Advanced Topics

This page contains a grab bag of various useful topics that don’t have an easy home elsewhere:

  • Ingress

  • Arbitrary extra code and configuration in

Most people setting up JupyterHubs on popular public clouds should not have to use any of this information, but these topics are essential for more complex installations.


The Helm chart can be configured to create a Kubernetes Ingress resource to expose JupyterHub using an Ingress controller.


Not all k8s clusters are setup with an Ingress controller by default. If you need to install one manually, we recommend using ingress-nginx.

The minimal example to expose JupyterHub using an Ingress resource is the following:

  enabled: true

Typically you should declare that only traffic to a certain domain name should be accepted though to avoid conflicts with other Ingress resources.

  enabled: true

Ingress and Automatic HTTPS with kube-lego & Let’s Encrypt

When using an ingress object, the default automatic HTTPS support does not work. To have automatic fetch and renewal of HTTPS certificates, you must set it up yourself.

Here’s a method that uses kube-lego to automatically fetch and renew HTTPS certificates from Let’s Encrypt. This approach with kube-lego and Let’s Encrypt currently only works with two ingress controllers: the community-maintained kubernetes/ingress-nginx and google cloud’s ingress controller.

  1. Make sure that DNS is properly set up (configuration depends on the ingress controller you are using and how your cluster was set up). Accessing <hostname> from a browser should route traffic to the hub.

  2. Install & configure kube-lego using the kube-lego helm-chart. Remember to change config.LEGO_EMAIL and config.LEGO_URL at the least.

  3. Add an annotation + TLS config to the ingress so kube-lego knows to get certificates for it:

      annotations: "true"
        - hosts:
            - <hostname>
          secretName: kubelego-tls-jupyterhub

This should provision a certificate, and keep renewing it whenever it gets close to expiry!

Arbitrary extra code and configuration in

Sometimes the various options exposed via the helm-chart’s values.yaml is not enough, and you need to insert arbitrary extra code / config into This is a valuable escape hatch for both prototyping new features that are not yet present in the helm-chart, and also for installation-specific customization that is not suited for upstreaming.

There are four properties you can set in your config.yaml to do this.


The value specified for hub.extraConfig is evaluated as python code at the end of You can do anything here since it is arbitrary Python Code. Some examples of things you can do:

  1. Override various methods in the Spawner / Authenticator by subclassing them. For example, you can use this to pass authentication credentials for the user (such as GitHub OAuth tokens) to the environment. See the JupyterHub docs for an example.

  2. Specify traitlets that take callables as values, allowing dynamic per-user configuration.

  3. Set traitlets for JupyterHub / Spawner / Authenticator that are not currently supported in the helm chart

Unfortunately, you have to write your python in your YAML file. There’s no way to include a file in config.yaml.

You can specify hub.extraConfig as a raw string (remember to use the | for multi-line YAML strings):

  extraConfig: |
    import time
            "CURRENT_TIME": str(time.time())

You can also specify hub.extraConfig as a dictionary, if you want to logically split your customizations. The code will be evaluated in alphabetical sorted order of the key.

    00-first-config: |
      import time
              "CURRENT_TIME": str(time.time())
    10-second-config: |
      # some other code

custom configuration

The contents of values.yaml is passed through to the Hub image. You can access these values via the z2jh.get_config function, for further customization of the hub pod. Version 0.8 of the chart adds a top-level custom field for passing through additional configuration that you may use. It can be arbitrary YAML. You can use this to separate your code (which goes in hub.extraConfig) from your config (which should go in custom).

For example, if you use the following snippet in your config.yaml file:

  myString: Hello!
    - Item1
    - Item2
    key: value
  myLongString: |

In your hub.extraConfig,

  1. z2jh.get_config('custom.myString') will return a string "Hello!"

  2. z2jh.get_config('custom.myList') will return a list ["Item1", "Item2"]

  3. z2jh.get_config('custom.myDict') will return a dict {"key": "value"}

  4. z2jh.get_config('custom.myLongString') will return a string "Line1\nLine2"

  5. z2jh.get_config('custom.nonExistent') will return None (since you didn’t specify any value for nonExistent)

  6. z2jh.get_config('custom.myDefault', True) will return True, since that is specified as the second parameter (default)

You need to have a import z2jh at the top of your extraConfig for z2jh.get_config(...) to work.


This property takes a dictionary that is set as environment variables in the hub container. You can use this to either pass in additional config to code in your hub.extraConfig or set some hub parameters that are not settable by other means.


A list of extra containers that are bundled alongside the hub container in the same pod. This is a common pattern in kubernetes that as a long list of cool use cases. Some example use cases are:

  1. Database Proxies, which are sometimes required for the hub to talk to its configured database (in Google Cloud) for example

  2. Servers / other daemons that are used by code in your hub.customConfig

The items in this list must be valid kubernetes container specifications.

Specifying suitable hub storage

By default, the hub’s sqlite-pvc setting will dynamically create a disk to store the sqlite database. It is possible to configure other storage classes under hub.db.pvc, but make sure to choose one that the hub can write quickly and safely to. Slow or higher latency storage classes can cause hub operations to lag which may ultimately lead to HTTP errors in user environments.

Referencing resources from a parent Helm chart’s templates

The k8s resources in this Helm chart should not be referenced by hardcoded name as they can be adjusted via fullnameOverride or nameOverride. So to reliably reference them, you can make use of defined Helm templates.

Below is an example of how you could define a RoleBinding binding some Role you have defined as well and the hub k8s ServiceAccount that may be named something slightly different if fullnameOverride or nameOverride has been used.

kind: RoleBinding
  name: my-role-binding
  - kind: ServiceAccount
    name: {{ include "jupyterhub.hub.fullname" . }}
    namespace: {{ .Release.Namespace }}
  kind: Role
  name: my-role