kept mixing these up so i wrote it down, figured it might help someone. from weakest to strongest isolation:
containers (runc/crun): namespaces for isolation, cgroups for resource limits. everything shares the host kernel
gvisor (runsc): a userspace kernel sits in between and handles most syscalls itself, so only a small subset ever reaches the host kernel
kata containers: each container (or pod) runs in a micro-vm with its own kernel, via kvm. you still use it like a normal container
full vms (qemu/kvm): a whole os, windows included, with its own disk, init, ssh etc
the thing that clicked for me: kata and a full vm give you basicaly the same isolation, both are real vms on kvm. the difference is what's inside and how you use it.…