Open Stack

On 06/18/2013 03:32 PM, Monty Taylor wrote:
>
>
> On 06/18/2013 03:14 PM, David Ripton wrote:
>> On 06/18/2013 12:43 PM, Martina Kollarova wrote:
>>> Jenkins keeps running all the tests, even if the basic pep8 test fails,
>>> and runs all of the (very slow) Tempest Quantum tests, even though
>>> almost all of them are failing.
>>>
>>> I propose that it should fail and stop all of the other tests once there
>>> is a failure in a voting test. For non-voting tests, it should stop only
>>> itself, not the others.
>>>
>>> This would decrease the feedback loop and we wouldn't have to wait for
>>> the non-voting Quantum tests to see that they failed as always.
>>
>> -1
>>
>> In addition to the other objections, we currently get a lot of false
>> positives (fail, retry, fail, retry, succeed), and it would be harder to
>> debug these problems if the output was truncated differently each time.
>>
>> Is anyone working on fixing the perma-failing Quantum test?  When the
>> Postgres test was perma-failing, one of the infrastructure folks gave us
>> an ultimatum that if nobody fixed it soon, it would be disabled.  (Happy
>> ending: Mauro fixed it before it got disabled.)
>
> That was brought up a little while ago, but we had already spent so much
> effort to get it working in the first place, none of us had the heart to
> put in such an ultimatum. But seriously- it might be time for an
> all-hands-on-deck dogpile to figure out what's up the the quantum gate.

The biggest cause of the Quantum vs. Full Tempest runs is that a lot of 
the network api's in nova currently don't do translation of errors. So 
under nova-network certain data validation and error codes are returned, 
when quantum is used others are returned.

This is a nova-api, so it needs to be consistent regardless of backend 
(i.e. we don't return different API responses on different databases).

Issues like this one - https://bugs.launchpad.net/nova/+bug/1160309

Jordan Pittier has been working on some of these issues (he's the only 
one I've seen working them from a Tempest / nova side), and got to the 
crux of the problem. It could use more hands though to organize the rest 
of those and get them banged out.

I'm sure there are other issues once we get past this class. But that 
would go a long way.

	-Sean

-- 
Sean Dague
http://dague.net