Hi,
I think certain things won’t be possible at all as all metadata that is pretty much a ‘source of truth’ is lost forever both for VMs and for example for ovirt hosts.
First of all check thread like:
https://lists.ovirt.org/archives/list/users@ovirt.org/thread/6LSE4MNEBGODIRPVAQCUNBO2KGCCQTM5/
1) Copy all images (like QCOW2, probably you don’t use Raw?):
Consider that you have certain number of Libvirt/QEMU VMs and imagine as if they were never controlled and try to have them somewhere somewhere else and maybe you can
reach the point where you can launch them without any Ovirt…
So save Libvirt XMLs somewhere as well, you can recreate that ‘definition in ovirt’ in case you are attentive or troubleshoot something later on.
Check out the notion of internal/external snapshots and you need to understand that snapshots that customer has done influence how those QCOW2 files are implemented
and influenced by snapshots.
2) Then just start create new cluster and it will have like 1 KVM host and use that import from QCOW2 feature and then gradually add hosts one-by-one and consider that
host must be ‘empty’ before you try to do something with it using new ovengine)
I am pretty much sure that trying to connect old hosts to new ovengine will not be possible and will easily lead to data loss.
Somehow I think that it might be useful to know about mounting QCOW2 as disk
https://gist.github.com/shamil/62935d9b456a6f9877b5 which might allow you to get information from VM and then you will just launch new VMs (that’s where infrastructure-as-code
comes in handy) and use only certain directories that are needed from disks of those VMs.
BR,
Konstantin