The libvirt snapshots I created in T329970 are copy-on-write snapshots that didn't use any disk space initially, but are using a growing amount of space that means we now have less than 5% free disk space on both cloudvirt1019 and cloudvirt1020.
We could merge the different qcow2 snapshot files into a single file (documentation here):
virsh blockcommit vm1 vda --verbose --pivot --active
I have two concerns:
- it's not clear if that operation requires a bigger amount of free disk space than what we have now
- while it can be performed live, it could have an impact on database performance
I'm thinking of just freeing up some space on disk instead, so that the snapshot can continue to grow for a few more weeks, until we decommission both servers.
There are about 1.5TB that are used by the disk images of clouddb1003 and clouddb1004. Those VMs were deprecated and stopped in T323159, so I believe that after 1 month we can now delete those disk images and free up 1.5TB on both hypervisors (cloudvirt1019 and cloudvirt1020).