ACA Side Topic - Proposal: templates.aeolusproject.org

Scott Seago sseago at redhat.com
Fri Jun 8 18:29:10 UTC 2012


On 06/08/2012 01:53 PM, Michael Orazi wrote:
>
> ----- Original Message -----
>> On 06/08/2012 09:24 AM, Hugh O. Brock wrote:
>>> Yeah. We have to do this -- it's a blindingly obvious way to
>>> generate
>>> community interest and a real benefit for users.
>>>
>>> I've asked Francesco to head up the effort of getting the thing
>>> built. We can host it on the Redmine machine for now -- if it gets
>>> bogged down with too many users, well, that goes in the category of
>>> problems we'd like to have.
>>>
>>> Andy Fitz, can you help with the design? I'll then appoint someone
>>> to
>>> help Francesco get it coded up.
>>>
>>> Good stuff...
>>>
>>> --Hugh
>>>
>> I agree with everyone else here -- it is an excellent idea. It will
>> require one change to the deployable definition (and possibly a
>> related
>> change to the image model) that we've discussed in the past -- this
>> may
>> be the actual motivator to get it done. Currently deployables
>> reference
>> image by UUID, which makes them completely un-shareable. For this to
>> work we need to change the deployable definition to allow referencing
>> images by name. Since we're sharing deployables with templates, it's
>> actually template name we'd want to use.
>>
> It might be quite interesting to be able to refer to template via a url (real or not for now).
>
Sure. This is one case where the "portable" format may differ from what 
we end up with in the database once imported. With a "portable" 
deployable file imported into conductor, if the template is referenced 
by URL we'll need to pull the template into conductor at the same time, 
since we need it locally to build. Once that's done, the in-database 
representation of the deployable would then point to this imported 
template (by name?) rather than the URL alone, although we may also want 
to keep the URL around so we know where we got it -- to eventually give 
alerts if something is out-of-date, etc.

>> There are a bunch of implementation edge cases to work through here
>> --
>> since it's valid to use the same template more than once -- i.e. more
>> than one image that happen to use the same template, the template
>> name
>> will not uniquely identify an image. When you take a set of
>> templates+deployables that were shared and import/build/push the
>> templates and then import the deployables, you could end up with a
>> situation where a deployable identifies a template/image by a name
>> that
>> maps to more than one image in the system -- either because the same
>> template has been used more than once, or because there are multiple
>> templates that happen to have the same name. We'll have to come up
>> with
>> some way to resolve this, but the answer isn't obvious to me at this
>> time.
>>
>> Scott
>>
>>




More information about the aeolus-devel mailing list