Missing GL version error on docker windows container while running pyvista

[IrakozeFD]

Hi,

I am trying to build a docker image based on windows container (mcr.microsoft.com/windows:20H2) to run on it a python project, and I am getting issues running pyvista in the built container. In fact I have used this repository script to set up OpenGL related features like headless displays in my windows container, but when trying to run tests involving Pyvista I am getting windows fatal exception error which I think it is related to this below :

Here the pyvista report about the windows container:

Date: Tue May 10 11:17:18 2022 Paris, Madrid (heure d’été)

            OS : Windows
        CPU(s) : 2
       Machine : AMD64
  Architecture : 64bit
   Environment : Python
   GPU Details : error

Python 3.8.6 (tags/v3.8.6:db45529, Sep 23 2020, 15:52:53) [MSC v.1927 64 bit
(AMD64)]

       pyvista : 0.34.1
           vtk : 9.0.3
         numpy : 1.22.3
       imageio : 2.19.0
       appdirs : 1.4.4
        scooby : 0.5.12
    matplotlib : 3.5.1
     pyvistaqt : 0.5.0
         PyQt5 : 5.15.6
         scipy : 1.8.0
        meshio : 5.3.4

---

Docker version 20.10.14, build a224086.

Thank you,

[banesullivan]

Do you have a repo where this CI is running?

[IrakozeFD]

The docker image that I want to build will be in fact used for CI in Pyleecan.

[banesullivan]

Can you point me to the source for that docker image? i.e., the Dockerfile?

[IrakozeFD]

Please find it here
Thank you.

[banesullivan]

So what exactly are you trying to do, where are you using this Action, and how can we reproduce the error you are encountering?

[banesullivan]

I'm not sure how you plan on using the mcr.microsoft.com/windows:20H2 image in GitHub Actions CI unless you have self-hosted runners?

[banesullivan]

Maybe just adding something like the following to your docker image would work?

Copy https://github.com/pyvista/setup-headless-display-action/blob/main/windows/install_opengl.ps1

COPY install_opengl.ps1 install_opengl.ps1
RUN Set-StrictMode -Version Latest \
    $ErrorActionPreference = "Stop" \
    $PSDefaultParameterValues['*:ErrorAction']='Stop' \
    powershell install_opengl.ps1

[IrakozeFD]

The docker file does not reflect, as it is right now, how I want to use the image but the testing process can be done with the following lines:

• clone pyleecan repository
• install all the requiremeents : python -m pip install -r pyleecan/requirements.tct
• run Tests with Pytest from the working Dir: python -m pytest pyleecan/Tests

I was just creating the environment and testing with the container CLI on docker desktop to validate the process.

[FelipeCybis]

Maybe just adding something like the following to your docker image would work?

Copy https://github.com/pyvista/setup-headless-display-action/blob/main/windows/install_opengl.ps1

COPY install_opengl.ps1 install_opengl.ps1
RUN Set-StrictMode -Version Latest \
    $ErrorActionPreference = "Stop" \
    $PSDefaultParameterValues['*:ErrorAction']='Stop' \
    powershell install_opengl.ps1

BTW, I had the same issue in a self-hosted windows container.

I am using FROM mcr.microsoft.com/windows/server:ltsc2022 and even the administrator profile cannot remove opengl32.dll.

My workaround (maybe I should say chatgpt's workaround :P) is downloading install-opengl.ps1 and using in my Dockerfile as you suggested, but changing this (commented lines are the current code)

    # takeown /F $filepath /A
    # icacls $filepath /grant "${env:ComputerName}\${env:UserName}:F"
    $acl = Get-Acl $filepath
    $acl.SetAccessRuleProtection($false, $false)
    $rule = New-Object System.Security.AccessControl.FileSystemAccessRule("Users", "FullControl", "Allow")
    $acl.AddAccessRule($rule)
    Set-Acl $filepath $acl

Maybe this comment #5 (comment) is also related to this in the first place and not directly to the old OSMesa version.

I am still using the action just for the part where it sets up the PYVISTA_OFF_SCREEN env variable and I am curious if this change would actually work in the action stage (instead of the docker build stage).