[RLlib] Subprocess in my customize env when tunning

How severe does this issue affect your experience of using Ray?

  • High: It blocks me to complete my task.

Hi! When I called subprocess.Popen to invoke a program with multithread optimization inside my customize environment, I found that the performance is as bad as singlethread. But when calling the program with a single python script, the performance becomes normal.
Does the RolloutWalker of ray.tune limits the leverage of cpu of its subprocess?
If so, how can I avoid this kind of problem?
Thank you!