On 04/03/2013 14:57, Ricky Schneberger wrote:
On 2013-03-04 13:24, Jiri Belka wrote:
> On Mon, 04 Mar 2013 12:52:53 +0100 Ricky Schneberger
> <ricky(a)schneberger.se> wrote:
>
>>> Live snapshot does not save memory state, IIRC. So either
>>> shutdown all VMs or do not think about this as solid backup
>>> solution.
>>>
>>
>> Hi, If I dont think about it as a solid solution, is it doable?
>> Even with a manual live snapshot I will lack this memory state,
>> right?
>>
>> All I want if I must restore a snapshot is to get the WM up on
>> track in an earlier state.
>
> The "snapshot" like in VMWare world which saves even memory is
> called checkpoint in qemu-kvm world. And IIRC it is not implemented
> yet.
>
> Live snapshot is doable, no problem. The problem can be your
> design. As you know a lot of apps do not write to filesystem
> immediately, so snapshot backup is useless. So either
> shutdown/think twice or forget it.
>
> But others can have different view... Backup is thing you _must_
> trust, if you design is broken, the consequences are terrific! If
> your goal is to have fast disaster recovery, then why not to have
> backup machine or some clustered solution?
>
> jbelka
>
Thanks for your answer.
We have both Netvault and Acronis on the most of the VMs, but then we
have other VMs in the cluster that we want a daily scheduled snapshot
implementation on, but we dont want to setup a complete backup
solution on this VMs. It's testmachines, different virtual desktops,
OCR-readers and so on. We have no needs of snapshots like in VMWare
world for them, nor even a complete backup solution. There I think
Ovirts snapshot handling should be perfect.
in general, you are correct - you don't care about live snapshot with
memory, since you probably mostly care about backing up the disks.
you may care, depending on the type of guest, on having a
qemu-guest-agent installed, which would help with syncing the writes to
the disks before the live snapshot is taken.
something to test though is if a daily snapshot does not degrade the
performance of the running VM, due to more COW layers (until the backup
api and live merge are in place).