Handle backup write errors and improve timer cancel

swanlab/data/porter/__init__.py

@@ -61,11 +61,16 @@ def backup():
     def wrapper(wrapped, instance, args, kwargs):
         result: Union[List[BaseModel], BaseModel] = wrapped(*args, **kwargs)
         if result is not None:
-            f = getattr(instance, "_f")
-            if isinstance(result, list):
-                [f.write(item.to_record()) for item in result]
-            else:
-                f.write(result.to_record())
+            try:
+                f = getattr(instance, "_f")
+                if isinstance(result, list):
+                    [f.write(item.to_record()) for item in result]
+                else:
+                    f.write(result.to_record())
+            except ValueError:
+                # 写入备份文件失败，可能是因为没有开启备份模式或者备份文件未打开
+                # TODO: 记录本地日志
+                pass
         return result
 
     return wrapper

swanlab/data/run/helper.py

@@ -159,6 +159,7 @@ def _():
     def cancel(self):
         if self.timer is not None:
             self.timer.cancel()
+            self.timer.join()
 
     @property
     def sleep_time(self):

swanlab/data/run/main.py

@@ -93,7 +93,7 @@ def __init__(
         # 定时采集系统信息
         # 测试时不开启此功能
         # resume时不开启此功能
-        if "PYTEST_VERSION" not in os.environ or run_store.resume == 'never':
+        if "PYTEST_VERSION" not in os.environ and run_store.resume == 'never':
             if monitor_funcs is not None and len(monitor_funcs) != 0:
                 swanlog.debug("Monitor on.")
 

test/unit/sync/v0/test_sync.py

@@ -30,7 +30,7 @@ def test_sync():
     record_logs: List[str] = []
     # ---------------------------------- 开启实验 ----------------------------------
     config = {
-        "epochs": 10,
+        "epochs": 10000,
         "learning_rate": 0.01,
         "offset": random.random() / 5,
     }