Open Stack

On Mon, Oct 12, 2015 at 7:46 PM, Cory Hawkless <Cory at hawkless.id.au> wrote:

> In my setup I have one Ceph cluster but 2 different pools, one for images
> which is on SATA disks and one for volumes which is on faster SAS disks.
>
> Given this setup I don’t imagine there is any way to thin provision an
> image to a volume? It would need to be a complete copy from one pool to
> another, yes?
>
>
>
> I’m certainly not in a position to be able to contribute to a code change,
> I’m surprised this hasn’t been done already, it seems terribly inefficient
> to have to copy the images twice?
>
>
>
> Regards,
>
> Cory
>
>
>
> *From:* Avishay Traeger [mailto:avishay at stratoscale.com]
> *Sent:* Monday, 12 October 2015 9:55 PM
> *To:* Cory Hawkless <Cory at Hawkless.id.au>
> *Cc:* openstack at lists.openstack.org
> *Subject:* Re: [Openstack] Volume from image question
>
>
>
> The flow for all images for this process is to download from Glance to a
> temporary file, and then write to volume.  This is not necessary for raw
> images, but that optimization has not been done.  I did leave a comment
> about that in the code though 2.5 years ago, but never implemented it - you
> can give it a go if you'd like :-)
>
>
>
> Are you using two different Ceph clusters, one for images and one for
> volumes?  Otherwise it should just be doing a thin provisioned clone of the
> image (no download, no temp space, no upload).
>
>
>
> On Mon, Oct 12, 2015 at 12:27 PM, Cory Hawkless <Cory at hawkless.id.au>
> wrote:
>
> Hi all,
>
>
>
> When creating a volume from an image(Using Horizon), why does the Cinder
> server need to do what appears to be a conversion of the image before it
> can create the volume?
>
> All of my images in Glance are uploaded in RAW format, images and volumes
> are stored in Ceph.
>
>
>
> The reason I know the images are being processed on my glance server is
> because it runs out of disk space when trying to make volumes from large
> images and the process fails. I can see the temporary file in
> /var/lib/cinder/conversion
>
>
>
> Is it not possible to have glance simply copy the image form the images
> store into the volumes store? I am going to regularly be creating new
> Windows Server instances, so it would take quite some time for a 20gb image
> to be processed by Cinder before it can be uploaded into Ceph.
>
>
>
> Regards,
>
> Cory
>
>
> _______________________________________________
> Mailing list:
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>
>
>
>
>
> --
>
> *Avishay Traeger, PhD*
>
> *System Architect*
>
>
>
> Mobile: +972 54 447 1475
>
> E-mail: avishay at stratoscale.com
>
>
>
>
>
> Web <http://www.stratoscale.com/> | Blog
> <http://www.stratoscale.com/blog/> | Twitter
> <https://twitter.com/Stratoscale> | Google+
> <https://plus.google.com/u/1/b/108421603458396133912/108421603458396133912/posts>
>  | Linkedin <https://www.linkedin.com/company/stratoscale>
>
> _______________________________________________
> Mailing list:
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>
> ​So trying to jog my memory a bit on this.  The initial work was done
here: [1]

A concern was raised but perhaps it was an IRC conversation as opposed to a
formal bug or review, but I reworked this after patch set 2 and added the
piece that does the conversion regardless of format.

Anyway, it is quite possible that Ceph can be configured so that this is
not an issue.  I seem to recall that you could in fact use it's internal
clones (that was actually what prompted me to work on this feature in the
first place).

Also, FWIW; in Liberty we fixed this up considerably by adding a cache
layer.  You'll still have this expense on initial creation so I guess it
won't really help you.

Sorry, I don't have better documentation for you on the reasoning behind
the change.  Perhaps some folks with more Ceph experience can chime in and
offer up a possible solution?

John
[1]: https://review.openstack.org/#/c/17437/2/cinder/volume/driver.py​
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20151012/86d578d0/attachment.html>