Skip to content

gesiscss/persistent_binderhub

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

This repository is no longer actively maintained!

However, the development of this functionality continues in cooperation with 2i2c.org. For more information, see our blog post:

👉 https://2i2c.org/blog/2022/gesis-2i2c-collaboration-update/

Persistent BinderHub

This is a Helm chart to install a persistent BinderHub. It simply configures and extends BinderHub chart to add persistent storage, it doesn't define any new component. Therefore before using this chart it is required that you read through BinderHub documentation, you know how to deploy a standard BinderHub and you are familiar with enabling authentication in BinderHub.

Prerequisites

First of all create a config.yaml file, everything explained here can go into that file and then it will be used for the installation. Before the installation there are configurations required to be done for User storage and Authentication.

User storage

To be able to offer a persistent storage to users, in your kubernetes cluster you need to have a storage class defined, which dynamically provisions persistent volumes. Please follow the user storage documentation of JupyterHub chart for more information.

Note that any configuration for JupyterHub chart goes under binderhub.jupyterhub in config.yaml that you created to install this chart. For example, if you want to specify the storage class, you have to add the following into your config.yaml:

binderhub:
  jupyterhub:
    singleuser:
      storage:
        dynamic:
          storageClass: <storageclass-name>

Authentication

This chart already includes some of the required changes for enabling authentication. But there are pieces that have to be manually configured. In your config.yaml:

  1. You have to set oauth_client_id:
binderhub:
  jupyterhub:
    hub:
      services:
        binder:
          # this is the default value
          oauth_client_id: "binder-oauth-client-test"
  1. You have to use config of your authenticator for binderhub.jupyterhub.auth. For more information you can check the authentication guide.
binderhub:
  jupyterhub:
    hub:
      config:
        JupyterHub:
          authenticator_class: dummy

Note that by default the authenticator is DummyAuthenticator and it is recommended to use it only for testing purposes.

Installing the chart

First of all you can find the list of charts here: https://gesiscss.github.io/persistent_binderhub/

The installation consists of 2 steps. As a first step we install the chart, then we will finalize the configuration of the Binder service and upgrade the chart to apply final changes in the config.

To install the chart with the release name pbhub into namespace pbhub-ns:

# add the persistent_binderhub helm chart repo
helm repo add persistent_binderhub https://gesiscss.github.io/persistent_binderhub/
# update charts
helm repo update

# you can change release name and namespace as you want
RELEASENAME=pbhub
NAMESPACE=pbhub-ns
kubectl create namespace $NAMESPACE
helm upgrade $RELEASENAME persistent_binderhub/persistent_binderhub \
             --version=0.2.0-n919 \
             --install --namespace=$NAMESPACE \
             --debug \
             -f config.yaml

After the first step, run kubectl get service proxy-public --namespace=$NAMESPACE and note down the IP address under EXTERNAL-IP, which is the IP of the JupyterHub. Then run kubectl get service binder --namespace=$NAMESPACE and again note down the IP address under EXTERNAL-IP, which is the IP of the Binder service.

With the IP addresses you just acquired update your config.yaml:

binderhub:
  jupyterhub:
    hub:
      services:
        binder:
          # where binder runs
          url: "http://<Binder_IP>"
          # when url is set, binder can be reached through JupyterHub
          oauth_redirect_uri: "http://<JupyterHub_IP>/services/binder/oauth_callback"

Finally upgrade the chart to apply this change:

helm upgrade $RELEASENAME persistent_binderhub/persistent_binderhub \
             --version=0.2.0-n919 \
             --install --namespace=$NAMESPACE \
             --debug \
             -f config.yaml

When the installation is done, the persistent BinderHub will be available at "http://<JupyterHub_IP>", and there (at JupyterHub home page) you will see a customized BinderHub UI for persistence, which is the place that users will interact with the system mostly. The standard BinderHub will be available at "http://<JupyterHub_IP>/services/binder" as a service of JupyterHub.

