notebook-on-kube

Create and manage your Jupyter notebooks on Kubernetes without JupyterHub :)

How and why?

You can check out this post.

notebook-on-kube provides the following features:

It provides the following features:

• Authn/authz based on Kubernetes'.
• Customize and create notebooks.
• Connect to notebooks.
• Pause/Resume notebooks.
• Get notebooks' events.
• See next steps.

How to use?

You need access to a Kubernetes cluster with an OIDC token.

• You can deploy notebook-on-kube on a Kubernetes cluster using Helm:

helm repo add notebook-on-kube https://machine424.github.io/notebook-on-kube
helm install nok notebook-on-kube/notebook-on-kube

• Or run the docker image directly from here.
• Or clone the repo and run:

pip install -e .
notebook-on-kube

You should land on:

Create, connect to and delete a notebook

Notes

• the Kubernetes OIDC token should contain an email claim and the local part of it should be unique as it's used to identify users. If you want to skip this validation and use any token to test, set the environment variable NOK_TEST_MODE=on (see values.yaml).
• notebook-on-kube is not meant to be exposed to the internet as some paths are not "protected" (/connect_notebook e.g.), use port-forwarding to interact with it, or use external authn (Oauth2 e.g.) or other, if you don't have a choice.
• By default, the notebooks have token-based authentication on, the token is set to the notebook's name.

Next steps

• Add JSON Schema for the Helm values (front + back (Python and/or Helm))
• Add a YAML Editor on /create_notebook (validation etc.)
• Replace /scale_notebook with a more generic /edit_noetbook (with YAML editor) that will helm upgrade with the new values.
• Enable culling support: Add Prometheus metric exporter + Kube HPA (prom adapter). Instead of JupyterHub idle culler
• Fastapi: More async?
• Maybe: Make this more generic to deploy other notebooks or even xxx-on-kube.