I have a RHV 4.1 virtualized guest-server with a number of rather large 
VirtIO virtual disks attached. The virtual disks are allocated from a 
fibre channel (block) storage domain. The hypervisor servers run RHEL 7.4.

When I take a snapshot of the guest, then it takes a long time to remove 
the snapshots again, when the guest is powered off (a snapshot of a 2 TiB 
disk takes around 3 hours to remove). However, when the guest is running, 
then snapshot removal is very quick (perhaps around five minutes per 
snapshot). The involved disks have not been written much to while they 
had snapshots.

I would expect the opposite: I.e., when the guest is turned off, then I 
would assume that oVirt can handle snapshot removal in a much more 
aggressive fashion than when performing a live snapshot removal?

When performing offline snapshot removal, then on the hypervisor having 
the SPM role, I see the following in output from "ps xauw":

vdsm 10255 8.3 0.0 389144 27196 ? S<l 13:40 7:04 /usr/bin/qemu-img 
convert -p -t none -T none -f qcow2 /rhev/data-center/mnt/blockSD/xxx/
images/yyy/zzz -O raw /rhev/data-center/mnt/blockSD/xxx/images/yyy/

I don't see the same kind of process running on a guest's hypervisor when 
online snapshot removal is in progress.

I've read most of https://www.ovirt.org/develop/release-management/
My interpretation from that document is that I should expect to see "qemu-
img commit" commands instead of "qemu-img convert" processes. Or?

The RHV system involved is somewhat old, having been upgrade many times 
from 3.x through 4.1. Could it be that it carries around old left-overs 
which results in obsolete snapshot removal behavior?

Troels Arvin

Users mailing list

Reply via email to