Open Stack

Thanks for sending out this message Malini.

I'm really pleased that the "image import" mechanism we've been working on
in Glance for a while is going to be helpful for supporting this kind of
use case.

The problem that I see is one of messaging. If we tell end users that
"OpenStack can import and run OVAs" I think we're probably setting
ourselves up for a serious problem with expectations. Since an OVA is *not*
an image, and actually could be much broader in scope or more constrained,
I'm worried that this import will fail for most users most of the time.
This just creates a negative impression of our cloud, and may cause a
significant support headache for some of our deployers.

The plan I propose to respond to this challenge is as follows:

1) develop the initial OVA image import out of tree
    - the basic functionality is just to grab the root disk out of the ova
and to set image properties based on some of the ovf metadata
2) assess what the median level of OVA complexity is out there in the wild
among OVA users
3) make sufficient progress with artifacts to ensure we can cover the
median level of OVA complexity in an OpenStack accessible way
    - openstack accessible to me means there probably has to be qemu-image
/ libvirt / heat support for a given OVA concept
4) Bring OVA import into the main tree as part of the "General Import" [1]
operation once that artifact progress has been made

However, I'm very interested to know if there are some folks more embedded
with operators and deployers who can reassure me that this OVA messaging
problem can be dealt with another way.

Thanks!


[1] As a reminder, the "General Import" item on our hazy future backlog is
different from "Image Import" in the following way. For an image import,
you are explicitly trying to create an image. For the general import, you
show up to the cloud with some information and just ask for it to be
imported, the import task itself will inspect the data you provide to
determine what, if anything, can be created for it. This works well for
OVAs because we may want to produce a disk image, a block device mapping
artifact, or even up to the level of a heat template.


On Fri, Jul 25, 2014 at 7:08 PM, Bhandaru, Malini K <
malini.k.bhandaru at intel.com> wrote:

> Hello Everyone!
>
> We were discussing the following blueprint in Glance:
> Enhanced-Platform-Awareness-OVF-Meta-Data-Import :
> https://review.openstack.org/#/c/104904/
>
> The OVA format is very rich and the proposal here in its first incarnation
> is to essentially
> Untar the ova package, andimport the first disk image therein and parse
> the ovf file and attach meta data to the disk image.
> There is a nova effort  in a similar vein that supports OVA, limiting its
> availability to the VMWare hypervisor. Our efforts will combine.
>
> The issue that is raised is how many openstack users and OpenStack cloud
> providers tackle OVA data with multiple disk images, using them as an
> application.
> Do your users using OVA with content other than 1 disk image + OVF?
> That is does it have other files that are used? Do any of you use OVAs
> with snapshot chains?
> Would this solution path break your system, result in unhappy users?
>
>
> If the solution will at least address 50% of the use cases, a low bar, and
> ease deploying NFV applications, this would be worthy.
> If so, how would we message around this so as not to imply that OpenStack
> supports OVA in its full glory?
>
> Down the road the Artefacts blueprint will provide a place holder for OVA.
> Perhaps even the OVA format may be transformed into a Heat template to work
> in OpenStack.
>
> Please do prov ide us your feedback.
> Regards
> Malini
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140726/9300c91f/attachment.html>