Refactor client utilities and add experiment heartbeat

Moved client utility functions and models to a new utils.py file, including the safe_request decorator. Introduced a client heartbeat mechanism to keep experiments active, with integration in CloudPyCallback. Updated imports and improved type hinting for better maintainability.

Author: SAKURA-CAT

.idea/SwanLab.iml

@@ -10,5 +10,5 @@
 # FIXME 存在循环引用，我们需要更优雅的代码结构
 # from . import auth
 # from . import uploader
-from .client import Client, create_client, reset_client, get_client, create_session
+from .client import *
 from .utils import timer

.idea/dictionaries/project.xml

@@ -5,13 +5,29 @@
 @description: 定义实验相关的后端API接口
 """
 
-from typing import Literal
+from typing import Literal, TYPE_CHECKING
 
-from swanlab.core_python.client import Client
+if TYPE_CHECKING:
+    from swanlab.core_python.client import Client
+
+
+def send_experiment_heartbeat(
+    client: "Client",
+    *,
+    cuid: str,
+    flag_id: str,
+):
+    """
+    发送实验心跳，保持实验处于活跃状态
+    :param client: 已登录的客户端实例
+    :param cuid: 实验唯一标识符
+    :param flag_id: 实验标记ID
+    """
+    client.post(f"/house/experiments/{cuid}/heartbeat", {"flagId": flag_id})
 
 
 def update_experiment_state(
-    client: Client,
+    client: "Client",
     *,
     username: str,
     projname: str,

swanlab/core_python/__init__.py

@@ -10,21 +10,15 @@
 from typing import Optional, Tuple, Dict, Union, List, AnyStr
 
 import requests
-from urllib3.exceptions import (
-    MaxRetryError,
-    TimeoutError,
-    NewConnectionError,
-    ConnectionError,
-    ReadTimeoutError,
-    ConnectTimeoutError,
-)
-
-from swanlab.error import NetworkError, ApiError
+
+from swanlab.error import ApiError
 from swanlab.log import swanlog
 from swanlab.package import get_package_version
-from .model import ProjectInfo, ExperimentInfo
 from .session import create_session
+from .utils import safe_request, ProjectInfo, ExperimentInfo
 from .. import auth
+from ..api.experiment import send_experiment_heartbeat
+from ..utils import timer
 from ...env import utc_time
 
 
@@ -114,11 +108,11 @@ def expname(self):
         return self.__exp.name
 
     @property
-    def web_proj_url(self):
+    def web_proj_url(self) -> str:
         return f"{self.__login_info.web_host}/@{self.groupname}/{self.projname}"
 
     @property
-    def web_exp_url(self):
+    def web_exp_url(self) -> str:
         return f"{self.web_proj_url}/runs/{self.exp_id}"
 
     # ---------------------------------- http方法 ----------------------------------
@@ -374,41 +368,31 @@ def reset_client():
     client = None
 
 
-def safe_request(func):
+def create_client_heartbeat():
     """
-    在一些接口中我们不希望线程奔溃，而是返回一个错误对象
+    创建客户端心跳定时器，保持实验处于活跃状态
+    :return: 心跳定时器实例
     """
+    cl = get_client()
 
-    def wrapper(*args, **kwargs) -> Tuple[Optional[Union[dict, str]], Optional[Exception]]:
+    # TODO 目前保证乡下兼容，如果报错也不提示用户，后续使用safe_request装饰器
+    # func = safe_request(func=send_experiment_heartbeat)
+    def func(c: Client, *, cuid: str, flag_id: str):
         try:
-            # 在装饰器中调用被装饰的异步函数
-            result = func(*args, **kwargs)
-            return result, None
-        except requests.exceptions.Timeout:
-            return None, NetworkError()
-        except requests.exceptions.ConnectionError:
-            return None, NetworkError()
-        # Catch urllib3 specific errors
-        except (
-            MaxRetryError,
-            TimeoutError,
-            NewConnectionError,
-            ConnectionError,
-            ReadTimeoutError,
-            ConnectTimeoutError,
-        ):
-            return None, NetworkError()
-        except Exception as e:
-            return None, e
-
-    return wrapper
+            send_experiment_heartbeat(c, cuid=cuid, flag_id=flag_id)
+        except ApiError as e:
+            swanlog.debug("Failed to send heartbeat: " + str(e))
+
+    task = lambda: func(cl, cuid=cl.exp.cuid, flag_id=cl.exp.flag_id)
+    return timer.Timer(task, interval=10, immediate=True).run()
 
