rm stop field(no need)

BalaBalaYi · Jun 26, 2024 · 4c20d5d · 4c20d5d
1 parent 494b046
commit 4c20d5d
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 5 deletions.
diff --git a/dlrover/python/master/node/job_auto_scaler.py b/dlrover/python/master/node/job_auto_scaler.py
@@ -75,7 +75,6 @@ class JobAutoScaler(metaclass=ABCMeta):
 
  def __init__(self):
  self._suggested_stop = False
- self._stop_autoscaling = False
  self._autoscaling_started = False
 
  def suggested_stop(self):
@@ -157,7 +156,7 @@ def start_auto_scaling(self):
  ).start()
 
  def stop_auto_scaling(self):
- self._stop_autoscaling = True
+ self._autoscaling_started = False
 
  def _periodic_optimize_running_resource(self):
  """Adjust job resource periodically and stop adjustment
@@ -167,7 +166,7 @@ def _periodic_optimize_running_resource(self):
  last_plan_time = 0
  opt_interval = _dlrover_context.seconds_interval_to_optimize
  while True:
- if self._stop_autoscaling:
+ if not self._autoscaling_started:
  logger.info("Stop auto-scaling thread for PS Training.")
  break
  if (
@@ -286,13 +285,13 @@ def start_auto_scaling(self):
  ).start()
 
  def stop_auto_scaling(self):
- self._stop_autoscaling = True
+ self._autoscaling_started = False
 
  def _periodic_adjust_worker(self):
  """Periodicaly adjust the number of worker."""
  logger.info("Start auto-scaling thread for AllReduce Training.")
  while True:
- if self._stop_autoscaling:
+ if not self._autoscaling_started:
  logger.info("Stop auto-scaling thread for AllReduce Training.")
  break
  time.sleep(self._scale_interval)

diff --git a/dlrover/python/tests/test_job_auto_scaler.py b/dlrover/python/tests/test_job_auto_scaler.py
@@ -81,7 +81,9 @@ def test_execute_job_optimization_plan(self):
  ps_addrs.append("test-edljob-ps-{}.default.svc:2222".format(i))
  self.assertListEqual(scale_plan.ps_addrs, ps_addrs)
  auto_scaler.start_auto_scaling()
+ self.assertTrue(auto_scaler._autoscaling_started)
  auto_scaler.stop_auto_scaling()
+ self.assertFalse(auto_scaler._autoscaling_started)
 
  def test_reduce_timeout_pending_node_resource(self):
  params = MockK8sPSJobArgs()
@@ -142,4 +144,6 @@ def test_execute_job_optimization_plan(self):
  alive_num = auto_scaler._get_alive_worker_num()
  self.assertEqual(alive_num, 16)
  auto_scaler.start_auto_scaling()
+ self.assertTrue(auto_scaler._autoscaling_started)
  auto_scaler.stop_auto_scaling()
+ self.assertFalse(auto_scaler._autoscaling_started)