冲突输出当{{ memory_verify }}是true的时候
 ### 事实冲突
 - **属性互斥**: 同一实体的相反属性（喜欢↔不喜欢）
 - **关系矛盾**: 同一实体在相同语境下的不同关系描述
 - **身份冲突**: 同一实体被赋予不同类型或角色
+- **隐私审核**: 存在隐私信息也作为冲突输出当{{ memory_verify }}是true的时候
 ### 混合冲突
 检测所有逻辑不一致或相互矛盾的记录。
 **检测原则**:
--- a/api/app/core/memory/utils/prompt/prompts/reflexion.jinja2
+++ b/api/app/core/memory/utils/prompt/prompts/reflexion.jinja2
@@ -171,7 +171,6 @@
  ]
 }
 ```
-
 **输出要求**:
 - 只输出JSON，不添加解释文本
 - 使用标准双引号，必要时转义
--- a/api/app/core/memory/utils/prompt/template_render.py
+++ b/api/app/core/memory/utils/prompt/template_render.py
@@ -7,7 +7,7 @@ from typing import List, Dict, Any
 prompt_dir = os.path.join(os.path.dirname(__file__), "prompts")
 prompt_env = Environment(loader=FileSystemLoader(prompt_dir))

-async def render_evaluate_prompt(evaluate_data: List[Any], schema: Dict[str, Any],
+async def render_evaluate_prompt(evaluate_data: List[Any], schema: Any,
                                 baseline: str = "TIME",
                                 memory_verify: bool = False,quality_assessment:bool = False,
                                 statement_databasets: List[str] = [],language_type:str = "zh") -> str:
@@ -16,7 +16,7 @@ async def render_evaluate_prompt(evaluate_data: List[Any], schema: Dict[str, Any

    Args:
        evaluate_data: The data to evaluate
-        schema: The JSON schema to use for the output.
+        schema: The Pydantic model class or JSON schema to use for the output.
        baseline: The baseline type for conflict detection (TIME/FACT/TIME-FACT)
        memory_verify: Whether to enable memory verification for privacy detection

@@ -25,9 +25,17 @@ async def render_evaluate_prompt(evaluate_data: List[Any], schema: Dict[str, Any
    """
    template = prompt_env.get_template("evaluate.jinja2")

+    # Convert Pydantic model to JSON schema if needed
+    if hasattr(schema, 'model_json_schema'):
+        json_schema = schema.model_json_schema()
+    elif hasattr(schema, 'schema'):
+        json_schema = schema.schema()
+    else:
+        json_schema = schema
+
    rendered_prompt = template.render(
        evaluate_data=evaluate_data,
-        json_schema=schema,
+        json_schema=json_schema,
        baseline=baseline,
        memory_verify=memory_verify,
        quality_assessment=quality_assessment,
@@ -36,14 +44,15 @@ async def render_evaluate_prompt(evaluate_data: List[Any], schema: Dict[str, Any
    )
    return rendered_prompt

-async def render_reflexion_prompt(data: Dict[str, Any], schema: Dict[str, Any], baseline: str, memory_verify: bool = False,
+
+async def render_reflexion_prompt(data: Dict[str, Any], schema: Any, baseline: str, memory_verify: bool = False,
                                  statement_databasets: List[str] = [],language_type:str = "zh") -> str:
    """
    Renders the reflexion prompt using the reflexion_optimized.jinja2 template.

    Args:
        data: The data to reflex on.
-        schema: The JSON schema to use for the output.
+        schema: The Pydantic model class or JSON schema to use for the output.
        baseline: The baseline type for conflict resolution.

    Returns:
@@ -51,7 +60,15 @@ async def render_reflexion_prompt(data: Dict[str, Any], schema: Dict[str, Any],
    """
    template = prompt_env.get_template("reflexion.jinja2")

-    rendered_prompt = template.render(data=data, json_schema=schema,
+    # Convert Pydantic model to JSON schema if needed
+    if hasattr(schema, 'model_json_schema'):
+        json_schema = schema.model_json_schema()
+    elif hasattr(schema, 'schema'):
+        json_schema = schema.schema()
+    else:
+        json_schema = schema
+
+    rendered_prompt = template.render(data=data, json_schema=json_schema,
                                      baseline=baseline,memory_verify=memory_verify,
                                      statement_databasets=statement_databasets,language_type=language_type)