repen heartbeat ut (#27684)

paddle/fluid/operators/distributed/heart_beat_monitor.cc

@@ -84,7 +84,7 @@ void HeartBeatMonitor::LostWorkerMonitor() {
       }
     }
 
-    std::this_thread::sleep_for(std::chrono::milliseconds(30 * 1000));
+    std::this_thread::sleep_for(std::chrono::milliseconds(10 * 1000));
   }
   VLOG(1) << "worker heartbeat monitor stopped, thread exit";
 }

paddle/fluid/operators/distributed/heart_beat_monitor_test.cc

@@ -23,32 +23,30 @@ namespace distributed {
 void run(HeartBeatMonitor* monitor) { monitor->LostWorkerMonitor(); }
 
 TEST(HeartBeatMonitor, All) {
-  // (tangwei12) fix it soon.
+  int trainers = 10;
-  return;
+  int pserver_id = 0;
-  //  int trainers = 10;
+  std::string var = "nce_w@GRAD.block0";
-  //  int pserver_id = 0;
+  std::string var2 = "nce_w@GRAD.block2";
-  //  std::string var = "nce_w@GRAD.block0";
+
-  //  std::string var2 = "nce_w@GRAD.block2";
+  HeartBeatMonitor::Init(trainers, pserver_id == 0, var);
-  //
+
-  //  HeartBeatMonitor::Init(trainers, pserver_id == 0, var);
+  auto* monitor = HeartBeatMonitor::GetInstance();
-  //
+
-  //  auto* monitor = HeartBeatMonitor::GetInstance();
+  std::vector<int> ids{1, 3, 5, 7};
-  //
+
-  //  std::vector<int> ids{1, 3, 5, 7};
+  for (auto& id : ids) {
-  //
+    monitor->Update(id, var, RUNNING);
-  //  for (auto& id : ids) {
+  }
-  //    monitor->Update(id, var, RUNNING);
+
-  //  }
+  monitor->Update(9, var2, RUNNING);
-  //
+  monitor->Update(2, var, COMPLETED);
-  //  monitor->Update(9, var2, RUNNING);
+
-  //  monitor->Update(2, var, COMPLETED);
+  std::thread t(run, monitor);
-  //
+  t.detach();
-  //  std::thread t(run, monitor);
+
-  //  t.detach();
+  std::this_thread::sleep_for(std::chrono::milliseconds(15 * 1000));
-  //
+
-  //  std::this_thread::sleep_for(std::chrono::milliseconds(45 * 1000));
+  monitor->Stop();
-  //
-  //  monitor->Stop();
 }
 
 }  // namespace distributed