@Adam, I'm not aware of any nova changes since pike that have resolved this issue, but I can say anecdotally that I don't think we're seeing this in the upstream CI system anymore once we started using newer versions of libvirt/qemu in our CI jobs (I'd have to probably dig into the details of when this was reported and what libvirt/qemu versions were used versus when we stopped since this as a result of using a newer ubuntu cloud archive release or using ubuntu bionic nodes).
It would be helpful if you posted which versions of libvirt/qemu you're using on your pike and queens clusters so we could compare with what we're using in the upstream CI jobs for stable/pike and stable/queens (and beyond).
@Adam, I'm not aware of any nova changes since pike that have resolved this issue, but I can say anecdotally that I don't think we're seeing this in the upstream CI system anymore once we started using newer versions of libvirt/qemu in our CI jobs (I'd have to probably dig into the details of when this was reported and what libvirt/qemu versions were used versus when we stopped since this as a result of using a newer ubuntu cloud archive release or using ubuntu bionic nodes).
It would be helpful if you posted which versions of libvirt/qemu you're using on your pike and queens clusters so we could compare with what we're using in the upstream CI jobs for stable/pike and stable/queens (and beyond).