myst build failing

[MartinSchobben]

myst build gets stuck after a number of notebooks without discernible error. @ktyle @r-ford

[MartinSchobben]

I will try to revert to jupyter book as I will be presenting a poster about the cookbook at the living planet symposium next week.

[ktyle]

@MartinSchobben Hmmm by any chance might you be running into rate limiting 429 errors accessing Hugging Face datasets? I noticed this when doing the initial testing of the MyST transition. So far, adding datasets to environment.yml, following huggingface/datasets#7344, has resolved this error.

[MartinSchobben]

This can't be it, as this is changed in a new version. We don't use huggingface as an external data source any longer. Now that i reverted to jupyter book it also builds again without problem.

[brian-rose]

@MartinSchobben I tried running myst build --execute --html locally on a Mac laptop. The build did run to completion although several of the notebooks raised errors like

RuntimeError: error during blosc decompression: 0

I wonder if the myst builds on GitHub Actions were timing out?

I was not able to successfully build with jupyter-book build . because of the same notebooks errors, and the fact that the jupyter-book configuration does not permit the build to pass if there are any execution errors in the notebooks.

Also it looks like this repo had been failing the nightly builds for the last three weeks, prior to the myst conversion. So maybe more than one issue going on here.

Reverting to JB1 in the short term makes sense if that gets you up and running again, but it would be great to figure out what's going wrong.

[MartinSchobben]

I really can't discern execution errors on my Linux machine or the GitHub actions. I only get some warnings that seem to originate from holoviews. Maybe the only thing I can think of after many attempts is the stdout print related to snaphu in notebook 9 of microwave remote sensing (starting from line 129) that should be dealt with in a better way.

[ktyle]

I've spent some time the last couple days looking into this. There look to be the following issues:

1. I'm continuing to get 429 Rate Limiting errors related to too many requests to Hugging Face-hosted datasets. These don't occur with every notebook, but notebooks 3-6 in notebooks/courses/microwave-remote-sensing all still point to Hugging Face as the repository (e.g. in exercise 3, https://huggingface.co/datasets/martinschobben/microwave-remote-sensing/resolve/main/microwave-remote-sensing.yml). One alternative that we can use is to host the various zarr files on our Project Pythia cloud storage on Jetstream2. I've successfully done this for the Exercise 5 notebook.
2. The blosc error that we get in Exercise 5 seems specific to when the dataset is accessed via Intake. Accessing it via our cloud storage using s3fs and fssepc eliminates the error.
3. Indeed, the myst build --execute hangs on the Exercise 9 notebook, and it is definitely tied to snaphu . This only happens when the notebook is executed via myst build --execute ... interactively running the notebook in the environment built from environment.yml runs as expected. I will plan to open an issue on mystmd regarding this.

I am currently working on a fork to address the first two items. See https://github.com/ktyle/eo-datascience-cookbook/tree/jun25 .

[ktyle]

Additionally @brian-rose I did a myst build on my Mac but it failed in a similar manner to what Martin and I have seen on Linux systems (which includes GitHub Actions as well).

[MartinSchobben]

Thanks for all the help! Do mind, however, that the last release does not use hugging face any longer. The rate limiting error has been abolished. The nightly build now also runs without error since three nights.