Autoscaling gets slapped onto nearly every deployment, and the usual results are surprise bills and cascading failures. Here's the uncomfortable part: autoscaling isn't a scaling solution at all — it's a cost optimization. You still have to design and test for peak load yourself. Worse, the moment you enable it you've adopted a feedback control system — with dead time, gain, and coupling to your load balancer — and almost nobody treats it as one. This talk is about what autoscaling actually is, why "just turn on the HPA" produces oscillation and outage amplification, and what responsible use looks like: design for max scale first, keep the scaling units well-behaved, protect upstream resources, and pick your control loop on purpose.
Independent researcher