Readthedocs.org: Build timeouts using nbconvert

[SaOgaz]

Details

Hi,

I've been using nbconvert for my builds for a long time, but recently the builds starting timing out. It looks like the sphinx-build commands (that are calling nbconvert) are taking about 10 times longer to run, which then causes my build to time out. We're talking ~30 seconds to ~200-300 seconds. I have only added one or two new notebooks, and even tried testing a previous commit of the notebooks that had completed successfully before. Any idea what's happening?

I've checked my version of nbconvert and ipython on the runs and it doesn't seem like they've changed. I've been updating to Python 3, but it looks like there are still runs failing under Python 2.

• Project URL: https://readthedocs.org/projects/stak-notebooks/
• Build URL (timing out): https://readthedocs.org/projects/stak-notebooks/builds/6396637/ *had wrong build here, updated
• Build URL (not timing out): https://readthedocs.org/projects/stak-notebooks/builds/6024721/
• Read the Docs username (if applicable): SaraOgaz

[willingc]

Hi @SaraOgaz,

I work on Jupyter's docs and JupyterHub. One thing that you may try is to start by updating the readthedocs.yml file to be structured more like:

name: jupyterhub
type: sphinx
conda:
    file: docs/environment.yml
python:
  version: 3

You may need to wipe the build in the settings on RTD, but I would try first without doing that. Ping me with the results.

[SaOgaz]

@willingc, thanks for the recommendation. Just tried it, and it didn't seem to have much effect on the build times, still around ~200 seconds for the sphinx-build portions.

[RichardLitt]

Thanks for following up, @SaraOgaz.

Mildly related note: @agjohnson This is one of many sphinx-related issues. Should we add a sphinx tag, or prepare some better docs for diagnosing issues like this?

[agjohnson]

So, I'm woefully behind on notebook usage -- I'm not yet able to tell what is unexpected behavior from notebooks. From my perspective, this looks like normal usage of nbcovert, but perhaps someone with more knowledge of Jupyter/notebooks/nbconvert can help provide us with some guidance (also, thanks @willingc for chiming in!)

If there is nothing that can be done about the increase in step time, which does seem troubling, we can always just increase the build timeout for the project.

@RichardLitt I assume that projects use Sphinx unless specified as a mkdocs project, which I'll generally give a Mkdocs label to help mkdocs developers find them. I have been organizing issues into groups using milestones as a way to apply development priority, though this usually excludes Support tagged issues, at least until they are a confirmed bug.

[willingc]

Hi @SaraOgaz and others, I forked Sara's repo and was able with a few updates to the config to get the docs to build for master. The docs say that they have timed out at 900s but they did build in 2 of 3 cases. Looking at the conf.py, I think that there is likely a way to further optimize the nbconvert usage in the build.

Test doc build on RTD: http://test-stak.readthedocs.io/en/latest/index.html

Repo fork with a few changes added to master: https://github.com/spacetelescope/stak-notebooks/compare/master...willingc:master

[SaOgaz]

Thanks for the additional help @willingc. I am still very interested in finding out why the build time for this suddenly became 10x longer. We may add more notebooks in the future and would find it frustrating to have our builds constantly on the verge of timing out.

[SaOgaz]

Hold the phone, I just restarted a build, and it's back to taking ~30 seconds for each nbconvert section... Did something change back on your guys's end? I guess I can close this issue now since it appears to be fixed?

[stsewd]

Your builds from latest are passing, I guess we can close this, feel free to reopen if you see something failing again. Also, you can check https://docs.readthedocs.io/en/latest/guides/build-using-too-many-resources.html