Merge pull request #829 from SuanmoSuanyangTechnology/fix/Timebomb_0210

fix(workflow)

api/app/core/workflow/nodes/llm/node.py

@@ -246,7 +246,10 @@ class LLMNode(BaseNode):
         logger.info(f"节点 {self.node_id} LLM 调用完成，输出长度: {len(content)}")
 
         # 返回 AIMessage（包含响应元数据）
-        return AIMessage(content=content, response_metadata=response.response_metadata)
+        return AIMessage(content=content, response_metadata={
+            **response.response_metadata,
+            "token_usage": getattr(response, 'usage_metadata', None) or response.response_metadata.get('token_usage')
+        })
 
     def _extract_input(self, state: WorkflowState, variable_pool: VariablePool) -> dict[str, Any]:
         """提取输入数据（用于记录）"""
@@ -305,15 +308,16 @@ class LLMNode(BaseNode):
 
         # 调用 LLM（流式，支持字符串或消息列表）
         last_meta_data = {}
+        last_usage_metadata = {}
         async for chunk in llm.astream(self.messages):
-            # 提取内容
             if hasattr(chunk, 'content'):
                 content = self.process_model_output(chunk.content)
             else:
                 content = str(chunk)
-            if hasattr(chunk, 'response_metadata'):
-                if chunk.response_metadata:
-                    last_meta_data = chunk.response_metadata
+            if hasattr(chunk, 'response_metadata') and chunk.response_metadata:
+                last_meta_data = chunk.response_metadata
+            if hasattr(chunk, 'usage_metadata') and chunk.usage_metadata:
+                last_usage_metadata = chunk.usage_metadata
 
             # 只有当内容不为空时才处理
             if content:
@@ -336,7 +340,10 @@ class LLMNode(BaseNode):
         # 构建完整的 AIMessage（包含元数据）
         final_message = AIMessage(
             content=full_response,
-            response_metadata=last_meta_data
+            response_metadata={
+                **last_meta_data,
+                "token_usage": last_usage_metadata or last_meta_data.get('token_usage')
+            }
         )
 
         # yield 完成标记

api/app/core/workflow/nodes/parameter_extractor/node.py

@@ -208,7 +208,10 @@ class ParameterExtractorNode(BaseNode):
             ])
 
         model_resp = await llm.ainvoke(messages)
-        self.response_metadata = model_resp.response_metadata
+        self.response_metadata = {
+            **model_resp.response_metadata,
+            "token_usage": getattr(model_resp, 'usage_metadata', None) or model_resp.response_metadata.get('token_usage')
+        }
         model_message = self.process_model_output(model_resp.content)
         result = json_repair.repair_json(model_message, return_objects=True)
         logger.info(f"node: {self.node_id} get params:{result}")

api/app/core/workflow/nodes/question_classifier/node.py

@@ -136,7 +136,10 @@ class QuestionClassifierNode(BaseNode):
 
             response = await llm.ainvoke(messages)
             result = self.process_model_output(response.content)
-            self.response_metadata = response.response_metadata
+            self.response_metadata = {
+                **response.response_metadata,
+                "token_usage": getattr(response, 'usage_metadata', None) or response.response_metadata.get('token_usage')
+            }
 
             if result in category_names:
                 category = result