OpenGL EGLError - EGL_NOT_INITIALIZED

[jxxtin]

Hello, @pointW

I am running a training script on a headless Ubuntu server, and I encounter the following OpenGL error during the validation phase:

OpenGL.raw.EGL._errors.EGLError: EGLError(
        err = EGL_NOT_INITIALIZED,
        baseOperation = eglDestroyContext,
        cArguments = (
                <OpenGL._opaque.EGLDisplay_pointer object at 0x7f440c46ebc0>,
                <OpenGL._opaque.EGLContext_pointer object at 0x7f4360c6f840>,
        ),
        result = 0
)

Could you provide guidance on how to properly resolve this issue?

[pointW]

Hi @jxxtin, did you install all of the required packages for mujoco? What version of Ubuntu are you running on?

One solution could be running with osmesa instead:

# to train with batch size of 64 and hidden dimension of 64
MUJOCO_GL=osmesa PYOPENGL_PLATTFORM=osmesa python train.py --config-name=train_equi_diffusion_unet_abs task_name=stack_d1 policy.enc_n_hidden=64 dataloader.batch_size=64

You can also check whether your code is using spawn for multi-processing:

equidiff/equi_diffpo/gym_util/async_vector_env.py

Line 11 in b67213a

mp.set_start_method('spawn', force=True)

I have seen some similar error without using spawn

[jxxtin]

hello @pointW

My environment setup is as follows:

Ubuntu 22.04.3  
Mujoco 2.3.2  

I have also checked that spawn is working.

When running the code you suggested, I encountered the following error:

Would you be able to help me identify what I might be doing wrong?

Thanks for your help!

[pointW]

Hmm, that's weird, could you add HYDRA_FULL_ERROR=1 and show me the full error message?

[jxxtin]

Hellow @pointW,

After exporting HYDRA_FULL_ERROR=1, when running the following command:

MUJOCO_GL=osmesa PYOPENGL_PLATFORM=osmesa python train.py --config-name=train_equi_diffusion_unet_abs task_name=stack_d1 policy.enc_n_hidden=64 dataloader.batch_size=64

I encountered the following error.

equidiff git:(main) ✗ MUJOCO_GL=osmesa PYOPENGL_PLATTFORM=osmesa python train.py --config-name=train_equi_diffusion_unet_abs task_name=stack_d1
ROBOMIMIC WARNING(
    No private macro file found!
    It is recommended to use a private macro file
    To setup, run: python /root/miniforge3/envs/equidiff/lib/python3.9/site-packages/robomimic/scripts/setup_macros.py
)
[2025-02-06 13:46:54,441][equi_diffpo.model.diffusion.conditional_unet1d][INFO] - number of parameters: 2.552895e+08
Enc params: 1.588464e+07
Diff params: 2.552909e+08
Acquiring lock on cache.
Loading cached ReplayBuffer from Disk.
Loaded!
[robosuite WARNING] No private macro file found! (macros.py:53)
[2025-02-06 13:46:59,409][robosuite_logs][WARNING] - No private macro file found!
[robosuite WARNING] It is recommended to use a private macro file (macros.py:54)
[2025-02-06 13:46:59,409][robosuite_logs][WARNING] - It is recommended to use a private macro file
[robosuite WARNING] To setup, run: python /root/miniforge3/envs/equidiff/lib/python3.9/site-packages/robosuite/scripts/setup_macros.py (macros.py:55)
[2025-02-06 13:46:59,409][robosuite_logs][WARNING] - To setup, run: python /root/miniforge3/envs/equidiff/lib/python3.9/site-packages/robosuite/scripts/setup_macros.py
Error executing job with overrides: ['task_name=stack_d1']

    from mujoco.gl_context import *
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/mujoco/gl_context.py", line 38, in <module>
    from mujoco.osmesa import GLContext as _GLContext
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/mujoco/osmesa/__init__.py", line 24, in <module>
    raise ImportError(
ImportError: Cannot use OSMesa rendering platform. The PYOPENGL_PLATFORM environment variable is set to 'egl' (should be either unset or 'osmesa').

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

Traceback (most recent call last):
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py", line 134, in _resolve_target
    target = _locate(target)
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/utils.py", line 653, in _locate
    raise ImportError(
ImportError: Error loading 'equi_diffpo.env_runner.robomimic_image_runner.RobomimicImageRunner':
ImportError("Cannot use OSMesa rendering platform. The PYOPENGL_PLATFORM environment variable is set to 'egl' (should be either unset or 'osmesa').")

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

Traceback (most recent call last):
  File "/root/projects/equidiff/train.py", line 78, in <module>
    main()
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/main.py", line 90, in decorated_main
    _run_hydra(
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/utils.py", line 389, in _run_hydra
    _run_app(
    lambda: hydra.run(
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/hydra.py", line 132, in run
    _ = ret.return_value
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/core/utils.py", line 260, in return_value
    raise self._return_value
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/core/utils.py", line 186, in run_job
    ret.return_value = task_function(task_cfg)
  File "/root/projects/equidiff/train.py", line 75, in main
    workspace.run()
  File "/root/projects/equidiff/equi_diffpo/workspace/train_equi_workspace.py", line 113, in run
    env_runner = hydra.utils.instantiate(
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py", line 222, in instantiate
    return instantiate_node(
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py", line 325, in instantiate_node
    _target_ = _resolve_target(node.get(_Keys.TARGET), full_key)
  File "/root/miniforge3/envs/equidiff/lib/python3.9/site-packages/hydra/_internal/instantiate/_instantiate2.py", line 139, in _resolve_target
    raise InstantiationException(msg) from e
hydra.errors.InstantiationException: Error locating target 'equi_diffpo.env_runner.robomimic_image_runner.RobomimicImageRunner', see chained exception above.
full_key: task.env_runner

Thank you for your help!

[pointW]

It looks like you had ‘PYOPENGL_PLATTFORM=osmesa’, there is an extra T

[jxxtin]

Thank you so much. The issue has been resolved! Since I copied the command from the README, I think updating its content should be enough.

[pointW]

Ah I see, this is a typo in the readme, I just fixed it. Sry about that