Has anyone scaled their custom agents deployment on k8s infra? Do you see CPU being the bottleneck or Memory? When I am running tests - the bottleneck seems to be memory rather than CPU but the default load_threshold function uses CPU as the metric. Can someone please confirm any scaling pieces that you have configured on your end?

on my bare machine is mem,

Thanks for replying. So have you scaled based on memory alone - meaning the HPA uses memory to scale up the pods?