We have three special-purpose hypervisors, cloudvirt-wdqs100[1-3].eqiad.wmnet which are reserved for wdqs workloads.
In theory there are host aggregates preventing regular VMs from getting scheduled there. In practice, this seems to not be working properly.