Marsha, a self-hosted open source video and document provider 🐠

Overview

Marsha is a video management & playback service. It is intended to be operated independently: it's like having your very own YouTube for education.

Marsha also supports hosting documents and distribute them on all your courses.

Instructors & organizations can use Marsha to upload and manage their videos (and associated files, such as subtitles or transcripts) or documents directly from a course as they are creating it.

Once the course is published, learners simply see a video player or documents in the course.

This seamless integration works with any LMS (Open edX, Moodle, ...) thanks to the LTI standard for interoperability.

Here is what Marsha offers out of the box:

Video:

• automatic transcoding of videos to all suitable formats from a single video file uploaded by the instructor;
• adaptive-bitrate streaming playback (both HLS and DASH);
• accessibility through the player itself and support for subtitles, closed captions and transcripts;

Document:

• upload any type of documents;
• prevent disk storage quota by using AWS S3;

Moreover, Marsha provides:

• access control to resources through LTI authentication;
• easy deployment & management of environments through Terraform;

Architecture

Marsha is made up of 3 building blocks: a container-native Django backend, an AWS transcoding and file storage environment, and a React frontend application.

The Django backend

The Django backend is tasked with serving the LTI pages that are integrated into the LMS. It also manages all the objects with their relationships, user accounts and all authentication concerns. It exposes a JSON API to communicate with the part of the infrastructure that operates on AWS lambdas and the React frontend.

It is defined using a docker-compose file for development, and can be deployed on any container environment (such as Kubernetes) for production.

The storage & transcoding environment

Source files (video, documents, subtitles,...) are directly uploaded to an S3 bucket by instructors. Depending the uploaded resource a lambda will be triggered to do different jobs:

• Launch MediaConvert to generate all necessary video files (various formats and fragments & manifests for adaptive-bitrate streaming) into a destination S3 bucket. Those files are then served through the CloudFront CDN.
• Convert any kind of subtitles (also captions and transcripts) in WebVTT format and encode them properly.
• Resize thumbnails in many formats.
• Copy documents from a source to a destination S3 Bucket accessible through the CloudFront CDN.

Lambdas are used to manage and monitor the process and report back to the Django backend.

This storage & transcoding environment requires AWS as it heavily relies on AWS MediaConvert to do the heavy lifting when it comes to transcoding. All the services it relies on are configured through Terraform and can be deployed effortlessly through a make command.

⚠️ Privacy concerns

Please note that the only objects we handle in AWS are the actual video, documents or subtitles files, from the upload to the distribution through transcoding and storage. It is not required to deploy any database or application backend to AWS or send any user's personal information there.

The React frontend

The React frontend is responsible for the interfaces with which users interact in the LTI Iframes. It gets an authenticated token with permissions from the view and interacts with the Django backend to manage objects and directly with AWS s3 to upload files.

It also powers the same resource view when loaded by a learner to display a video player (thanks to videojs) or a document reader.

⚠️ Iframe management

To have the best possible user experience for instructors, we need to be able to change the size of the <iframe> depending on its contents. This can be done through the iframe-resizer library.

iframe-resizer requires to run some JS inside the <iframe> (which we include with our React frontend bundle) and some JS inside the host page. It then communicates through message-passing to adjust the size of the <iframe>.

This means that to have the best interfaces for instructors, you need to include the host-side iframe-resizer JS in your LMS pages. For Open edX, this is already done in our custom LTI consumer Xblock.

If you cannot or do not want to include this host-side JS, you can still run Marsha. It will work exactly the same for learners (provided you adjust the size of the LTI <iframe> for video), and instructors will only have to scroll inside the <iframe> on some occasions.

Getting started

Make sure you have a recent version of Docker and Docker Compose installed on your laptop:

$ docker -v
  Docker version 19.03.6, build 369ce74a3c

$ docker-compose --version
  docker-compose version 1.24.1, build 4667896b

⚠️ You may need to run the following commands with sudo but this can be avoided by assigning your user to the docker group.

The storage & transcoding environment

All tasks related to this environment are run from the ./src/aws directory. We use Terraform to keep this infrastructure configuration as code and easily manage several independent deployments of the whole AWS infrastructure.