Known issues

  1. If you don't know the url of the JupyterHub (binderhub.config.BinderHub.hub_url) and it is not set during the first step of the installation, you will get an error similar to Error: render error in "persistent_binderhub/charts/binderhub/templates/deployment.yaml": template: persistent_binderhub/charts/binderhub/templates/deployment.yaml:98:74: executing "persistent_binderhub/charts/binderhub/templates/deployment.yaml" at <"/">: invalid value; expected string
    To fix it, you can use a dummy value for the hub_url, e.g. "http://127.0.0.1",
    and after the first step when you have the correct url of the hub, you can replace it.
    GitHub issue: #5
    Potential fix: jupyterhub/binderhub#1139

Uninstalling the chart

# to delete the Helm release
helm delete $RELEASENAME --purge
# to delete the Kubernetes namespace
kubectl delete namespace $NAMESPACE

Customization

As mentioned before, this chart extends the BinderHub chart in order to bring persistency in. To do that this chart also uses extraConfig from JupyterHub and BinderHub charts. While using persistent BinderHub chart, you should use another name for your extraConfigs, unless you want to overwrite defaults of this chart and you know what you are doing. Here is the list of extraConfigs used:

  • binderhub.extraConfig:

    • 20-launcher
    • 10-repo-providers
  • binderhub.jupyterhub.hub.extraConfig:

    • 20-template-variables
    • 10-project-api
    • 00-binder

For more information check values.yaml.

BinderHub customization

Anything you want to customize in BinderHub chart you can refer to the BinderHub documentation. The only thing you have to pay attention is that you put that configs under binderhub in your config.yaml. For example, if you want to use another version of repo2docker to build repos, add following into your config.yaml:

binderhub:
  config:
    BinderHub:
      build_image: quay.io/jupyterhub/repo2docker:2021.08.0-8.gf1d01b6

Note: repo2docker:2021.08.0-8.gf1d01b6 is the repo2docker version used in this chart.

Default project

A project is simply a binder-ready repository that you launch in a persistent BinderHub.

Default project is this repo itself by default (check .binder folder, there is intro_to_persistent_binderhub notebook).

Assuming that you have a binder-ready repo with the following information

  • repo url: https://github.com/user_name/repo_name
  • branch or tag or commit: ref

you can set it as default project by adding the following into your config.yaml:

binderhub:
  jupyterhub:
    custom:
      default_project:
        repo_url: "https://github.com/user_name/repo_name"
        ref: "ref"

Warning: Default project must be a binder-ready repo, e.g. https://github.com/binder-examples/requirements.

Projects limit per user

Number of projects concurrently stored per user. By default it is 5. For example, if you want to increase it to 10, add the following into your config.yaml:

binderhub:
  extraEnv:
    - name: PROJECTS_LIMIT_PER_USER
      value: "10" # change this value as you wish

Repo providers

Only the following repo providers are supported in this chart:

  • GitHubRepoProvider
  • GistRepoProvider
  • GitLabRepoProvider
  • GitRepoProvider

Other providers (ZenodoProvider, FigshareProvider, HydroshareProvider, DataverseProvider) are currently not supported. If you enable them, persistent BinderHub is not going to work as expected for these providers.

For example, if you want to only enable GitHubRepoProvider and GistRepoProvider, add the following into your config.yaml:

binderhub:
  extraConfig:
    10-repo-providers: |
      from binderhub.repoproviders import GitHubRepoProvider, GistRepoProvider
      c.BinderHub.repo_providers = {
          'gh': GitHubRepoProvider,
          'gist': GistRepoProvider,
      }

Spawner

This chart uses the PersistentBinderSpawner. If you want to customize it, you can subclass it in extraConfig. For example:

binderhub:
  jupyterhub:
    hub:
      extraConfig:
        00-binder: |
          from persistent_bhub_config import PersistentBinderSpawner
          MyPersistentBinderSpawner(PersistentBinderSpawner):
            ...
          c.JupyterHub.spawner_class = MyPersistentBinderSpawner

Local development

In local/minikube folder you can find instructions and configuration file to install this chart in minikube.

Migrating from JupyterHub chart

Be aware that this is not tested widely, but it should be safe to migrate from JupyterHub chart to this chart.

Here are the differences compared to fresh installation of this chart:

  • after migration, existing users will have no default project in the beginning
  • files of existing users won't be copied to anywhere, existing users can find them under /projects dir and they should manage them manually via terminal

Limitations

  1. Binder pod (binderhub.replicas) must be 1, otherwise there are authentication errors (jupyterhub/jupyterhub#2841).

Funded by the German Research Foundation (DFG). FKZ/project number: 324867496.