This directory contains a Helm chart for a default configuration to use MLHub.

It is inspired and partially based on the great Zero to JupyterHub K8s project. However, we made some modifications for two reasons; first, to get it work with ml-hub and and ml-workspace. Second, to simplify the setup. Zero to Jupyterhub K8s is very sophisticated but adds an extra layer of complexity by, for example, not directly enabling you to configure KubeSpawner as you are used to by Jupyterhub via it's c.Spawner config but via mixing it with infrastructure configuration in the same values.yaml file. They have a great documentation about pre-pulling images, auto-scaling the cluster etc. So if this simple version is not enough, we recommend to check out their project and documentation.

Most prominent changes to the "original" project:

• change of the command fields in hub and proxy yamls
• modifying ports to make tunnelling of ssh possible
• changes of default values, e.g. the used images
• changes of paths, e.g. the ssl secret mount path
• separated the deployment configuration and the configuration for JupyterHub & the Spawner
• removed a few Kubernetes resources such as the image puller etc.

We do not push the helm chart to a repository for now, so feel free to download it from the mlhub releases page or to create the package yourself via the helm package command.

You can then deploy the chart via helm upgrade --install mlhub packaged-chart.tgz --namespace $namespace --values config.yaml --set-file userConfig=./jupyterhub_user_config.py. The config.yaml can be used to overrride default deployment values, the userConfig can be used to configure JupyterHub and the Spawner.

For more details, check out the main readme.

You find the default values for the deployment in the mlhub/values.yaml file. Some values should be self-explanatory; if not we encourage you to have a look at the chart files or at the Zero to JupyterHub K8s documentation, on which our setup is based.