Ask HN: You want to build new Twitter. Your Architecture/Tech stack? | Hacker News Reader

Ask HN: You want to build new Twitter. Your Architecture/Tech stack?

5 pointswg09 comments

Hey Architects and Engineers! Lend me your ears!

Let's say you have your savings. And you want to build a new Twitter with pretty much the same feature set.

Where would you do that and how? What your tech stack would look like? Postgres, RabbitMQ, Kafka, Redis, OpenSearch, Cockroach DB etc and what piece would do what?

Where would you host it intially and eventually? AWS? GCP? Azure? Hetzner?

How much time to the MVP? Backend in Golang/Python or something else? Frontend React/Sevelte or what?

I'm looking for a complete comprehensive plan that is practical. Think of it as something you want to showcase as your pragmatic rational technical excellence.

With these words, I leave the stage to you.

Take a look at whats up. I would use something like erlang.

The front end is less of a concern.

But in general the site is huge (at least 500K machines).

It will be roughly like Mastodon, but for large files, I will use ipfs or BitTorrent.

I like the idea of bluesky where user data stay in version-controlled repos and can be migrated between servers. But I guess I will use crdt for that.

I will use rust for the backend, VUE for the frontend.

I'd use shell scripts... Bash is the only way to ensure stability and backwards compatibly so that aliens can see the "public (yet commercially private) town square".

Just do it in with react/MERN and get it done with quickly. A twitter clone MVP can be made in a day. It’s one of the most basic tutorials.

Scaling the back end is tough, the problem of handling the notifications when somebody with a few million followers tweet particularly.

If I remember right, the first version of Twitter was written in Ruby on Rails and they had the hardest time keeping it up, they got VC money and brought in some distributed systems experts quickly.

Complexity is probably in generating timelines. A clone that works for let's say 100 users would be doable maybe in a day or two.

BBC Basic and Econet.

The only legit answer.

Switch workflow orchestration to temporal!

Reply on news.ycombinator.com