CodeOcean

Introduction

CodeOcean is an educational, web-based execution and development environment for practical programming exercises designed for the use in Massive Open Online Courses (MOOCs).

The programming courses offered on openHPI include practical programming exercises that are provided on a web-based code execution platform called CodeOcean. The platform has many advantages for learners and teachers alike:

1. CodeOcean offers a predefined view on the exercises with scaffolded source code including syntax highlighting and allows editing these files directly within the browser.
2. The platform allows executing the code on the server and streams the output back to the learner’s web browser. Together with the customized view, learners are supported to get up with the programming in no time instead of taking care of the development environment. Learners do not need to download and install any compiler or runtime. Thus, MOOC instructors minimize technical help requests about correct machine setup and can focus forum posts on the content provided in the course.
3. CodeOcean includes unit tests to provide feedback for learners and score their code. A unit test is defined as a program that either runs the learner’s code in a pre-defined way and compares the provided result with an expectation or the unit test parses the student’s source code and matches it against an exercise-defined string. While the code of the unit tests are hidden, learners can run the unit tests at any time and get instant feedback whether they passed or failed. If the unit tests fail the result is shown together with an error message defined by the MOOC instructors. On the one hand, this feedback helps people to help themselves and provides learners with a hint of their mistake. On the other hand, the automated scoring using unit tests is required to indicate progress for the learners. In the context of a MOOC with thousands of active learners, a manual review by the instructors is not feasible and peer-review of source code has not been implemented in CodeOcean so far.
4. In CodeOcean, learners can ask questions about their program directly within the platform and in context of their current program. Usually, MOOC platforms provide a forum to discuss questions. While this concept also works great for source code in general outside of a MOOC (cf. StackOverflow), it is an additional barrier for novices to summarize their problem externally. To understand the problem, contextual information is generally of help for others to provide the current solution. When using a dedicated forum, learners are required to provide as much information as necessary to reproduce the issue which beginners might find difficult to identify. As a result, they might copy too few or too much information. In addition, early iterations of the Java courses showed that learners did not format their source code appropriate in forum posts (but as plain text), making it difficult to read. With Request for Comments, CodeOcean provides a built-in feature to ask a question in the context of an exercise, thus lowering the barriers to get help. CodeOcean presents the learner’s source code and error message together with the question to fellow students and allows them to add a comment specifically to one line of code. Hence, the previously described issue is solved with a dedicated forum.

CodeOcean is mainly used in the context of MOOCs (such as those offered on openHPI and mooc.house) and has been used by more than 60,000 users as of June 2020. CodeOcean is a stand-alone tool implementing the Learning Tools Interoperability (LTI) standard to be used in various learning scenarios. By offering an LTI interface, it is accessible from MOOC providers as well as other providers, such as the HPI Schul-Cloud. CodeOcean itself cannot be used directly by learners or other users than the MOOCs instructors or administrators.

Development Setup

Please refer to the Local Setup Guide for more details.

Mandatory Steps

• install the Docker client
• run bundle install
• create config/action_mailer.yml
• create config/database.yml
• create config/secrets.yml
• customize config/docker.yml.erb

Exemplary configuration files are available in the config directory.

In order to execute code submissions using the DockerContainerPool, source code files are written to the file system and are provided to a dedicated Docker container. These files are temporarily written to Rails.root/tmp/files/. Please make sure that workspace_root in config/docker.yml.erb

• corresponds to that directory or to a linked directory if using a remote Docker server.
• is always writeable by the user executing the web server (in this case the codeocean user): setfacl -Rdm user:codeocean:rwx /var/www/app/current/tmp/files.

Optional Steps

• Use Docker Machine or vagrant if there is no native support for docker on your OS
• If you want to use the app without docker (and hence without code execution) comment the validation validate :working_docker_image? in models/execution_environments.rb otherwise the seed will fail (because of missing docker connection)
• Create seed data by executing rake db:seed
• If you already created a configuration for your local installation and want to use vagrant, too, be sure to log into the vagrant instance via ssh and add your database user manually to the database. Afterwards, create, migrate and seed.

Production Setup

• We recommend using Capistrano for deployment.
• Once deployed, CodeOcean assumes to run exclusively under a (sub)domain. If you want to use it under a custom subpath, you can specify the desired path using an environment variable: RAILS_RELATIVE_URL_ROOT=/codeocean. Please ensure to rebuild all assets and restart the server to apply the new path.

Monitoring

• We use a Prometheus Exporter and a Telegraf Client
• The Telegraf client collects the data from the Prometheus endpoint, adds its own datasets and forwards them to an InfluxDB
• The Prometheus Exporter must be started separately before running the Rails server via bundle exec prometheus_exporter
• The InfluxDB data can be visualized using Grafana, for example. There is also an adapted dashboard for this purpose

Additional Note

• If you want to change the default port of the underlying rails server, you can use authbind to bind it to the regular 80/443 port.