Using mujoco rendering on CircleCI

[vmoens]

We're trying to use mujoco (specifically dm_control)-based environments in our tests on torchrl.

We'd like to use the egl backend, as osmesa is slow (it often stays idle for a very long time) and glfw is not available in headless environments.

We have installed all the required dependencies as pointed here but the following error still occurs:

Collecting dm_control
  Downloading dm_control-1.0.4.tar.gz (38.2 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 38.2/38.2 MB 67.6 MB/s eta 0:00:00
  Installing build dependencies: started
  Installing build dependencies: finished with status 'done'
  Getting requirements to build wheel: started
  Getting requirements to build wheel: finished with status 'error'

Pip subprocess error:
  error: subprocess-exited-with-error
  
  × Getting requirements to build wheel did not run successfully.
  │ exit code: 1
  ╰─> [25 lines of output]
      Traceback (most recent call last):
        File "/root/project/env/lib/python3.8/site-packages/pip/_vendor/pep517/in_process/_in_process.py", line 363, in <module>
          main()
        File "/root/project/env/lib/python3.8/site-packages/pip/_vendor/pep517/in_process/_in_process.py", line 345, in main
          json_out['return_val'] = hook(**hook_input['kwargs'])
        File "/root/project/env/lib/python3.8/site-packages/pip/_vendor/pep517/in_process/_in_process.py", line 130, in get_requires_for_build_wheel
          return hook(config_settings)
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/setuptools/build_meta.py", line 177, in get_requires_for_build_wheel
          return self._get_build_requires(
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/setuptools/build_meta.py", line 159, in _get_build_requires
          self.run_setup()
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/setuptools/build_meta.py", line 174, in run_setup
          exec(code, locals())
        File "<string>", line 26, in <module>
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/mujoco/__init__.py", line 52, in <module>
          from mujoco.egl import GLContext
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/mujoco/egl/__init__.py", line 31, in <module>
          from mujoco.egl import egl_ext as EGL
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/mujoco/egl/egl_ext.py", line 27, in <module>
          from OpenGL import EGL
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/OpenGL/EGL/__init__.py", line 2, in <module>
          from OpenGL.raw.EGL._types import *
        File "/tmp/pip-build-env-012xq985/overlay/lib/python3.8/site-packages/OpenGL/raw/EGL/_types.py", line 87, in <module>
          raw_eglQueryString = _p.PLATFORM.EGL.eglQueryString
      AttributeError: 'NoneType' object has no attribute 'eglQueryString'

The strangest thing is that ssh-ing on the machine and executing the same code resolves the issue.

Any help would be appreciated.

[saran-t]

The strangest thing is that ssh-ing on the machine and executing the same code resolves the issue.

This implies that your machine has everything needed installed already. I don't know what to suggest since I don't know CircleCI, but if I'm investigating it I'd start by dumping all environment variables from your SSH session and on the CI workflow and see if there's any difference there.

[vmoens]

Thanks for the tip! We solved the issue by not relying on docker to run the tests (which was messing up the env variables).

See this for more context.