[openstack-dev] [Ceph] Is it necessary to flatten Copy-on-write cloning for RBD-backed disks?

Sebastien Han seb at redhat.com
Tue Jun 9 09:52:10 UTC 2015

Hi Kun,

> On 09 Jun 2015, at 05:34, Kun Feng <fengkun32 at gmail.com> wrote:
> Hi all,
> I'm using ceph as storage backend for Nova and Glance, and merged the rbd-ephemeral-clone patch into Nova. As VM disks are Copy-on-write clones of a image, I have some concerns about this:
> 1. Since hundreds of vm disks based on one base file, is there any performance problems that IOs are loaded on this one paticular base file?

This may be an issue but as Clint mentioned you’ll get reads served from OSD memory anyway, so this should be expectable.

> 2. Is it possible that the data of base file is damaged or PG/OSD containing data of this base file out of service, resulting as all the VMs based on that base file malfunctioned?

This assumption is correct, if the parents gets a corrupted block that hasn’t been written to the clone yet, your clone will get the corrupted block on a write request.

> If so, flattening Copy-on-write clones may do some help. Is it necessary to do it?

People have expressed the concern, but no one has ever seen such thing happening (as far as I know), so it’s really up to you to flatten the clone.
There is a patch that needs to be reworked that will allow to flatten all the clone after their creation in Nova. Hopefully this will get into Liberty.

> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

Sébastien Han 
Senior Cloud Architect 

"Always give 100%. Unless you're giving blood."

Mail: seb at redhat.com 
Address: 11 bis, rue Roquépine - 75008 Paris

More information about the OpenStack-dev mailing list