---
title: "Comment by Raunak Singh on \"I Built a Real-Time Collaborative Whiteboard to Understand Stateful Backends\" | Hashnode"
description: "Splitting HTTP auth and history from WebSocket broadcast makes the architecture easy to follow. The next scaling challenge would be moving room membership out o"
url: "https://hashnode.com/posts/i-built-a-real-time-collaborative-whiteboard-to-understand-stateful-backends/6a9aa49b942b543f25d5f353/comment/6a9c3c699ccbacac1dad4e15"
---

[Discussion](https://hashnode.com/posts/i-built-a-real-time-collaborative-whiteboard-to-understand-stateful-backends/6a9aa49b942b543f25d5f353)

Comment [I Built a Real-Time Collaborative Whiteboard to Understand Stateful Backends](https://hashnode.com/posts/i-built-a-real-time-collaborative-whiteboard-to-understand-stateful-backends/6a9aa49b942b543f25d5f353)

[Raunak Singh](https://hashnode.com/@raunakbuilds)

[Sep 5](https://hashnode.com/posts/i-built-a-real-time-collaborative-whiteboard-to-understand-stateful-backends/6a9aa49b942b543f25d5f353/comment/6a9c3c699ccbacac1dad4e15)

Splitting HTTP auth and history from WebSocket broadcast makes the architecture easy to follow. The next scaling challenge would be moving room membership out of a single process. Are you planning Redis pub/sub or sticky sessions first?

[Anshuman Praharaj](https://hashnode.com/@anshcds)

[Sep 5](https://hashnode.com/posts/i-built-a-real-time-collaborative-whiteboard-to-understand-stateful-backends/6a9aa49b942b543f25d5f353/comment/6a9c3c699ccbacac1dad4e15#reply-6a9c75996473ececd44c2a7b)

yes will using pub/sub for scale is ofc the next thing

[Raunak Singh](https://hashnode.com/@raunakbuilds)

[Sep 16](https://hashnode.com/posts/i-built-a-real-time-collaborative-whiteboard-to-understand-stateful-backends/6a9aa49b942b543f25d5f353/comment/6a9c3c699ccbacac1dad4e15#reply-6aaa47e37e86753e945323bf)

Nice. Redis pub/sub is a natural next step. I would keep room state behind a small adapter so you can compare in-memory and Redis implementations without changing the WebSocket handlers. That should make the scaling path easier to test.
