/canvas-docker

Containerized Canvas instance for dev/testing

Primary LanguageDockerfileApache License 2.0Apache-2.0

canvas-docker

Overview

docker-canvas aims to provide a simple, disposable, containerized Canvas instance for fast(ish) integration testing of LTI applications.

Prerequisites

  • docker (developed & tested w/ v1.12.1)

Running

docker run --name canvas-docker -p 3000:3000 -d lbjay/canvas-docker

This repo is registered at Docker Hub as an automated build. So you should also be able to docker pull lbjay/canvas-docker to get the pre-built image.

Building

  1. Clone this repo somewhere.
  2. Build the image: docker build -t canvas-docker .
  3. Start the container: docker run -t -i -p 3000:3000 --name canvas-docker canvas-docker
  4. Point your browser to http://localhost:3000. The admin user/pass login is canvas@example.edu / canvas-docker.

The "fat" container

The Dockerfile and associated build scripts create a resulting docker image where all necessary services of the Canvas instance are run within a single container. This approach is sometimes called a "fat" container. This admittedly goes against the "Docker Philosophy" of one concern per container, but for the intended purposes of the image it offers a couple of advantages, chief among them, faster spin-up times. The functionality focus is on creating a tool for integration testing of external (LTI) apps, not general canvas development, scalability, or, god forbid, actual deployment.

Default developer_key & API access token

The image build includes the injection of default developer_key and access_token entries into the database.

  • developer key: test_developer_key
  • access token: canvas-docker

API requests should be possible, e.g.,

curl -H "Authorization: Bearer canvas-docker" http://localhost:3000/api/v1/courses

The developer key is for use with Canvas's OAuth2 Token Request Flow. For example, if you're making use of harvard-dce/django-canvas-api-token.

Outgoing Email

By default the instance's outgoing email delivery_method will be set to "test", meaning outgoing emails, such as user registration messages, will be sent to the container's stdout.

To configure 'smtp' delivery set the following $ENV values at runtime:

  • EMAIL_DELIVERY_METHOD (set this to "smtp")
  • SMTP_ADDRESS
  • SMTP_PORT
  • SMTP_USER
  • SMTP_PASS

Example using Mandrill:

docker run -d --name=canvas -p 3000:3000 -e EMAIL_DELIVERY_METHOD=smtp -e SMTP_ADDRESS=smtp.mandrillapp.com -e SMTP_PORT=587 -e SMTP_USER=<mandrill_user> -e SMTP_PASS=<mandrill_api_key> lbjay/canvas-docker

Details

  • The resulting canvas image is built and run using RAILS_ENV=development. At some point I might try creating a separate "production" flavor, but, because docker doesn't allow the setting of build-time variables except in the Dockerfile, it would require a separate Dockerfile. Also, when I did try building with RAILS_ENV=production, the resulting instance had issues with routing errors to the compiled assets, and the db:initial_setup rake task threw lots of warnings about missing triggers (?). So that.
  • Everything is currently somewhat "opinionated" in that things that would be nice to have configurable are hard-coded, e.g., postgres and canvas usernames, postgres network settings, path to the postgres data, etc.
  • The Dockerfile build process mostly follows Canvas's Quick Start guidelines with a few exeptions:
    • as mentioned above, RAILS_ENV=development
    • redis is installed, configured and used
    • the delated_job background task is executed
    • postgres is configured to not require a password for local connections, or for connections originating within a network defined by Docker's default network bridge setup: 172.17.0.0/16.

Contributors

License

Apache 2.0

Copyright

2016 President and Fellows of Harvard College