Why We Moved a Client Off Lambda: A Cold-Start Case Study (2% → 0.01% Error Rate)

TL;DR — A client’s core Lambda function was cold-starting under uneven traffic, and those cold starts were failing outright rather than just running slow — a 2% error rate that propagated into 5 downstream services. Moving the function off Lambda onto an always-on EC2 instance eliminated the cold start entirely and cut the error rate […]