also, how much "virtual machine complexity" is added by running the VM's inside of a container? does this result in extra layers of translation of vm machine code? or is it still pretty "flat"? i.e. is it basically a container running as a process that spawns the qemu process? i would imagine that there is more overhead here than with proxmox, but as long as the container isn't doing much other than spawning a VM, that it's not adding too much overhead. my other questions are more important, but i'm just curious about this. I have an AMD 5950x and harvester is using 0.5% but has reserved about 25%. i plan on running a single node cluster. if i ran a multinode cluster, would this reserved CPU number drop for the non-primary nodes?

There is no container involved in the way you're thinking. It may be easier to think of it this way: the cluster is treating libvirt (running on each node) as if it were replacing the docker daemon (it's not, but think of it that way).

ah thank you. that clears things up. i'll have to look into it more. i've been listening to Devops and Docker off/on, but i'm just now getting back into development, so everything is new again.