 
 __all__ = [
     "get_client",
     "reset_client",
     "create_session",
     "create_client",
+    "create_client_heartbeat",
     "safe_request",
     "decode_response",
     "Client",

swanlab/core_python/api/experiment.py

@@ -1,11 +1,53 @@
 """
 @author: cunyue
-@file: model.py
-@time: 2025/6/16 14:55
-@description: 实验、项目元信息
+@file: utils.py
+@time: 2025/12/31 13:29
+@description: 客户端工具函数
 """
 
-from typing import Optional
+from typing import Tuple, Optional, Union
+
+import requests
+from urllib3.exceptions import (
+    MaxRetryError,
+    TimeoutError,
+    NewConnectionError,
+    ConnectionError,
+    ReadTimeoutError,
+    ConnectTimeoutError,
+)
+
+from swanlab.error import NetworkError
+
+
+def safe_request(func):
+    """
+    在一些接口中我们不希望线程奔溃，而是返回一个错误对象
+    """
+
+    def wrapper(*args, **kwargs) -> Tuple[Optional[Union[dict, str]], Optional[Exception]]:
+        try:
+            # 在装饰器中调用被装饰的异步函数
+            result = func(*args, **kwargs)
+            return result, None
+        except requests.exceptions.Timeout:
+            return None, NetworkError()
+        except requests.exceptions.ConnectionError:
+            return None, NetworkError()
+        # Catch urllib3 specific errors
+        except (
+            MaxRetryError,
+            TimeoutError,
+            NewConnectionError,
+            ConnectionError,
+            ReadTimeoutError,
+            ConnectTimeoutError,
+        ):
+            return None, NetworkError()
+        except Exception as e:
+            return None, e
+
+    return wrapper
 
 
 class ProjectInfo:

swanlab/core_python/client/__init__.py

@@ -27,6 +27,7 @@
 from ..run import get_run
 from ...core_python import *
 from ...core_python.api.experiment import update_experiment_state
+from ...core_python.utils.timer import Timer
 from ...log.type import LogData
 
 
@@ -36,6 +37,7 @@ class CloudPyCallback(SwanLabRunCallback):
     def __init__(self):
         super().__init__()
         self.executor = ThreadPoolExecutor(max_workers=1)
+        self.heartbeat: Optional[Timer] = None
 
     def __str__(self):
         return "SwanLabCloudPyCallback"
@@ -62,12 +64,15 @@ def _converter_summarise_metric():
         pass
 
     def on_init(self, *args, **kwargs):
-        _ = self._create_client()
+        self._create_client()
         # 检测是否有最新的版本
         U.check_latest_version()
+        # 挂载项目、实验
         with Status("Creating experiment...", spinner="dots"):
             with Mounter() as mounter:
                 mounter.execute()
+        # 创建客户端心跳
+        self.heartbeat = create_client_heartbeat()
 
     def _terminal_handler(self, log_data: LogData):
         self.porter.trace_log(log_data)
@@ -100,6 +105,10 @@ def on_metric_create(self, metric_info: MetricInfo, *args, **kwargs):
         self.porter.trace_metric(metric_info)
 
     def on_stop(self, error: str = None, *args, **kwargs):
+        # 删除心跳
+        self.heartbeat.cancel()
+        self.heartbeat.join()
+        # 删除终端代理和系统回调
         success = get_run().success
         # FIXME 等合并 swankit 以后优化一下 interrupt 的传递问题
         interrupt = kwargs.get("interrupt", False)