Little experiment using neo4j to power a twitter like backend database

author: Dalton Oliveira

Install dependencies

yarn

Development

yarn start

nodemon restarts the application upon any file change in .env or src/* files. This is useful as it can speed up development process.

Linting

yarn lint

Test

yarn test

Setup

Go to https://neo4j.com/cloud/platform/aura-graph-database/?ref=get-started-dropdown-cta and create a free databse to you. It will be given an url with user and password. Copy .env.local to .env and fill those fields with corresponding values. Then run the script npx ts-node src/scripts/loadUsers.ts

Planning

A feature reply-to-post is a relation that would looks like this: (author:User)-[:REPLIES]->(p:Post) and thus won't show on the existing list/listAll/listFollowing endpoints. No much change is required besides mapping this new endpoint and code the query:

MATCH (author:User)-[reply:REPLIES]->(p:Post)
OPTIONAL MATCH (p)-[:QUOTES]->(quoted:Post)<-[:POSTS]-(quotedUser:User)
RETURN author, p, reply, quoted, quotedUser
ORDER BY p.createdAt DESC SKIP $skip LIMIT 10

This is how it will be crested

MATCH (author:User { username: $username }), (p:Post { id: $postId })
CREATE author-[:REPLIES {text: $text, createdAt: datetime()}]->(p)

Critique

There's some important missing tests and express-validator is not so easy to test with jest. I'd need more time to elaborate better this validation chain pattern. It's also lacking a full integration test with a working database to be queried. These tests could acelerate further development.

I'm not completely sure that all indexes to have performant database were created. Having more time I would create a stress scenario to validate the response times.

The project is missing logs and on a real product I'd integrate it with a log analysis tool such as datadog or sentry.

I'm unsure what needs to be done to make it more scalable. Neo4j sells itself as a high performant database that can handle a petabyte scale. It was my first attempt do build somethis with it and I really enjoyed it.

I hadn't used a proper pattern for authorization and I did it just for sake of brevity to manually test. On a real application I'd do some research to check if jwt is still the best to use or if there are better ways of doing it.

On a production backend I'd also start it with pm2 for process monitoring.