The TL;dr of how Vercel's Fluid compute works: It's a variant of the traditional Node.js load balancing strategies with a per VM instance hypervisor[1]. More detailed post to come soon [1] not the correct technical term, but a good metaphor
The TL;dr is that we run a type of "hypervisor" written in Rust inside of the function. It measures how responsive the node and python processes are, and handles communication to the load balancer. The LB makes decision based on the historic data but the hypervisor also gets to