[viewer][macOS] Can't use env.replay/render because of a python multiprocessing error

[mwulfman]

I am using the tagged version 1.8.15.post1. I see only this issue on macOS (mac M1 pro) and it could be due to my laptop config or some issue with the C/Cpp compiler. I've created a simple hello_world.py reproduction script:

from gym_jiminy.envs.acrobot import AcrobotJiminyEnv

print("hello world")
env = AcrobotJiminyEnv()
env.reset()
env.replay()

This is the ouput I get from:

python hello_world.py

WARNING:root:Boost.Python cannot be shared between modules. Importing bundled jiminy dependencies instead of system-wide install as fallback.
hello world
WARNING:root:Boost.Python cannot be shared between modules. Importing bundled jiminy dependencies instead of system-wide install as fallback.
hello world
Traceback (most recent call last):
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/viewer.py", line 1214, in connect_backend
    client = Panda3dViewer(window_type='onscreen',
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/panda3d/panda3d_visualizer.py", line 1928, in __init__
    self._app = Panda3dProxy(config)
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/panda3d/panda3d_visualizer.py", line 1755, in __init__
    self.start()
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/process.py", line 121, in start
    self._popen = self._Popen(self)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/context.py", line 224, in _Popen
    return _default_context.get_context().Process._Popen(process_obj)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/context.py", line 288, in _Popen
    return Popen(process_obj)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/popen_spawn_posix.py", line 32, in __init__
    super().__init__(process_obj)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/popen_fork.py", line 19, in __init__
    self._launch(process_obj)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/popen_spawn_posix.py", line 42, in _launch
    prep_data = spawn.get_preparation_data(process_obj._name)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/spawn.py", line 154, in get_preparation_data
    _check_not_importing_main()
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/spawn.py", line 134, in _check_not_importing_main
    raise RuntimeError('''
RuntimeError: 
        An attempt has been made to start a new process before the
        current process has finished its bootstrapping phase.

        This probably means that you are not using fork to start your
        child processes and you have forgotten to use the proper idiom
        in the main module:

            if __name__ == '__main__':
                freeze_support()
                ...

        The "freeze_support()" line can be omitted if the program
        is not going to be frozen to produce an executable.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/viewer.py", line 611, in __init__
    Viewer.connect_backend(backend)
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/viewer.py", line 292, in fun_safe
    return fun(*args, **kwargs)
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/viewer.py", line 1221, in connect_backend
    raise RuntimeError(
RuntimeError: Something went wrong. Impossible to instantiate viewer backend.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/spawn.py", line 116, in spawn_main
    exitcode = _main(fd, parent_sentinel)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/spawn.py", line 125, in _main
    prepare(preparation_data)
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/spawn.py", line 236, in prepare
    _fixup_main_from_path(data['init_main_from_path'])
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/multiprocessing/spawn.py", line 287, in _fixup_main_from_path
    main_content = runpy.run_path(main_path,
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/runpy.py", line 289, in run_path
    return _run_module_code(code, init_globals, run_name,
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/runpy.py", line 96, in _run_module_code
    _run_code(code, mod_globals, init_globals,
  File "/Users/mathiasw/.pyenv/versions/3.10.14/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/Users/mathiasw/Code/misc/github/jiminy/hello_world.py", line 7, in <module>
    env.replay()
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/gym_jiminy/common/envs/generic.py", line 1109, in replay
    self.simulator.render(
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/simulator.py", line 747, in render
    viewer = Viewer(
  File "/Users/mathiasw/.pyenv/versions/jiminy/lib/python3.10/site-packages/jiminy_py/viewer/viewer.py", line 648, in __init__
    raise RuntimeError(
RuntimeError: Impossible to create backend or connect to it.

render/replay triggers another run of the script ?? If I copy-paste the lines in an ipython, there is no problem. But if I do run hello_world.py the same problem appears.

I have no issue running the examples/rllib/acrobot_ppo.pyscripts either. I am not sure how to interpret this problem and can't find much help online. The problem remains when I change the backend to meshcat. I was hoping someone would know something about it or encountered the same issue.

Thanks for your help!

EDIT, this script reproduce the problem, the call self.start in Panda3dProxy.init inPanda3dViewer(window_type='onscreen', window_title=Viewer.window_name) called by connect_backend restarts the script:

from gym_jiminy.envs.acrobot import AcrobotJiminyEnv
from jiminy_py.viewer import Viewer

print("hello world")
env = AcrobotJiminyEnv()
env.reset()

robot_name = env.engine.robots[0].name
robot = env.engine.robots[0]
viewer_kwargs = {'backend': None,
                 'delete_robot_on_close': True,
                 'camera_pose': ((0.0, 8.0, 0.0),
                                 (1.5707963267948966, 0.0, 3.141592653589793),
                                 None),
                 'robot_name': None}

viewer = Viewer(robot,
                use_theoretical_model=False,
                open_gui_if_parent=False,
                **{"robot_name": robot_name, **viewer_kwargs})

or even simpler (what does Panda3dProxy):

import multiprocessing as mp

print("hello world")
proc = mp.Process()
proc._host_conn, proc._proc_conn = mp.Pipe()
proc.start()

[duburcqa]

Yes, this behaviour is unfortunately expected on MacOS and Windows. The viewer is running in another process to prevent the GIL from getting on the way and slowdowning the main thread when the viewer is running. fork multiprocessing mode is not available on Windows and not the default on MacOS because it is not reliable on this OS (and pretty fragile overall). Because of this limitation, spawn is generally recommended. However, the latter requires guarding the main script behind if __name__ == '__main__':. This is not necessary with ipython because it hacking the multiprocessing library to use some kind of asyncio mechanism instead, so it manages sub-processes completely differently.

[mwulfman]

Thank you very much for your explanation and your help ! It works now ! 🤖