Hey everyone! I'm working on an application that needs real-time data from an API using the WebSocket protocol. As of June 2025, I'm curious about the best library to go for WebSockets in Python. Currently, the module I'm using isn't too complex—it just needs to handle about 50-100 concurrent connections from the API with a data flow of around 10 bytes per second per connection. I anticipate that the number of concurrent connections could grow to 3000 or even more, so scalability is crucial. I did some searching here and in other forums about the websockets library, but surprisingly, I didn't find much discussion on it. I thought this library was quite popular! What do you all think about using it?
5 Answers
It really depends on whether this is backend-to-backend or client-to-backend. For backend communications, gRPC might serve you better than WebSockets. If you must stick with sockets (because HTTP/2 isn't an option), then websockets can work, especially with frameworks like FastAPI or Starlette that provide nice wrappers around it.
Are you building just the API to serve socketed endpoints or are you working on the client side too? It's not entirely clear from your post.
I'm actually writing a client script that connects to the socket endpoints provided by the API.
We use FastAPI's WebSocket integration in our production environment. It's got a bit of a reputation for its heavy emoji use in the documentation, but people seem to love its ease of use. I've seen mixed feelings about its bandwidth capabilities on Reddit. My app isn't super high-bandwidth right now, but it could upgrade in the future, so scalability is my top priority. Have you or your team run into any issues with FastAPI on both regular HTTP requests and WebSocket implementation?
If you're looking for performance, aiohttp has solid WebSocket support. It's pretty fast and efficient, so that might be worth checking out!
From what I've seen, Python-based WebSocket servers can struggle with scaling beyond about 100 concurrent connections, especially using FastAPI and Uvicorn. If high load is expected, I’d suggest considering a language that handles concurrency more efficiently, like Go or Rust. In our experience, we’ve had issues like dropped connections at scale. Though we use a Kubernetes setup for weight balancing, it can still get tricky with Python's limitations in handling many persistent connections effectively.
My Python script needs to connect to the socket endpoints that the API provides, so it's more about fetching the data from that.