Note for Mac users: Marsha's AWS development setup uses getopt. The version that comes with macOS is not suitable for our use case. You need to install the GNU version and add it to your path so it is used by default.

brew install gnu-getopt

echo 'export PATH="/usr/local/opt/gnu-getopt/bin:$PATH"' >> ~/.zshrc

There are 2 Terraform projects in Marsha with two different purposes:

• ./src/aws/shared_resources: this project manages resources common to all marsha environments on the same AWS account. These resources must not live in different workspaces so you must work in the default workspace. To ease the use of this project, a dedicated script is available in ./src/aws/bin/shared-resources which uses and configures the Terraform docker image. You have to run a Terraform command as if you were using the terraform cli. (eg: ./bin/shared-resources plan will execute Terraform's "plan" command).
• ./src/aws: this is the main project we use, most of the infrastructure is managed here (in all *.tf files). This project must use Terraform workspaces and we highly recommand you to not use the default one. With multiple workspaces, you can manage multiple environments for your Marsha instance with a single AWS account. To ease the use of this project, a dedicated script is available in ./src/aws/bin/terraform which uses and configures the Terraform docker image. You have to run a Terraform command as if you were using the terraform cli. (eg: ./bin/terraform plan will execute Terraform's "plan" command).

Terraform state management

Terraform manages a state of your infrastructure. By default this state is stored locally on your machine but it is highly recommanded to use a remote backend.

You will find all you need to configure a remote backend in the Terraform documentation: https://www.terraform.io/docs/configuration/blocks/backends/index.html

⚠ You must configure your state management before running any of the commands hereafter. The first init will initiate your state and after that you will have to deal with state migration if you want to modify it. You can create a file src/aws/state.tf and src/aws/shared-resources/state.tf to configure a backend, there is an example in each project (state.tf.dist file).

🔧 Before you go further, you need to create ./src/aws/env.d/development and replace the relevant values with your own. You can take a look at the environment documentation for more details on this topic. You should use this command to create the file from the existing model:

$ cp ./src/aws/env.d/development.dist ./src/aws/env.d/development

Initialize your Terraform config:

$ cd src/aws
$ make init

The make init command will also create an ECR repository. Before going further you have to build and publish the lambda docker image. Unfortunately AWS doesn't allow to use a public image, so you have to host this one on a private ECR instance. Copy the output of the init command, you will use them in the next step.

Build and publish the lambda image

For this step, we cooked a script to help you build, tag and deploy images. All the scripts are run from the marsha root directory.

🔧 Before you go further, you need to create ./env.d/lambda and replace the relevant values with your own. The ECR url is available in the shared_resources terraform output you copied earlier. You should use this command to create the file from the existing model:

$ cp ./env.d/lambda.dist ./env.d/lambda

You have to successively run these commands :

Build the image:

$ ./bin/lambda build

Tag the image:

$ ./bin/lambda tag

And then publish it:

$ ./bin/lambda publish

Apply all terraform plans

Terraform is split in two parts. The main one, directly in src/aws can work on multiple Terraform workspaces. You will use this feature if you want separate environments (development, staging, preprod and production). We also need some resources available across all workspaces. For this we have an other terraform in src/aws/shared_resources. To apply all plans at once run this command in the src/aws directory.

$ make apply-all

Everything should be set up! You can check on your AWS management console.

You may have noticed that the AWS development environment requires a URL where the Django backend is running. You can easily get a URL that points to your locally running Django app using a tool such as ngrok.

The Django Backend

All tasks related to the Django backend are run from the project root (where this README.md is located).

The easiest way to start working on the project is to use our Makefile:

$ make bootstrap

This command builds the app container, installs back-end dependencies and performs database migrations. It's a good idea to use this command each time you are pulling code from the project repository to avoid dependency-related or migration-related issues.

🔧 Before you go further, you should take a look at the newly created ./env.d/development file and replace the relevant values with your own. You can take a look at the environment documentation for more details on this topic.

Now that your Docker services are ready to be used, start the application by running:

$ make run

You should be able to view the development view at localhost:8060/development/.

The React frontend

All tasks related to the React frontend are run from the ./src/frontend directory.

We use yarn for all those tasks. Make sure you have a recent version installed:

$ yarn --version
  1.13.0

If you need to install yarn, please take a look at the official documentation.

Install all the dependencies:

$ yarn install

The LTI frontend

Run the build and copy the iframe-resizer host-side JS into your local Django assets:

$ yarn build-lti
$ yarn copy-iframe-resizer

The "Standalone" frontend

Run the build for libraries and start the standalone frontend:

$ yarn build-tests
$ yarn start-site

Now the standalone frontend is available at localhost:3000.

Testing frontend

The front application is tested using jest. Every js module has its corresponding spec file containing related tests:

├── VideoPlayer
│   ├── index.spec.tsx
│   ├── index.tsx

Run the tests (this will run all tests: for both LTI and standalone frontends and libraries):

$ yarn test

Browser testing provided by:

🗝 Before you go further, you need to create a Consumer Site and Passport in Marsha's admin panel.

You should be all set to make the LTI request on the development view and access Marsha's frontend interface!

Contributing

This project is intended to be community-driven, so please, do not hesitate to get in touch if you have any question related to our implementation or design decisions.

We try to raise our code quality standards and expect contributors to follow the recommandations from our handbook.

License

This work is released under the MIT License (see LICENSE).