How to configure custom kernels

[stefanseefeld]

Hello,
I'm looking for information about setting up custom jupyter kernels with nebari. In particular, I want to configure my notebooks to run in custom docker containers, and thus need a way to pull the corresponding docker images into nebari so they can be started on demand. I was able to set this up in a TLJH instance, but am unsure of how to do the same with nebari. Any pointers would be highly appreciated. Thanks !

[Adam-D-Lewis]

Hi @stefanseefeld, are there any docs on the littlest Jupyter hub that explain what you're describing? I'm Nebari, building your environments in conda store is definitely the most supported, but I have seen some deployments with conda environments in the docker image. I'll try to provide some more info later.

Can you further describe why building your environment in conda store doesn't work for your use case? Maybe there's a feature we could add to conda store.

[stefanseefeld]

Yes, the above kernel config is exactly how I defined my own docker-based kernels. That worked fine, at least in TLJH where everything was local. I don't know whether or how this would work on a cluster.

In fact, the main reason I'm exploring Nebari is because of the JHub App loader. Could I use that outside of Nebari, e.g. within TLJH ?

[Adam-D-Lewis]

JHub apps is intended to be usable on other Jupyterhub deployments besides Nebari, but I'm not sure how much testing has gone into testing it on TLJH for example. @aktech could likely comment more on that.

[dharhas]

jhub-apps should work with any JupyterHub deployment with the caveat that the default theme has Nebari colors. If you try it out and get it working we'd love any documentation pr's if there is anything specific you encounter when using TLJH.

[stefanseefeld]

Thanks, that's great to hear. We are still having some issues setting things up (see nebari-dev/jhub-apps#409), but will hopefully sort this out quickly.

[aktech]

Hey @stefanseefeld 👋

I responded to that issue, earlier today. Jhub-apps should surely work on TLJH, I deployed it in the pas several times while developing. There might be some nitpicks required, which I might be able to help you with.