Segfault upon calling vedo.show on server

[ldouteau]

Hi,

I'm currently implementing some reg tests on an application that uses Vedo for visualization purposes. The test is very simple, just trying to make sure that my object gets created properly. I set up Vedo objects, call show (must not be very clever), that's it. Tests run properly on my laptop, both on Windows & WSL2. Using Python 3.11 & Pytest, Vedo 2024.5.1

When i tried to run CI on our Gitlab server, I got a segmentation fault at test collection. For some reason, upon collecting tests that include a vedo.show, Pytest segfaults immediately. After some testing, running the following piece of code on our sever (Ubuntu 22.04 LTS) generates a segfault, even though it takes a bit longer to get it. Thus my current workaround is to skip these reg tests when running CI.

import vedo

def test_show():
    pts = vedo.Points([[1, 2, 3]])
    vedo.show(pts).close()
    print("success")

if __name__ == "__main__":
    test_show()

The logs I get are below. The visualization libraries (libGL) are installed on the server. As i don't get this failure in other platform i assume it's related to the absence of X11 forwarding or something (I know nothing about it sorry). I tried to track what happens in VTK, but couldn't find a better minimal example.

vedo_test.py Fatal Python error: Segmentation fault

Current thread 0x00007f87749e7000 (most recent call first):
  File "~/vedo_test/.venv/lib/python3.10/site-packages/vedo/plotter.py", line 892 in initialize_interactor
  File "~/vedo_test/.venv/lib/python3.10/site-packages/vedo/plotter.py", line 3373 in show
  File "~/vedo_test/.venv/lib/python3.10/site-packages/vedo/plotter.py", line 344 in show

I understand that the procedure i followed has several caveats, from the reg test running show to the server not having display tools. Do you have any improvement to suggest ? I tried to turn off the interactivity of show, but that had no impact. More generally, do you have any suggestion for nice reg tests with Vedo ?

Sorry for the very unprecise question, and happy friday

[marcomusy]

Hi @ldouteau yes - it's related to X11 forwarding have you already had a look at https://vedo.embl.es/autodocs/content/vedo/vedo.html#running-on-a-server ?

From inside a python script you can also call:

import vedo
vedo.settings.start_xvfb()
# vedo.settings.initi_colab()

# ...

Hope this helps!

[ldouteau]

Hi Marco, Thanks for the answer & the reference to the docs, i missed it. It solves the issue i have with segfault at test collection. However, i'm still getting seg faults with the minimal example provided (screenshot of a cube). I tested both the python & bash versions. I get the following log, that is not very helpful. I tested that on Ubuntu 22.04 LTS with Python 3.10 & a fresh new virtual environment

/usr/bin/Xvfb
starting xvfb (can take a minute) ... xvfb started.
Aborted (core dumped)

[marcomusy]

Hi, does this script also crash?

import vtk

cone = vtk.vtkConeSource()

mapper = vtk.vtkPolyDataMapper()
mapper.SetInputConnection(cone.GetOutputPort())

actor = vtk.vtkActor()
actor.SetMapper(mapper)

ren = vtk.vtkRenderer()
ren.AddActor(actor)

renWin = vtk.vtkRenderWindow()
renWin.SetOffScreenRendering(True)
renWin.SetSize(500, 500)
renWin.AddRenderer(ren)
renWin.Render()

[ldouteau]

Hi Marco. Sorry for the late answer. Yes this script crashes too, with the following logs:

2024-03-12 11:57:47.543 (   0.181s) [    7F9ABA305000]vtkXOpenGLRenderWindow.:456    ERR| vtkXOpenGLRenderWindow (0x56447770bfc0): bad X server connection. DISPLAY=
Aborted (core dumped)

[marcomusy]

Hi, I' afraid then it's a problem with the server not starting.. unrelated to vedo. I would manually check on your server the individual steps in https://github.com/marcomusy/vedo/blob/9e68ef2a7ae7f3632067590319875fcf9fce9ebc/vedo/settings.py#L728

[ldouteau]

Ok, that's kind of what i expected when i tested your suggestion earlier today. As it's not critical for me i guess i won't investigate more on this. Thanks for your efficient support, as usual !