Merge remote-tracking branch 'origin/develop-fix-stall' into asymm-envs

ml-agents/mlagents/trainers/trainer/rl_trainer.py

 from typing import Dict, List
 from collections import defaultdict
 import abc
+import time
 
 from mlagents.trainers.optimizer.tf_optimizer import TFOptimizer
 from mlagents.trainers.buffer import AgentBuffer
                 # We grab at most the maximum length of the queue.
                 # This ensures that even if the queue is being filled faster than it is
                 # being emptied, the trajectories in the queue are on-policy.
+                _queried = False
+                    _queried = True
+                if self.threaded and not _queried:
+                    # Yield thread to avoid busy-waiting
+                    time.sleep(0.0001)
         if self.should_still_train:
             if self._is_ready_update():
                 with hierarchical_timer("_update_policy"):

ml-agents/mlagents/trainers/trainer_controller.py

 
         parsed_behavior_id = BehaviorIdentifiers.from_name_behavior_id(name_behavior_id)
         brain_name = parsed_behavior_id.brain_name
+        trainerthread = None
+            if trainer.threaded:
+                # Only create trainer thread for new trainers
+                trainerthread = threading.Thread(
+                    target=self.trainer_update_func, args=(trainer,), daemon=True
+                )
+                self.trainer_threads.append(trainerthread)
 
         policy = trainer.create_policy(
             parsed_behavior_id, env_manager.external_brains[name_behavior_id]
 
         trainer.publish_policy_queue(agent_manager.policy_queue)
         trainer.subscribe_trajectory_queue(agent_manager.trajectory_queue)
-        if trainer.threaded:
-            # Start trainer thread
-            trainerthread = threading.Thread(
-                target=self.trainer_update_func, args=(trainer,), daemon=True
-            )
+
+        # Only start new trainers
+        if trainerthread is not None:
-            self.trainer_threads.append(trainerthread)
 
     def _create_trainers_and_managers(
         self, env_manager: EnvManager, behavior_ids: Set[str]