This is a demo of using Node.js with TypeScript to expose a HTTP API. A 'user' can request to start a video, end a video or ask for a count of how many videos they have started.
The application can either be hosted using the Node.js HTTP server (index.ts) or AWS Lambda (index-aws-lambda.ts).
The requests will be routed to the appropriate action on a controller, where the watch count for the given user will either be incremented, decremented or just returned as a count. If using the Node.js http server, an in memory repository is used for storing watch counts for users - this is ideal for local development. Clearly this isn't a scalable solution, thus when deployed to AWS a DynamoDB repository is injected instead.
If a user tries to start more than 3 videos at once, the API will return a 403 forbidden status code. Otherwise the user will get a 200 OK.
The following curl commands will make a request to the endpoint and return the headers, along with the number of videos in the return body.
Where Adrian is the username for the user, used for tracking the number of videos being watched.
Start Video (increments count): $ curl -i -H "x-username: Adrian" -X GET https://m6l3t9dvwl.execute-api.us-east-1.amazonaws.com/Prod/startVideo
Stop Video (decrements count): $ curl -i -H "x-username: Adrian" -X GET https://m6l3t9dvwl.execute-api.us-east-1.amazonaws.com/Prod/endVideo
Count (returns count): $ curl -i -H "x-username: Adrian" -X GET https://m6l3t9dvwl.execute-api.us-east-1.amazonaws.com/Prod/count
Start Video (increments count): $ curl -i -H "x-username: Adrian" -X GET http://localhost:3000/videos/start
Stop Video (decrements count): $ curl -i -H "x-username: Adrian" -X GET http://localhost:3000/videos/end
Count (returns count): $ curl -i -H "x-username: Adrian" -X GET http://localhost:3000/videos/count
- Run
$ npm installat the root of the repository
- Open repository in VS Code
- Hit F5
$ npm start
You can run the code in this repository using node directly, or with Docker. The application itself is deployed to AWS Lambda, and described using the aws-lambda.sam-template.yaml SAM template. This section covers deployment to AWS using the CLI. You must have the CLI configured against your account before continuing.
- Create a bucket: A bucket needs to exist before you can upload the package to it, you can create the bucket with this command:
aws s3 mb s3://node-videostream-api - Upload package:
$ npm run aws:package - Deploy package:
$ npm run aws:deploy
AWS Lambda has been used to provide automatic scalability of the functions. AWS will auto provision any resources required depending on load.
For tracking how many videos are being watched by a user, data is persisted using DynamoDB within AWS - which is also autoscaled. The username of the user has been used as the primary key (thus used for partitioning).
If hosting outside of AWS with DynamoDB, consideration is needed around the repository used for persistenting the watch count. One solution is to ensure that multiple instances of the API only use a single database, but this database will become a bottleneck as load increases. I'd recommend sharding on the username to allow for scaling out the database.
In a production system, a token would be used to authorise whether the watch count can be de/incremented. For simplicity auth has been omitted, and the user ID is placed in the HTTP header instead. Using the user ID is a better solution to say an IP address, as there could be multiple users behind the same shared internet connection.
Unit tests have been written throughout where deemed sensible. They can be run using: $ npm run test. I have used Mocha (test runner), Chai (assertion), Sinon (spies) and TypeMoq (Mocks). Note: Mocks are available within Sinon but type safety compatability with TypeScript is better with TypeMoq.
The application has been tested on:
- Ubuntu 17.10
- Windows 10
- AWS Lambda
TSLint has been used for code linting. You can run this using: $ npm run lint.
- If the scope of the logic within the controller got any more complicated, it would be worthwhile encapsulating the logic in it's own service / provider, rather than in the controller. For simplicity of the demo it has been left as is.