Updated | 5000 Rps
This is the golden rule of horizontal scaling. Your application servers must not store user data locally (local sessions or file uploads). If a server stores state, and that server dies, the user’s session is lost. By storing state in an external store (like Redis), you can add or remove application servers at will without disrupting the user experience.
No single server can handle 5,000 RPS of dynamic content reliably. A Load Balancer (such as NGINX or HAProxy) sits in front of your application servers, distributing traffic across a cluster of nodes. If one node crashes, the balancer routes traffic to the healthy ones, ensuring zero downtime. 5000 rps