ver 1.0.3 一点小修复加更新一点逻辑

api/glm.py

@@ -51,18 +51,10 @@ async def proxy_glm_chat(request: ChatRequest):
     if request.max_tokens:
         payload["max_tokens"] = request.max_tokens
 
-    # 处理 GLM web_search 功能
-    if request.web_search:
-        tool = {"type": "web_search", "web_search": {}}
-        if request.web_search is True:
-            # 如果是简单的 true,则启用默认的网络搜索
-            tool["web_search"] = {"enable": True}
-        elif isinstance(request.web_search, dict):
-            # 如果是字典,则使用用户提供的配置
-            tool["web_search"] = request.web_search
-            # 确保 enable 标志为 True
-            tool["web_search"]["enable"] = True
-        payload["tools"] = [tool]
+    # 如果客户端请求 web_search=True，则为 GLM 启用网页搜索工具
+    if request.web_search is True:
+        payload["tools"] = [{"type": "web_search", "web_search": {"enable": True}}]
+        logger.info(f"[{request_id}] Enabled web search for GLM.")
 
     logger.debug(
         f"[{request_id}] Upstream Payload:\n"

api/mthreads.py

@@ -2,6 +2,7 @@ import json
 import logging
 import uuid
 import httpx
+import asyncio
 
 from fastapi import APIRouter
 from fastapi.responses import StreamingResponse
@@ -74,49 +75,85 @@ async def proxy_mthreads_chat(request: ChatRequest):
 
     async def stream_generator():
         """
-        第一次请求获取 event_id 后，轮询流式数据
+        第一次请求获取 event_id 后,轮询流式数据;总共尝试3次,每次间隔15秒
         """
-        # --- 第二步: GET 请求流式数据 ---
         stream_url = f"{MTHREADS_UPSTREAM_URL}/{event_id}"
         logger.debug(f"[{request_id}] Streaming from URL: {stream_url}")
-        try:
-            async with httpx.AsyncClient(timeout=300.0) as client:
-                async with client.stream("GET", stream_url) as response:
-                    await handle_upstream_response(response)
-                    line_iterator = response.aiter_lines()
-                    async for line in line_iterator:
 
