Open Stack

Hi Caius,

This has existed in the rbd cinder driver since volume-to-image was added:

https://github.com/openstack/cinder/blob/stable/kilo/cinder/volume/drivers/rbd.py#L823

Cinder falls back to doing the full copy if glance doesn't report the
location, or it's not marked as raw format.

If glance doesn't have show_image_direct_url = True, or cinder doesn't
have glance_api_version = 2, cinder won't be able to do the clone. See

http://ceph.com/docs/master/rbd/rbd-openstack/#configure-openstack-to-use-ceph

for more details.

Josh


On 07/29/2015 07:36 AM, Caius Howcroft wrote:
> Hi,
>
> We (bloomberg) are preparing to roll out kilo into production and one
> thing is causing a lot of grief. I wonder if anyone else has
> encountered it.
>
> We run BCPC (https://github.com/bloomberg/chef-bcpc) which is ceph
> backed. When we boot an instance from volume the cinder create volume
> from image function (
> https://github.com/openstack/cinder/blob/stable/kilo/cinder/volume/drivers/rbd.py#L850)
> ends up pulling the entire image through the glance API, so lots of
> tenants doing this creates quite a bit of load on our API nodes.
>
> We were confused why it did this, when its way more efficient to go
> directly via rbd clone, we created a patch and tested and it seems to
> work just fine (and an order of magnitude faster)
> https://github.com/bloomberg/chef-bcpc/pull/742
>
> So, the question is: what are other ceph backed installations doing ?