I'm reluctant to give any advice in a situation as fragile as that, but I believe VDSM
doesn't want you to mess with virsh in oVirt: You're supposed to make things
happen with hosted-engine and potentially vdsm-tool.
Try to see if you can get the management VM down with hosted-engine --vm-shutdown, enable
maintenance with hosted-engine --set-maintenance --mode=global (as well as --mode=local
for each surviving node) and if that doesn't work then perhaps you'd want to stop
the vdsm broker/agent daemons while you do gluster repairs.