Autolab OneClick Installation

This is a customized installer for CMU DL "passing together" course (https://dlcmu.datagym.ru/). The idea is to make it very simple for users to join. Users can only sign in using Github OAuth,and then automatically subscribed to the only course: CMUDL. There is no need in sign-ups, confimation email, password changes etc.

Differences in installer

removed sudo ops from install.sh, installation can be done as a normal user with no special rights except for docker.
presume that docker is installed and current user is in docker group.
docker compose is modernized to version 3
database volume is created as docker volume
courses volume is created as docker volume.
do not copy Autopopulated course

Patches to Autolab

uglifier version upgraded to use harmony: true
Autopopulate with "CMUDL" course
create only 1 admin user
github-oauth added to Gemfile
github-oauth is the only auth method left. When user is logged in with github oauth, he/she is automatically subsribed to "CMUDL"
Attention: github app id and secret must be added to the live container!

Known issues

User can not edit his/her personal information: vising profile leads to 500 error. Most likely related to using some email methods that are disabled elsewhere.
User's name and surname are set to his/her Github login and id respectively. This is to avoid potential database problems if user's Github name is not in ascii.

Customization

Autolab

edit cover/autolab.rack to set your admin_email
edit cover/autolab.rack to set custom pre-created course name
edit cover/create_ods_user.rb to set a course name to automatically subscribe a newly signed in user.
edit cover/new.html.erb to customize login page.

Once your images are built and the containers are up, login to the local_web_1 container:

docker exec -it local_web_1 bash

Find devise.rb and put there your github app id and secret. If you can suggest a better way, please do. I could not make .env working here.

Assignments

Copy existing assignments (exported from other instances of Autolab) into the docker volume:

docker cp homework0.tar local_web_1:/home/app/webbapp/courses/CMUDL/

docker exec -it local_web_1 bash

Change right on newly copied files, if different:

chmod -R app:app /home/app/webbapp/courses/CMUDL/homework0

Go to Course Admin interface and import this assignment.

Grading images

Go to live tango container:

docker exec -it local_tango_1 bash

and then pull your image, i.e.:

docker pull datamove/cmudlimage:torch100

Then you need to retag the image, since Tango doesn't like / in the image:

docker tag datamove/cmudlimage:torch100 cmudlimage_torch100

cmudlimage_torch100 is what you use in Autograder section of your assignment settings.

Debugging

Note, Autolab is cloned from its repo during the installation, so as it evolves, patches applied by the installer may become incompatible.

You do not have to clone Autolab every time, i.e. instead of running install.sh with your changes, you can just use docker:

docker-compose up -d

and optionally web rake commands to reset the db.

Debugging users

Use rails console inside the container:

User.all
User.first
User.where(email: "ermakov.pd@gmail.com").first
User.new(...).save
User.where(...).first.delete

To create a new user use code sample from autolab.rake.

Acknowledgements

Petr Ermakov (datagym.ru, dlcourse.ru) for driving this project.
Nick Bienko (@bikolya at ODS Slack) for adapting Autolab to this work mode.

Original README from Autolab team

Autolab-OneClick is the fastest way to install a full autograding environment and course management system!

You can setup Autolab in a local or server environment

Contact autolab-dev@andrew.cmu.edu for questions. Please fill out Github issues if you encounter any problems.

datamove/autolab-oneclick