Depends on the workload characteristics. If there’s a chance one request could balloon into a huge RAM consumer, you want to be able to survive that without run

Depends on the workload characteristics. If there’s a chance one request could balloon into a huge RAM consumer, you want to be able to survive that without running out of RAM. But if you are hitting swap regularly across your fleet, it’s time to tweak your scaling parameters. But if you never ever hit it, you may be overscaled.