Can't select GPU for EGL headless rendering

[guy4261]

TL;DR - how can I select the GPU that the EGL headless rendering runs on? I want to use /dev/nvidia2 but it seems only /dev/nvidia0 is supported.

===

I have two machines in the cloud that are Docker containers of the same image (Ubuntu). When I run the code from the EGL offscreen rendering colab notebook as a Python script, one renders the bottle well, while the other gives me the error below. The working machine has its GPU mounted on /dev/nvidia0 while on the faulty machine it is on /dev/nvidia2.

It seems that egl_display = egl.eglGetDisplay(egl.EGL_DEFAULT_DISPLAY) tries picking on /dev/nvidia0 and fails if it is not available. Since I can't control where will the GPU get mounted on my machine, is there a way to make OpenGL/EGL pick the other device? I tried multiple solutions found around the web, from different environment variables to creating a symlink, but to no avail.

The error:

Traceback (most recent call last):
  File "colab", line 30, in <module>
    r = pyrender.OffscreenRenderer(640, 480)
  File "/miniconda/envs/foo/lib/python3.6/site-packages/pyrender/offscreen.py", line 31, in __init__
    self._create()
  File "/miniconda/envs/foo/lib/python3.6/site-packages/pyrender/offscreen.py", line 134, in _create
    self._platform.init_context()
  File "/miniconda/envs/foo/lib/python3.6/site-packages/pyrender/platforms/egl.py", line 177, in init_context
    assert eglInitialize(self._egl_display, major, minor)
  File "/miniconda/envs/foo/lib/python3.6/site-packages/OpenGL/platform/baseplatform.py", line 402, in __call__
    return self( *args, **named )
  File "/miniconda/envs/foo/lib/python3.6/site-packages/OpenGL/error.py", line 232, in glCheckError
    baseOperation = baseOperation,
OpenGL.error.GLError: GLError(
	err = 12290,
	baseOperation = eglInitialize,
	cArguments = (
		<OpenGL._opaque.EGLDisplay_pointer object at 0x7f7c9f107400>,
		c_long(0),
		c_long(0),
	),
	result = 0
)

[guy4261]

Does it make sense that if my nvidia device is /dev/nvidia1 or /dev/nvidia2 then it doesn't get picked up as the egl.EGL_DEFAULT_DISPLAY? Is there an environment variable I should set for this case? (Like I'm setting PYOPENGL_PLATFORM=egl before import pyrender?)

[Guptajakala]

@guy4261 Hi, I'm meeting the same issue. Have you resolve it?
On my machine I indeed have /dev/nvidia0 but still get that error. Are you sure your two machines have the only difference in that /dev/nvidia id?

[guy4261]

@Guptajakala sorry - 1. I didn't solve this, 2. I do believe that on my machine it's the /dev/nvidiaN issue, perhaps you are experiencing something else I can't help with :(

[keunhong]

I submitted a PR (#60) a couple months ago that does this.

Just set the environment variable EGL_DEVICE_ID e.g.

export EGL_DEVICE_ID=1
python run.py

[guy4261]

Great @keunhong !!! Thanks!!!
Does this mean I should for now build from master rather then install with pip?

[Guptajakala]

@guy4261 I dont think you need. Just run export EGL_DEVICE_ID=1 before you run your python file.
However, it still doesn't work in my case, maybe my server has some wierd things. Please let me know if it works for you.

[keunhong]

Great @keunhong !!! Thanks!!!
Does this mean I should for now build from master rather then install with pip?

Looks like the commit should be in 0.1.33 so shouldn't be necessary.

[oarriaga]

I also had the following issue:

OpenGL.error.GLError: GLError( err = 12290,

It was related to incompatible versions of pyrender and PyOpenGL.

However, is not working for me using headless rendering with EGL. I now have the following error

[grgkopanas]

@oarriaga What was the incompatibility issue and what was eventually the compatible versions that you used?

[vstadnytskyi-FDA]

I submitted a PR (#60) a couple months ago that does this.

Just set the environment variable EGL_DEVICE_ID e.g.

export EGL_DEVICE_ID=1
python run.py

How do I do this on Windows 11?