LVM thin provisioning and virt-manager

[Chris Murphy]

On Oct 15, 2013, at 3:46 PM, Richard W.M. Jones <rjones at redhat.com> wrote:

> On Tue, Oct 15, 2013 at 03:29:38PM -0600, Chris Murphy wrote:
>> The better snapshots sound ideal for VM testing. Snapshot a
>> successful install and then try to break the snapshot. Etc.
>> 
>> Presently virt-manager ignores thinp pools and only creates
>> conventional LV's. I haven't tried using virsh to force it to use an
>> already created virtualsize LV as backing, but I'm wondering if it
>> should work. If not, is there a rough time frame on such support?
> 
> Is using LVs for this over-thinking things?

Could be.

> Creating a snapshot of a regular file which efficiently shares the
> backing disk is easy, and doesn't require root or special support:
> 
>  qemu-img create -f qcow2 -b original snapshot.qcow2

Yes but unallocated qcow2 is slower than either kind of LV. And it's noticeable when doing something like guest using XFS in a qcow2 file on btrfs host, or guest using btrfs in a qcow2 file on btrfs host. It's double the file system activity. On an LV there's pretty close to no hit.

I'm happy to hear it argued I should instead use the LV space for a regular partition formatted ext2 and drop the qcow2 file there. There's still overhead of two file systems there, but may compare to LV performance.

> 
> Then you can import this as a new guest in libvirt, again *without*
> needing root:
> 
>  virt-install --import --name snapshot \
>    --ram 1024 --disk path=snapshot.qcow2,format=qcow2
> 
> And in Fedora 20 we'll have virt-builder, which makes creating the
> original images quick too.  (Not to mention virt-sysprep and all the
> other tools to manipulate disk images easily, without root)

Hmm good to know.


Chris Murphy