-                        if line.startswith("event: complete"):
-                            data_line = await line_iterator.__anext__()
-                            logger.debug(f"[{request_id}] Upstream Raw Data Line: {data_line}")
-                            raw_text = json.loads(data_line.split(":", 1)[1].strip())[0]
-                            logger.debug(f"[{request_id}] Upstream Parsed Raw Text: '{raw_text}'")
-                            
-                            # 清理 MThreads 返回的非标准文本
-                            final_text = sanitize_mthreads_response(raw_text)
-                            logger.debug(f"[{request_id}] Agent Cleaned Text: '{final_text}'")
-                            
-                            formatted_data = json.dumps([final_text], ensure_ascii=False)
-                            chunk_for_client_event = f"event: complete\n"
-                            chunk_for_client_data = f"data: {formatted_data}\n\n"
-                            logger.debug(
-                                f"[{request_id}] Yielding to Client:\n{chunk_for_client_event}{chunk_for_client_data}"
-                            )
-                            yield chunk_for_client_event
-                            yield chunk_for_client_data
-                            # 收到 complete 事件后,认为流已结束
-                            break
-        except httpx.RequestError as e:
-            async for item in generate_network_error_sse(e, request_id):
-                yield item
-        except (UpstreamAPIError, StopAsyncIteration, json.JSONDecodeError) as e:
-            if isinstance(e, json.JSONDecodeError):
-                err = UpstreamAPIError(status_code=502, message="Failed to parse MThreads stream response.")
-            elif isinstance(e, StopAsyncIteration):
-                err = UpstreamAPIError(status_code=502, message="MThreads stream was interrupted.")
-            else:
-                err = e
-            async for item in generate_error_sse(err, request_id):
-                yield item
+        for attempt in range(1, 4):
+            logger.info(f"[{request_id}] Attempt {attempt}/3 to stream from MThreads.")
+            try:
+                async with httpx.AsyncClient(timeout=150.0) as client:
+                    async with client.stream("GET", stream_url) as response:
+                        await handle_upstream_response(response)
+                        line_iterator = response.aiter_lines()
+
+                        async for line in line_iterator:
+                            if line.startswith("event: error"):
+                                try:
+                                    data_line = await line_iterator.__anext__()
+                                    if 'data: "404: Session not found."' in data_line:
+                                        logger.warning(
+                                            f"[{request_id}] MThreads session not found on attempt {attempt}. "
+                                            f"{'Will retry.' if attempt < 3 else 'This was the last attempt.'}"
+                                        )
+                                        break
+                                except StopAsyncIteration:
+
+                                    logger.warning(f"[{request_id}] Stream ended after 'event: error' on attempt {attempt}.")
+                                    break
+
+
+                            if line.startswith("event: complete"):
+                                data_line = await line_iterator.__anext__()
+                                logger.debug(f"[{request_id}] Upstream Raw Data Line: {data_line}")
+                                raw_text = json.loads(data_line.split(":", 1)[1].strip())[0]
+                                logger.debug(f"[{request_id}] Upstream Parsed Raw Text: '{raw_text}'")
+
+                                final_text = sanitize_mthreads_response(raw_text)
+                                logger.debug(f"[{request_id}] Agent Cleaned Text: '{final_text}'")
+
+                                formatted_data = json.dumps([final_text], ensure_ascii=False)
+                                chunk_for_client_event = f"event: complete\n"
+                                chunk_for_client_data = f"data: {formatted_data}\n\n"
+                                logger.debug(
+                                    f"[{request_id}] Yielding to Client:\n{chunk_for_client_event}{chunk_for_client_data}"
+                                )
+                                yield chunk_for_client_event
+                                yield chunk_for_client_data
+
+                                return
+
+            except httpx.RequestError as e:
+                logger.error(f"[{request_id}] Network error on attempt {attempt}: {e}")
+                if attempt == 3:
+                    async for item in generate_network_error_sse(e, request_id):
+                        yield item
+                    return
+            except (UpstreamAPIError, StopAsyncIteration, json.JSONDecodeError) as e:
+                logger.warning(f"[{request_id}] Handled error on attempt {attempt}: {type(e).__name__} - {e}")
+                if isinstance(e, StopAsyncIteration):
+                     logger.warning(f"[{request_id}] MThreads stream was interrupted on attempt {attempt}. Retrying...")
+                
+                if attempt == 3:
+                    if isinstance(e, json.JSONDecodeError):
+                        err = UpstreamAPIError(status_code=502, message="Failed to parse MThreads stream response.")
+                    elif isinstance(e, StopAsyncIteration):
+                        err = UpstreamAPIError(status_code=502, message="MThreads stream was interrupted unexpectedly on final attempt.")
+                    else:
+                        err = e
+                    async for item in generate_error_sse(err, request_id):
+                        yield item
+                    return
+
+            if attempt < 3:
+                logger.info(f"[{request_id}] Waiting 15 seconds before next attempt.")
+                await asyncio.sleep(15)
+
+        logger.error(f"[{request_id}] Failed to get response from MThreads after 3 attempts. Returning 'No content' message.")
+        error_message = "No content returned from model"
+        error_payload = json.dumps({"error": {"message": error_message, "type": "proxy_error", "code": 504}}, ensure_ascii=False)
+        yield f"event: error\n"
+        yield f"data: {error_payload}\n\n"
 
     return StreamingResponse(stream_generator(), media_type="text/event-stream") 

api/qwen.py

@@ -50,6 +50,12 @@ async def proxy_qwen_chat(request: ChatRequest):
         "top_p": request.top_p,
         "max_tokens": request.max_tokens,
     }
+
+    # 如果客户端请求 web_search=True，则为 Qwen 启用网页搜索
+    if request.web_search is True:
+        payload["enable_search"] = True
+        logger.info(f"[{request_id}] Enabled web search for Qwen.")
+
     logger.debug(
         f"[{request_id}] Upstream Payload:\n"
         f"{json.dumps(payload, indent=2, ensure_ascii=False)}"

main.py

@@ -71,7 +71,7 @@ logger = logging.getLogger("proxy")
 app = FastAPI(
     title="LLM Chat API Proxy",
     description="A proxy for LLM APIs, supporting GLM, Qwen, and MThreads Playground.",
-    version="1.0.2",
+    version="1.0.3",
 )
 
 # 包含来自各个 API 模块的路由
@@ -82,7 +82,7 @@ app.include_router(glm.router, prefix="/api")
 
 if __name__ == "__main__":
     logger.info(
-        f"Starting LLM API Proxy v1.0.2 on http://{DEFAULT_LISTEN_HOST}:{listen_port}"
+        f"Starting LLM API Proxy v1.0.3 on http://{DEFAULT_LISTEN_HOST}:{listen_port}"
     )
     if config.DEBUG:
         logger.info("Debug mode is enabled. Logging to debug.log")