Open Stack

On 06/05/2013 03:35 PM, Dan Prince wrote:
>
>
> ----- Original Message -----
>> From: "Sean Dague" <sean at dague.net>
>> To: openstack-dev at lists.openstack.org
>> Sent: Wednesday, June 5, 2013 3:15:57 PM
>> Subject: Re: [openstack-dev] Nova RPC regression
>>
>> On 06/05/2013 02:25 PM, Dan Prince wrote:
>>>>
>>>> Just to follow up.  This was caused by a change triggering a limitation
>>>> in qpid.  This is being tracked at
>>>> https://bugs.launchpad.net/nova/+bug/1175808.  The fix which has already
>>>> merged into oslo is now proposed against Nova.
>>>
>>> Thanks Laski.
>>>
>>> To my knowledge we have nothing gating on qpid. Given it is the prefered
>>> deployment option on some distributions (RHEL for example) it seems
>>> important so perhaps we should?
>>>
>>> SmokeStack uses qpid for some configurations but it isn't a gate so this
>>> slipped through the cracks.
>>
>> Did SmokeStack -1 this on checks?
>>
>> I've seen people pretty respectful of SmokeStack -1s on the check side,
>> so as long as it was getting back with results within an hour I think it
>> would cover the case fine.
>
> Yes. I knew about this ahead of time. I caught patchset 21. Patchset 22 came in while I was out at dinner. Perhaps I need to be more diligent about highlighting these sorts of failures... but as a non-gate SmokeStack is only so scalable. In general I don't mind this sort of thing (I'm used to it), and we can always do a quick revert... until we can't when the patch is like 1500 lines long :(.

Wait, I'm confused. SmokeStack doesn't report back by itself?

> I'm happy enough with the turn around here... just wondering if a slight config change to our upstream gating system might help out here. Right now we already spin up two devstack-gate instances for each zul run (one with quantum, one without). Why not take one these and spin it with a Fedora 18 Qpid setup? Seems like a simple change and we'd have this covered without too much issue.

I don't know that anyone's specifically opposed to that. The work just 
needs to be done. I do think there was concern from the CI team of 
chasing ephemeral distros. The roll back from 18 - 9 months of support 
on Ubuntu releases has meant our devstack nodes are going to hang out on 
LTSes. Fedora support windows are still < 1 year?

I do however wonder if on the RPC subsystem it's something that could be 
better accomplished through isolating via a unit test like approach run 
against real rpc daemons. There really is a lot that's untested in the 
RPC layer that would be good to get exercised better.

	-Sean

-- 
Sean Dague
http://dague.net