BinderHub test mode

This document explains how to contribute to BinderHub from a bare-metal server. If you are a Neurolibre dev, you don’t need to follow First time setup section, just jump directly to Code integration section.

First time setup

Create an instance with openstack using bionic image, don’t forget to assign a floating IP. After, you can ssh to this instance.

Note

You can find detailed instructions on how to create an openstack instance in Bare-metal to local Docker registry and volumes.

All the following should be run as root :

sudo su - root

Now install docker.

Install npm and other dependencies :

apt-get install libssl-dev libcurl4-openssl-dev python-dev python3 python3-pip curl socat
curl -sL https://deb.nodesource.com/setup_13.x | sudo -E bash -
apt-get install -y nodejs

Install minikube for a bare-metal server.

Install kubectl.

Warning

Don’t forget to let kubectl run commands as your own user: sudo chown -R $USER $HOME/.kube $HOME/.minikube.

Install binderhub repo:

git clone https://github.com/jupyterhub/binderhub
cd binderhub

You can now follow the contribution guide from step 3.

Note

Since you are in a bare-metal environment like, you don’t need to use eval $(minikube docker-env)

You can now connect and verify the binderhub installation by accessing http://localhost:7777/.

Code integration

To make changes to the K8s integration of BinderHub, such as injecting repo2data specific labels to a build pod, we need to bring up a BinderHub for development.

The following guidelines are inhereted from the original BinderHub docs. This documentation assumes that the development is to be done in a remote node via ssh access.

  1. ssh into the previously configured node

Note

Ask any infrastructure admin for the current binderhub debug instance, if authorized.

  1. Launch shell as the root user:

    sudo su - root

  2. Make sure that the following apt packages are installed

    • npm

    • git

    • curl

    • python3

    • python3-pip

    • socat

  3. Ensure that the minikube is installed, if not follow these instructions.

  4. Clone the BinderHub repo and cd into it:

    git clone https://github.com/jupyterhub/binderhub
cd binderhub

  5. Start minikube:

    minikube start

  6. Install helm to the minikube cluster:

    curl https://raw.githubusercontent.com/kubernetes/helm/master/scripts/get | bash

  7. Initialize helm in the minikube cluster:

    helm init

  8. Add JupyterHub to the helm charts:

    helm repo add jupyterhub https://jupyterhub.github.io/helm-chart/
helm repo update

    The process is successful if you see the Hub is up message.

  9. Install BinderHub and its development requirements:

python3 -m pip install -e . -r dev-requirements.txt

  1. Install JupyterHub in the minikube with helm:

./testing/minikube/install-hub

  1. Make minikube use the host Docker daemon :

eval $(minikube docker-env)

Expect 'none' driver does not support 'minikube docker-env' command message. This is intended behavior.

  1. Run helm list command to see if the JupytherHub is listed. It should look like:

binder-test-hub 1 DEPLOYED jupyterhub-0.9.0-beta.4 1.1.0

Now, you are ready to start BinderHub with a config file. As done in the reference doc, start the binderhub with the config in the testing directory:

python3 -m binderhub -f testing/minikube/binderhub_config.py

Note

You are starting BinderHub with module name. This is possible thanks to the step-10 above. In that step, -e argument is passed to pip to point the local ../binderhub directory as the project path via . value. This is why the changes you made in the /binderhub directory will take effect.

There are some details worth knowing in the testing/minikube/binderhub_config.py file, such as:

c.BinderHub.hub_url = 'http://{}:30123'.format(minikube_ip)

This means that upon a successful build, the BinderHub session will be exposed to your_minikube_IP:30123. To find out your minikube IP, you can Simply run minikube ip command.

The port number 30123 is described in jupyterhub-helm-config.yaml.

If everything went right, then you should be seeing the following message:

[I 200318 23:53:33 app:692] BinderHub starting on port 8585

Just leave this terminal window as is. Open a new terminal and do ssh forward the port 8585 to the port 4000 of your own computer by:

ssh -L 4000:127.0.0.1:8585 ubuntu@<floating-ip-to-the-node>

Open your web browser and visit http://localhost:4000/. BinderHub should be running here.

When you start a build project by pointing BinderHub to a GitHub repo, a pod will be associated with the process. You can see this pod by opening a third terminal in your computer. Do not login shell as root in the second terminal, which is used for ssh 8585-->4000 port forwarding.

In the 3rd terminal, do the steps 1 and 2 (above), then:

kubectl get pods -n binder-test

If you injected some metadata, label etc. to a pod, you can see by:

kubectl get describe -n binder-test <pod_name>

It is expected that you’ll receive a 404 response after a successful Binder build. This is because the user is automatically redirected from 8585 to the instance served at your_minikube_IP:30123.

If you would like to interact with a built environment, you need to forward your_minikube_IP:30123 to another port in your laptop using another terminal.

Finally, Docker images created by Binder builds in the minikube host can be seen simply by docker images. If you’d like to switch docker environment back to the default user, run eval $(docker-env -u).

Terminate the BinderHub running on port 8585 by simply ctrl+c.

To delete the JupyterHub running on minikube, first helm list, then helm delete --purge <whatever_the_name_is>.

Further tips such as using a local repo2docker installation instead of the one comes in a container, enabling debug logging (really useful) and more, please visit the original resource.

To see how BinderHub automates building and publishing images for helm charts, please visit the chartpress.