fix: improve robustness of request handling and shared memory linking

lightllm/server/detokenization/manager.py

@@ -76,7 +76,10 @@ def handle_loop(self):
                     for _ in range(recv_max_count):
                         recv_obj: GroupReqIndexes = self.zmq_recv_socket.recv_pyobj(zmq.NOBLOCK)
                         assert isinstance(recv_obj, GroupReqIndexes)
-                        self._add_new_group_req_index(recv_obj=recv_obj)
+                        try:
+                            self._add_new_group_req_index(recv_obj=recv_obj)
+                        except Exception:
+                            logger.exception("add new group req index has exception")
 
                     # 当队列中存在较多的请求时，将一次接受的数量上调
                     recv_max_count = min(int(recv_max_count * 1.3), 256)

lightllm/server/httpserver/manager.py

@@ -284,6 +284,14 @@ async def generate(
         # 用于等待 pd_master 下发的交换信息
         nixl_pd_event: asyncio.Event = None,
     ) -> AsyncGenerator[Tuple[int, str, dict, FinishStatus], None]:
+        if isinstance(prompt, str):
+            max_prompt_chars = self.max_req_total_len * 8
+            if len(prompt) > max_prompt_chars:
+                raise ValueError(
+                    f"prompt text length {len(prompt)} exceeds the character limit {max_prompt_chars}, "
+                    f"the request is rejected before tokenization."
+                )
+
         start_time = time.time()
         request_headers = request.headers if request is not None else {}
         group_request_id = self.alloc_req_id(sampling_params, is_health_req)