Problem
Twice a day, a product launch email drives an 800 RPS flash spike that lasts about 90 seconds. Every time it hits, the fleet scales out reactively, and by the time a new instance finishes loading the model into memory; roughly 45 seconds; the spike is already over. Design the serving tier so the spike is actually absorbed, without paying for peak capacity sitting idle the other 23+ hours of the day.
Required components
Click to add a component to the canvas.
Additional components
Browse by category. Click or drag a component onto the canvas.