Open Stack

On Mon, Jul 1, 2013 at 11:14 AM, Vishvananda Ishaya
<vishvananda at gmail.com>wrote:

>
> On Jul 1, 2013, at 3:35 AM, Sheng Bo Hou <sbhou at cn.ibm.com> wrote:
>
> Hi Mate,
>
> First, thanks for answering.
> I was trying to find the way to prepare the bootable volume.
> Take the default image downloaded by devstack, there are three images:
> cirros-0.3.0-x86_64-uec, cirros-0.3.0-x86_64-uec-kernel and
> cirros-0.3.0-x86_64-uec-ramdisk.
> cirros-0.3.0-x86_64-uec-kernel is referred as the kernel image and
> cirros-0.3.0-x86_64-uec-ramdisk is referred as the ramdisk image.
>
> *Issue:* If only the image(cirros-0.3.0-x86_64-uec) is copied to the
> volume when creating a volume) from an image, this volume is unable to boot
> an instance without the references to the kernel and the ramdisk images.
> The current cinder only copies the image cirros-0.3.0-x86_64-uec to one
> targeted volume(*Vol-1*), which is marked as bootable but unable to do a
> successful boot with the current nova code, even if image-id is removed in
> the parameter.
>
> *Possible solutions:* There are two ways in my mind to resolve it. One is
> we just need the code change in Nova to let it find the reference images
> for the bootable volume(*Vol-1*) and there is no need to change anything
> in cinder, since the kernel and ramdisk id are saved in the
> volume_glance_metadata, where the references point to the images(kernel and
> ramdisk) for the volume(*Vol-1*).
>
>
> You should be able to create an image in glance that references the volume
> in block device mapping but also has a kernel_id and ramdisk_id parameter
> so it can boot properly. I know this is kind of an odd way to do things,
> but this seems like an edge case and I think it is a valid workaround.
>
> Vish
>
> The other is that if we need multiple images to boot an instance, we need
> a new way to create the bootable volume. For example, we can create three
> separate volumes for three of the images and set the new references in
> volume_glance_metadata with the kernel_volume_id and ramdisk_volume_id. The
> benefit of this approach is that the volume can live independent of the
> existence of the original images. Even the images get lost accidentally,
> the volumes are still sufficient to boot an instance, because all the
> information have been copied to Cinder part.
>
> I am trying to looking for the "another way to prepare your bootable
> volume" as you mentioned and asking for the suggestions.
> And I think the second approach could be one way. Do you think it is a
> good approach?
>
> Best wishes,
> Vincent Hou (侯胜博)
>
> Staff Software Engineer, Open Standards and Open Source Team, Emerging
> Technology Institute, IBM China Software Development Lab
>
> Tel: 86-10-82450778 Fax: 86-10-82453660
> Notes ID: Sheng Bo Hou/China/IBM at IBMCN    E-mail: sbhou at cn.ibm.com
> Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang
> West Road, Haidian District, Beijing, P.R.C.100193
> 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编：100193
>
>
>  *Mate Lakat <mate.lakat at citrix.com>*
>
> 2013/07/01 04:18
>  Please respond to
> OpenStack Development Mailing List <openstack-dev at lists.openstack.org>
>
>   To
> OpenStack Development Mailing List <openstack-dev at lists.openstack.org>,
> cc
> jsbryant at us.ibm.com, "Duncan Thomas <duncan.thomas at gmail.com> John
> Griffith" <duncan.thomas at gmail.comduncan.thomas>
> Subject
> Re: [openstack-dev] [cinder] Propose to add copying the reference images
> when creating a volume
>
>
>
>
> Hi,
>
> I just proposed a patch for the boot_from_volume_exercise.sh to get rid
> of --image. To be honest, I did not look at the various execution paths.
> My initial thought is that boot from volume means you boot from volume.
> If you only have a kernel + ramdisk image, I simply assumed that you
> can't do it.
>
> I would not do any magic. Boot from volume should boot from volume. If
> you only have 3 part images, you need to find another way to prepare
> your bootable volume.
>
> btw, here is my change:
>
> https://review.openstack.org/34761
>
> Cheers,
> Mate
>
> On Mon, Jul 01, 2013 at 01:25:23AM -0400, Sheng Bo Hou wrote:
> > Hi Cinder folks,
> >
> > I am currently fixing the bugs related to booting the instance from the
> > volume. I found there are bugs both in Nova and
> > Cinder.
> >
> > Cinder: https://bugs.launchpad.net/cinder/+bug/1159824
> > Nova: https://bugs.launchpad.net/nova/+bug/1191069
> >
> > For the volumes created from the image, I propose to copy the reference
> > image during the creation of
> > the main image. For example, an image may refer to a kernel image and a
> > ramdisk image. When we create a volume
> > from this image, we only copied this one to the volume. The kernel and
> > ramdisk images are still in glance, and
> > the volume still refers to the kernel and ramdisk images.
> >
> > I think if an image has other reference images, the reference images
> also
> > need to be copied to the volumes(kernel volume and ramdisk volume),
> > and then set the volume referring to the kernel volume and the ramdisk
> > volume. This feature will make booting from
> > a volume completely independent of the existence of the glance image.
> >
> > Do you think we can firstly add this feature to cinder? Do folks have
> any
> > comments on it?
> >
> > Thanks.
> >
> > Best wishes,
> > Vincent Hou (侯胜博)
> >
> > Staff Software Engineer, Open Standards and Open Source Team, Emerging
> > Technology Institute, IBM China Software Development Lab
> >
> > Tel: 86-10-82450778 Fax: 86-10-82453660
> > Notes ID: Sheng Bo Hou/China/IBM at IBMCN    E-mail: sbhou at cn.ibm.com
> > Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang
> > West Road, Haidian District, Beijing, P.R.C.100193
> > 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编：100193
>
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
> --
> Mate Lakat
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
While I like the idea of having Nova inspect the image data on the volume
and doing what it needs, we don't expose that level of info to Nova on boot
currently and I think there would be a good deal of overhead just for these
specific cases.

The option of multiple cinder volumes for boot from volume I don't quite
follow how that would work and it seems like it would be confusing to a
user.  Would you just automatically create the other two volumes when a
user requested 'create --image-id 123' if 123 was one of the image types
requiring the additional volumes? The other question is how would that be
interpreted and put back together on the nova side?

Anyway, it seems like the suggestion made by Vish is the best place to
start.  We can build off of that if there's feed-back that warrants it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20130702/4dab90f3/attachment.html>