FIX AutoDL修复exception导致worker停止问题

AutoDL/autodl_scheduling/server.py

@@ -61,13 +61,16 @@ class Server:
     def introspect_instance(self):
         loguru.logger.info("Introspecting worker started || Scaledown Window: %ds" % self.instance_pool.scaledown_window)
         while True:
-            self.instance_pool.introspection()
-            time.sleep(1)
+            try:
+                self.instance_pool.introspection()
+                time.sleep(1)
+            except:
+                traceback.print_exc()
 
     def scaling_worker(self):
         loguru.logger.info("Scaling worker started")
-        try:
-            while True:
+        while True:
+            try:
                 # 提交任务
                 self.instance_pool.scale_instance(self.waiting_queue.get_size()+self.running_pool.get_running_size(), disable_shrink=True)
                 for instance in self.instance_pool.instances:
@@ -104,8 +107,8 @@ class Server:
                                     instance.active = False
                                     instance.last_active_time = time.time()
                 time.sleep(0.5)
-        except:
-            traceback.print_exc()
+            except:
+                traceback.print_exc()
 
 
 if __name__=="__main__":