Production goes down because there's a memory leak due to scale.

When you say it in one sentence, it sounds too easy. Being developers we know how it all goes. It starts with an alert ping, then one server instance goes down, then the next. First you start debugging from your code, then the application servers, then the web servers and by that time, you're already on the tips of your toes. Then you realize that the application and application servers have been gradually losing memory over a period of time. If the application is one that don't get re-deployed ever so often, the complexity grows faster. No anomaly / change detection monitor can detect a gradual decrease of memory over a period of months.