first try

swanlab/data/cloud/log_collector.py

@@ -15,12 +15,17 @@
 from swanlab.error import SyncError
 from .task_types import UploadType
 
+from psutil import cpu_percent
+from swanlab.data.run.main import get_run
+from swanlab.data.modules import MediaType, DataWrapper, FloatConvertible, Line
+
 
 class LogCollectorTask(ThreadTaskABC):
     """
     日志聚合器，负责收集所有线程注册的日志信息
     并且定义日志上传接口
     """
+
     UPLOAD_TIME = 1
     """
     每隔多少秒上传一次日志
@@ -114,6 +119,21 @@ def task(self, u: ThreadUtil, *args):
                 swanlog.error(f"upload error: {e}")
             self.lock = False
 
+    def hardware_task(self, u: ThreadUtil, *args):
+        """
+        定时任务，读取硬件信息添加到记录
+        :param u: 线程工具类
+        """
+        cpu_percent_now = float(cpu_percent())
+        run = get_run()
+        if run == None:
+            return
+        else:
+            exp = run.exp
+            print(f'My exp = {exp}')
+            exp.add(key="CPU", data=DataWrapper("CPU", [Line(cpu_percent_now)]), step=None)
+        print(f'Now CPU usage: {cpu_percent_now}%')
+
     def callback(self, u: ThreadUtil, *args):
         """
         回调函数，用于结束时的回调

swanlab/data/cloud/start_thread.py

@@ -46,20 +46,24 @@ def __init__(self, upload_sleep_time: float = None):
             args=(),
             name=self.UPLOAD_THREAD_NAME,
             sleep_time=upload_sleep_time,
-            callback=self.collector.callback
+            callback=self.collector.callback,
         )
+
+        # 生成硬件指标上传线程，此线程包含数据记录任务，作为和log子线程同级的子线程
+        self.hardware_thread = self.create_thread(
+            target=self.collector.hardware_task,
+            args=(),
+            name="HardwareUploader",
+            sleep_time=upload_sleep_time,
+        )
+
         self.queue = LogQueue(queue=self.__queue, readable=False, writable=True)
         """
         一个线程安全的队列，用于主线程向数据上传线程通信
         """
 
     def create_thread(
-        self,
-        target: Callable,
-        args: Tuple = (),
-        name: str = None,
-        sleep_time: float = None,
-        callback: Callable = None
+        self, target: Callable, args: Tuple = (), name: str = None, sleep_time: float = None, callback: Callable = None
     ) -> threading.Thread:
         """
         创建一个线程
@@ -83,11 +87,7 @@ def create_thread(
         thread_util = ThreadUtil(q, name)
         callback = ThreadUtil.wrapper_callback(callback, (thread_util, *args)) if callback is not None else None
         task = self._create_loop(name, sleep_time, target, (thread_util, *args))
-        thread = threading.Thread(
-            target=task,
-            daemon=True,
-            name=name
-        )
+        thread = threading.Thread(target=task, daemon=True, name=name)
         self.thread_pool[name] = thread
         if callback is not None:
             self.__callbacks.append(callback)
@@ -101,13 +101,7 @@ def sub_threads(self):
         """
         return {name: thread for name, thread in self.thread_pool.items() if name != self.UPLOAD_THREAD_NAME}
 
-    def _create_loop(
-        self,
-        name: str,
-        sleep_time: float,
-        task: Callable,
-        args: Tuple[ThreadUtil, ...]
-    ) -> [Callable]:
+    def _create_loop(self, name: str, sleep_time: float, task: Callable, args: Tuple[ThreadUtil, ...]) -> [Callable]:
         """
         创建一个事件循环，循环执行传入线程池的任务
         :param name: 线程名称

swanlab/data/run/main.py

@@ -25,6 +25,7 @@
 
 MAX_LIST_LENGTH = 108
 
+
 class SwanLabRunState(Enum):
     """SwanLabRunState is an enumeration class that represents the state of the experiment.
     We Recommend that you use this enumeration class to represent the state of the experiment.
@@ -321,6 +322,7 @@ def log(self, data: dict, step: int = None):
                 v = DataWrapper(k, [Line(v)])
             # 数据类型的检查将在创建chart配置的时候完成，因为数据类型错误并不会影响实验进行
             metric_info = self.__exp.add(key=k, data=v, step=step)
+            print(f'Corr exp = {self.__exp}')
             self.__operator.on_metric_create(metric_info)
             log_return[metric_info.key] = metric_info