test_plotting failures on M2 Mac due to memory issues

Describe the bug, what’s wrong, and what you expected.

I’m seeing ~93 test failures in tests/plotting/test_plotting.py on my M2 Mac.

I think all of them are raising the same error: RuntimeError: Unable to extract image from Plotter as it has already been closed. from

https://github.com/pyvista/pyvista/blob/adfb940d9e4d9d7d01530c7453c91e82921abea2/pyvista/utilities/regression.py#L119-L122

Here is one of the test failures:

___________________________________ test_plot_points_gaussian_scalars ___________________________________
[gw3] darwin -- Python 3.9.13 /opt/miniconda3/envs/pyvista-dev/bin/python

sphere = PolyData (0x29ba6cb80)
  N Cells:	840
  N Points:	422
  N Strips:	0
  X Bounds:	-1.260e+02, 1.250e+02
  Y Bounds:	-1.270e+02, 1.260e+02
  Z Bounds:	-1.270e+02, 1.270e+02
  N Arrays:	2


    @skip_windows
    def test_plot_points_gaussian_scalars(sphere):
>       sphere.plot(
            scalars=sphere.points[:, 2],
            style='points_gaussian',
            render_points_as_spheres=False,
            point_size=20,
            opacity=0.5,
            show_scalar_bar=False,
            before_close_callback=verify_cache_image,
        )

tests/plotting/test_plotting.py:3092:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
pyvista/plotting/helpers.py:302: in plot
    return pl.show(
pyvista/plotting/plotting.py:6201: in show
    self.close()
pyvista/plotting/plotting.py:4027: in close
    self._before_close_callback(self)
tests/plotting/test_plotting.py:250: in verify_cache_image
    error = pyvista.compare_images(image_filename, plotter)
pyvista/utilities/regression.py:131: in compare_images
    im2 = remove_alpha(to_img(im2))
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

img = <pyvista.plotting.plotting.Plotter object at 0x29e0c3190>

    def to_img(img):
        if isinstance(img, UniformGrid):  # pragma: no cover
            return img
        elif isinstance(img, _vtk.vtkImageData):
            return wrap(img)
        elif isinstance(img, str):
            return read(img)
        elif isinstance(img, np.ndarray):
            return wrap_image_array(img)
        elif isinstance(img, Plotter):
            if img._first_time:  # must be rendered first else segfault
                img._on_first_render_request()
                img.render()
            if not hasattr(img, 'ren_win'):
>               raise RuntimeError(
                    'Unable to extract image from Plotter as it has already been closed.'
                )
E               RuntimeError: Unable to extract image from Plotter as it has already been closed.

pyvista/utilities/regression.py:120: RuntimeError
----------------------------------------- Captured stderr call ------------------------------------------
UNSUPPORTED (log once): POSSIBLE ISSUE: unit 0 GLD_TEXTURE_INDEX_2D is unloadable and bound to sampler type (Float) - using zero texture because texture unloadable
2022-10-14 12:17:18.079 python[82376:1047634] GLDRendererMetal command buffer completion error: Error Domain=MTLCommandBufferErrorDomain Code=8 "Insufficient Memory (00000008:kIOGPUCommandBufferCallbackErrorOutOfMemory)" UserInfo={NSLocalizedDescription=Insufficient Memory (00000008:kIOGPUCommandBufferCallbackErrorOutOfMemory)}

If I run with pytest -v -n 6 . I see ~93 failures. If I run with pytest -v ., I see ~82 failures. If I just run pytest -v tests/plotting/test_plotting.py, I see ~21 failures.

This pattern makes me think there is a memory management issue or memory leak an eventually the plotting tests just start failing

Steps to reproduce the bug.

pytest -v .

On an M2 mac

System Information

--------------------------------------------------------------------------------
  Date: Fri Oct 14 12:18:07 2022 MDT

                OS : Darwin
            CPU(s) : 8
           Machine : arm64
      Architecture : 64bit
               RAM : 24.0 GiB
       Environment : Python
       File system : apfs
        GPU Vendor : Apple
      GPU Renderer : Apple M2
       GPU Version : 4.1 Metal - 76.3

  Python 3.9.13 (main, Aug 25 2022, 18:24:45)  [Clang 12.0.0 ]

           pyvista : 0.37.dev0
               vtk : 9.2.2
             numpy : 1.23.1
           imageio : 2.22.1
           appdirs : 1.4.4
            scooby : 0.5.12
        matplotlib : 3.6.1
           IPython : 8.5.0
          colorcet : 3.0.1
           cmocean : 2.0
        ipyvtklink : 0.2.3
             scipy : 1.9.2
              tqdm : 4.64.1
            meshio : 5.3.4
        jupyterlab : 3.4.8
         pythreejs : Version unknown
--------------------------------------------------------------------------------

This is an Apple M2 chip with 8‑core CPU, 10‑core GPU, 24GB unified memory

Screenshots

No response

[adeak]

The test suite ran without any issues with a bit more than 9 GB of free RAM. The memory monitor graph barely increased during the run. The largest memory use estimate from pytest is 210 MB, and even if I add up all the numbers above 1 MB (which gives a nonsensical upper bound) I get 5.2 GB. This is just to confirm that it is extremely unlikely that you are actually running out of RAM.

[adeak]

My laptop (running linux) has 16 gigs total and I can usually run the test suite fine. I’ll check later tonight. I’d expect this to be some memory allocation bug on the weird M2 rather than an actual pyvista bug.