I set up autoscaling based on CPU usage but our app is memory-bound so the servers run out of RAM while CPU stays low and autoscaling never triggers and users get errors.
More people experience this than they realize.
Automated systems built on generic metrics can systematically misunderstand the specific characteristics of the workloads they manage.
“Where Are You with Cloud computing and distributed everything?”
If this sounds familiar, the Library can help you find the bigger picture.