Readthedocs for PyLops library - memory issue

[mrava87]

Details

Hello,
I have been trying to build my documentation from this project https://github.com/Statoil/pylops in readthedocs.

I get an error about memory comsumption (build report below) but I do not seem to find more details onto at which point of the build this happens. I am aware of the fact that you limit the build to 1GB memory, if you could help me understand where the build fails I could try to simplify the example to be able to avoid reaching the limit :)

Basically the documentation uses sphinx-gallery to create various examples for each class that I implement and some of them require quite some computation to be meaningful, but we are still talking about less than 10 minutes on a normal OSX laptop.

I have sent an email to your support account and Manuel replied suggesting to open an issue.

Looking forward to hearing from you,
MR

• Read the Docs project URL: https://readthedocs.org/projects/pylops
• Build URL (if applicable): https://readthedocs.org/projects/pylops/builds/8181469/
• Read the Docs username (if applicable): mrava87

Expected Result

Building documentation

A description of what you wanted to happen
I would like to get my documentation built succesfully

Actual Result

Command killed due to excessive memory consumption

[mrava87]

Hi humitos,
looking forward to hearing from you, hope you can help me troubleshoot this problem :)

[stsewd]

I guess you could reduce the number of images or generate the images outside rtd (I think someone did this in the past)

[mrava87]

Thanks for suggestions. Could you explain how that would help? Would it help if I have a plt.close('all')
at the start/end of each Sphinx-gallery example? I see something like this is done here https://matplotlibguide.readthedocs.io/en/latest/Matplotlib/basic.html#id1

And if I generate them outside of rtd how can I then keep the Sphinx-examples in the documentation?

[stsewd]

#3468

[mrava87]

Mmh this sounds a poor man job trying to replicate what Sphinx-gallery (https://sphinx-gallery.readthedocs.io/en/latest/) does for you. This is used by many libraries like Matplotlib and sciiti-learn and they have tons of images. How do they get around?

[mrava87]

Quick update. Things work now:

• added plt.close(‘all’), this didn’t solve the problem but the build run for longer before giving up
• remove most heavy computational examples and build is successful.

I guess the problem is that memory really goes above limit, but I wonder if it would be possible to explain how we can access the stdout and stderr of the Sphinx-build command, otherwise it’s a bit difficult to troubleshoot :)

[mrava87]

Also I think the solution in the issue you linked should not be encouraged as it makes things way more cumbersome than they can be and just does not take advantage of a fantastic library which sphinx-gallery is :)

[stsewd]

Looks like we replace the output when it's killed

https://github.com/rtfd/readthedocs.org/blob/85d9a1fb236e2c2149358b1cf462159e523e80bb/readthedocs/doc_builder/environments.py#L274-L276

I guess we could keep the original output.

[mrava87]

Up to you to decide of course but I personally think that would help a lot :)

Thanks for help!

[stsewd]

I think it's a good idea, I opened #4957 to track that, closing this issue as the problem was solved.