Requests and limits that actually work

For memory, yes: memory is incompressible, so a limit above the request means the pod gets killed rather than throttled. For CPU, set requests and leave limits off unless you need hard isolation, because CPU limits throttle in ways that look like random latency.