I’m not sure about operator, some resources especially during instrumentation has to be consumed but it is rather small number. We’ve never faced impact because of this.
On pod level consumption is like for any other monitored process. From my experience it’s much less than 0.5% cpu and up to 70-100mb ram (when you have more than one process). In general it should be 40mb ram per process.