fix(server): stop bare excepts from swallowing cancellation and shutdown

lightllm/server/build_prompt.py

@@ -58,15 +58,15 @@ def tokenizer_supports_force_thinking() -> bool:
         logger.debug(f"chat_template: {tokenizer.chat_template}")
         logger.info(f"tokenizer_supports_force_thinking : {ans}")
         return ans
-    except:
+    except (AttributeError, TypeError):
         pass
 
     try:
         ans = "thinking" in tokenizer.tokenizer.chat_template or "enable_thinking" in tokenizer.tokenizer.chat_template
         logger.debug(f"tokenizer.tokenizer.chat_template: {tokenizer.tokenizer.chat_template}")
         logger.info(f"tokenizer_supports_force_thinking : {ans}")
         return ans
-    except:
+    except (AttributeError, TypeError):
         pass
 
     logger.info("tokenizer_supports_force_thinking : False")

lightllm/server/core/objs/py_sampling_params.py

@@ -125,7 +125,7 @@ def _cfg(key, default):
             cls._top_p = _cfg("top_p", 1.0)
             cls._top_k = _cfg("top_k", -1)
             cls._stop_sequences = generation_cfg.get("stop", None)
-        except:
+        except Exception:
             pass
 
     def verify(self):

lightllm/server/core/objs/sampling_params.py

@@ -415,7 +415,7 @@ def _cfg(key, default):
             cls._temperature = _cfg("temperature", 1.0)
             cls._top_p = _cfg("top_p", 1.0)
             cls._top_k = _cfg("top_k", -1)
-        except:
+        except Exception:
             pass
 
     def verify(self):

lightllm/server/embed_cache/afs_utils.py

@@ -50,7 +50,7 @@ def save_tensor_afs(self, name: str, tensor: torch.Tensor) -> bool:
         finally:
             try:
                 tmp_path.unlink(missing_ok=True)
-            except:
+            except OSError:
                 pass
 
     def load_tensor_afs(self, name: str) -> Optional[torch.Tensor]:
@@ -132,7 +132,7 @@ def insert(self, md5: str, tensor: torch.Tensor) -> bool:
             ans = self.afs_utils.save_tensor_afs(md5, tensor)
             self.redis_client.update(md5)
             return ans
-        except:
+        except Exception:
             return False
 
     def load(self, md5: str) -> Optional[torch.Tensor]:

lightllm/server/httpserver_for_pd_master/manager.py

@@ -103,7 +103,7 @@ async def update_req_status(self, upkv_status: PDUpKVStatus):
             up_status_event = self.req_id_to_out_inf[group_request_id].up_status_event
             up_status_event.upkv_status = upkv_status
             up_status_event.set()
-        except:
+        except Exception:
             pass
         return
 
@@ -414,7 +414,7 @@ async def _generate_one_attempt(
 
             try:
                 await self.abort(block_group_request_id, p_node=p_node, d_node=d_node)
-            except:
+            except Exception:
                 await self.abort(block_group_request_id)
             raise e
 
@@ -724,25 +724,25 @@ async def abort(
             del self.req_id_to_out_inf[group_request_id]
             p_node = req_status.p_node
             d_node = req_status.d_node
-        except:
+        except Exception:
             pass
 
         try:
             await p_node.websocket.send_bytes(pickle.dumps((ObjType.ABORT, group_request_id)))
-        except:
+        except Exception:
             pass
 
         try:
             await d_node.websocket.send_bytes(pickle.dumps((ObjType.ABORT, group_request_id)))
-        except:
+        except Exception:
             pass
 
         return
 
     async def remove_req(self, group_request_id):
         try:
             del self.req_id_to_out_inf[group_request_id]
-        except:
+        except Exception:
             pass
 
     async def timer_log(self):
@@ -782,7 +782,7 @@ async def handle_loop(self):
                                 async with req_status.lock:
                                     req_status.out_token_info_list.append((sub_req_id, text, metadata, finish_status))
                                     req_status.event.set()
-                            except:
+                            except Exception:
                                 pass
                     elif obj[0] == ObjType.PD_UPLOAD_PREFILL_PROMPT_IDS:
                         _, group_req_id, prompt_ids = obj
@@ -791,7 +791,7 @@ async def handle_loop(self):
                             async with req_status.lock:
                                 req_status.prefill_prompt_ids_event.prompt_ids = prompt_ids
                                 req_status.prefill_prompt_ids_event.set()
-                        except:
+                        except Exception:
                             logger.error(
                                 f"PD_UPLOAD_PREFILL_PROMPT_IDS fail find req status for group_req_id: {group_req_id}"
                             )

lightllm/server/metrics/manager.py

@@ -70,7 +70,7 @@ def push_metrics(self):
                 if time_counter >= 60:
                     logger.info("push metrices success")
                     time_counter = 0
-            except:
+            except Exception:
                 pass
             finally:
                 time.sleep(self.interval)

lightllm/server/multi_level_kv_cache/shm_objs.py

@@ -296,7 +296,7 @@ def _create_shm(name: str, byte_size: int, auto_cleanup: bool = False):
         if auto_cleanup:
             register_posix_shm_for_cleanup(name)
         logger.info(f"create lock shm {name}")
-    except:
+    except Exception:
         shm = shared_memory.SharedMemory(name=name, create=False, size=byte_size)
         logger.info(f"link lock shm {name}")
     return shm

lightllm/server/router/model_infer/mode_backend/pd/decode_node_impl/decode_trans_process.py

@@ -258,7 +258,7 @@ def accept_peer_task_loop(
                     for notify in _notify_list:
                         try:
                             notify_obj = pickle.loads(notify)
-                        except:
+                        except Exception:
                             notify_obj = None
 
                         if not isinstance(notify_obj, PDChunckedTransTask):

lightllm/server/router/model_infer/mode_backend/pd/trans_process_obj.py

@@ -55,7 +55,7 @@ def is_trans_process_health(self):
                 return False
             else:
                 return True
-        except:
+        except Exception:
             return False
 
     def killself(self):