Apparently internal snapshots of a qcow2-based vm are problematic when
the vm boots from UEFI. I found that out when my nightly snapshots
stopped working after I upgraded a server from stretch to buster.
Testing the effective line of the script with actual values, I got this
result:
#virsh snapshot-create-as --domain server --name serverS0
error: Operation not supported: internal snapshots of a VM with
pflash based firmware are not supported
An internet search of the error message revealed that this problem has
been going on for a long time. The new behaviour is to simply not allow
the snapshot to take place rather than risk creating a corrupted one.
Fortunately there is a simple workaround. I simply replaced the virsh
snapshots with qemu-img snapshots. In my script, this meant changing a
few lines like this (VM is the name of the VM while DOW is the day of
the week number (0-6) - I put an 'S' between them in naming the snapshot):
from: virsh snapshot-create-as --domain $VM --name ${VM}S$DOW
to: qemu-img snapshot -c ${VM}S$DOW $VM.qcow2
While virsh works with the xml "domain" definition, qemu-img works
directly with the qcow2 file. This means the script also has to know
where the qcow2 file is. This can be accomplished either by adding the
full path to the VM name (which would prevent it from being used in the
virsh command to shut down the VM) or cd'ing to the folder. I chose the
latter as the simplest.
There's always more than one to do something. In this case, I'm glad
qemu-img was there to overcome a bug in libvirt. The other option would
be to switch to external snapshots which, apparently, are different
enough to not be affected by the internal snapshot bug. However they
operate a lot differently and I really don't want to get into them for
this application.
