Round Robin Is One Index
How do you spread requests evenly across backends when you are willing to know nothing else about them?
Open Session 01Sessions
Load Balancer track sessions
Least Connections Is a Counter
If you're willing to track one number per backend, what does that number buy you that round robin's blind index can't?
Open Session 02Hashing Keeps You on the Same Server
How do you guarantee the same client lands on the same backend every time, with no state at all — and what does that guarantee cost you when the backend list changes?
Open Session 03The Ring Survives a Server Change
Session 03 showed that folding a client key onto len(backends) with modulo remaps almost every key when membership changes. What has to change about the fold so that only the keys that truly need to move actually move?
Health Is a State Machine
None of the picking strategies taught so far know whether a backend is actually answering — only that it exists. What turns a stream of probe results into a routing decision, and why is going down fast and coming back slow the same design, not two different ones?
Open Session 05Build the Toy Load Balancer Loop
Five earlier files each answer one question in isolation — how to pick among backends assumed to be up, and how to know which ones actually are. What does the smallest object look like that filters a live request through health *before* handing it to whichever picking strategy is configured?
Open Session 06