Hanging Manager or worker

[deepdad]

Hi Adam, when both the manager and the worker seem to just be staring each other down, nothing much will happen.

I have cobbled together a main program here using CpuSampler: https://github.com/deepdad/SharedWorldModels/blob/parallel/dreamer-pytorch/main_parallel.py

This ties together a number of ideas, but where it hangs is in RLPyt. That doesn't mean that RLPyt is to blame, but I'm stuck as well.

In rlpyt/samplers/buffer.py, line 17, I change

if subprocess:

to

if False:

because the code in that block hangs on mp.Manager(). I guess the manager starts to wait for worker threads to send (sampled) data to it, but they never send anything.
We use RLBench, which is a set of CoppeliaSim tasks. With the above commented out, sampling_process in rlpyt/samplers/parallel/worker.py gets called but it hangs on line 50 envs = [c.EnvCls(**c.env_kwargs) for _ in range(w.n_envs)]

I can run it with headless=False. I see the robot make one action and then nothing happens.

It's difficult to debug. I would be nice if I could get you to log in to our Gcloud machine over vnc where PyCharm is running.