JupyterHub Enabled LMS

The purpose of this document is to create a bare-minimum learning management system that can offer interactive computing labs with the use of jupyterhub. From jupyterhub's documentation, jupyterhub can be used to serve sandboxed computing environments to students by the use of the Learning Tools Interoperability (LTI) standard. The LTI was created by MS Global Learning Consortium for connecting external service tools to learning management systems like openEdx, canvas, blackboard, moodle and etc. Officially, jupyterhub has only tested their LTI implementation on openeEdx and canvas. Since Canvas is a lighter system than the openEdx platform, it will be used in order to test the integration of Jupyterhub with an LMS.

There is a commercial software as a service (SAAS) that has an implementation of this, called IllumiDesk with their pricing here. For a video of how Illumidesk works, and how this testing environment should work similarly, check out the video here.

Setup: JupyterHub (Tested in DigitalOcean Droplet - 4GB 2VCPUs)

1. Create virtualenv env for running sandboxed jupyterhub:

    virtualenv -ppython3 env

2. Enter env environment:

   source env/bin/activate

3. Install jupyterhub for multiuser-session jupyter notebooks:

    pip install jupyterhub

4. Install jupyterhub-ltiauthenticator for multiuser-session jupyter notebooks:

    pip install jupyterhub-ltiauthenticator

5. Install nbgrader for auto-grading jupyter notebooks:

    pip install git+git://github.com/samhinshaw/nbgrader.git

6. Install nbgrader for pulling jupyter notebooks from git repositories:

    pip install nbgitpuller

7. Generate LTI_CLIENT_KEY and LTI_CLIENT_SECRET then save to .env:

    LTI_CLIENT_KEY=(`openssl rand -hex 32`)
    LTI_CLIENT_SECRET=(`openssl rand -hex 32`)
    echo LTI_CLIENT_KEY=$LTI_CLIENT_KEY > .env
    echo LTI_CLIENT_SECRET=$LTI_CLIENT_SECRET >> .env
    echo export LTI_CLIENT_KEY >> .env
    echo export LTI_CLIENT_SECRET >> .env
    source .env

8. Generate config file for jupyterhub:

    jupyterhub --generate-config

9. Set ltiauthenticator as jupyterhub's authenticator class in jupyter_config.py generated above:

   #c.JupyterHub.authenticator_class = 'jupyterhub.auth.PAMAuthenticator'
   import os
   c.JupyterHub.authenticator_class = 'ltiauthenticator.LTIAuthenticator'
   c.LTIAuthenticator.consumers = {
       os.environ['LTI_CLIENT_KEY']: os.environ['LTI_CLIENT_SECRET']
   }

10. install configurable-http-proxy as it is needed by jupyterhub:

    sudo npm install -g configurable-http-proxy

11. Add machine's hostname to /etc/hosts :

    sudo echo 127.0.0.1 $(hostname) >> /etc/hosts

12. Start jupyterhub with specified jupyterhub_config.py:

    jupyterhub -f /path/to/jupyterhub_config.py

Setup: Canvas-LMS (Tested in DigitalOcean Droplet - 8GB 4VCPUs)

1. Clone canvas-lms repository:

    git clone https://github.com/instructure/canvas-lms

2. Enter canvas-lms directory:

    cd canvas-lms

3. Perform quick install script:

    ./script/docker_dev_setup.sh

Connecting Canvas-LMS with Jupyterhub (WIP: Canvas LMS Server still configuring)

References:

• https://github.com/jupyterhub/ltiauthenticator

• https://www.slideshare.net/SamuelHinshaw1/integrating-canvas-jupyterhub

• https://github.com/UBC-DSCI/rudaux

• https://github.com/instructure/canvas-lms