We’re going to need to free up a hypervisor and put its load on other hypervisors, in order to pull out the one hypervisor and have some of its faulty hardware replaced — but there’s two problems;
The hypervisor to free up has asserted required CPU capabilities most of the eligible targets do not have — this prevents a migration that does not involve a shut down, reconfiguration, and restart of the guest.
Furthermore, we’ve configured some of the disks with writeback caches, which also prevents migration without a proper nudge. Sigh.
This means I’m going to need to adjust configuration for, shut off, move over and fire back up some guests that are — you’ve got it — not redundant.
Some of these changes have to do with the VM’s CPU architecture that virt-install appears to choose by default (including SandyBridge on top of which IvyBridge features are added explicitly). These things are incompatible. Surprise.
The application of the emergency service window will last from 23:00 CEST to 23:45 CEST. It’ll impact some ~20% of users, in chunks of ~4% for ~5 minutes at most, each.