Depends … KVM hosts are very usefull when you want to mutualize hardware to deploy some specific workload where the underlying virtualization host don’t have to bother about high availability because HA is done at the application level.
For example, in our use case, we use KVM hosts to deploy openstack/Kubernetes control plane components.
Those components are HA by design (or through the help of some other components) and that way, we can mutualize 3 big hypervisors to support those VMs for multiple clusters while we would need to deploy at least 3 bare metal machines per cluster to do the same with MaaS/Juju.
Cherry on the cake : with MaaS, you can very easily deploy a bare metal hypervisor that will be automatically be configured as a KVM host you can use right away to deploy those virtualised control planes I was talking about.
No more chicken and eggs dilemma.
So just to sumarize, I love this KVM host feature but it lacks just a few things to be a lot more usable (like mass compose or more precise hardware tuning)
Makes me think that a kind of “hardware template” would be nice … let’s say there would a tab where you can create a hardware template where you set things like CPU type, vcpu numbers, disk type and layout, network configuration, … and then, at compose time, you just select a hardware template to compose “n machines of type x with base name ZZZ”.
That would be awesome and would avoid the “clickodrome” effect.
Same thing for hardware commissioning would be really awesome too !