Open Stack

On 05/22/2013 08:38 AM, Sean Dague wrote:
> On 05/22/2013 07:02 AM, Angus Salkeld wrote:
>> On 21/05/13 23:50 -0300, Sandy Walsh wrote:
>>> Howdy ....
>>>
>>> Given the timezone changes on this project I thought I'd send this to
>>> the list in hopes someone will see it overnight.
>>>
>>> I was close to getting my branch landed when a rebase brought the testr
>>> changes in ... https://review.openstack.org/#/c/29047/
>>>
>>> Now, my tests are failing in an odd way (and they used to pass).
>>>
>>> The errors are all of this form:
>>> http://paste.openstack.org/show/37560/
>>>
>>> I've investigated and it seemed to be related to sqlalchemy and
>>> eventlet:
>>>
>>> http://stackoverflow.com/questions/9525220/sqlalchemy-raises-none-causes-typeerror
>>>
>>>
>>>
>>>
>>> and I thought that perhaps the testr patch is doing something funky with
>>> eventlet that isn't making sqlalchemy happy.
>>>
>>> But when I apply the 'raise e' patch as suggested, it shows the
>>> underlying issue as being:
>>> http://paste.openstack.org/show/37563/
>>>
>>> Which might suggest the migrations aren't being set up correctly?
>>> (odd as the other sqlalchemy tests are passing)
>>>
>>> I'm sort of stumped here ... suggestions?
>>
>> Hi Sandy,
>>
>> I ran tox and testr run ~ 50 times on my laptop (with your patch).
>> There was no issue with tox but for some reason I got some
>> errors using testr run (I don't think they are related to
>> what you are seeing).
>>
>> https://review.openstack.org/30077
>> https://review.openstack.org/30078
>>
>> Are you using tox or "testr run" directly?
>> If you are using "testr run" try
>> testr run --concurrency=1
> 
> Actually, I'd really recommend doing:
> 
> testr run --concurrency=32
> 
> In the conversion of any unit tests you'll find people were very bad
> about keeping the tests issolated from each other. Driving the
> concurrency up will expose those breaks early, otherwise they'll show up
> at some statistical rate in the future when tests are reordered be
> subsequent runs.
> 
> Those fails are real bugs in our tests. If nose changed hash ordering
> they'd show up there as well. Best to drive out the test bugs now.

Thanks guys,

I'd be happy to get it running single process for now. Then, I agree, we
should be running randomized and in parallel.

Baby steps for me.

Angus, I'll try running testr directly, I'd been running it via tox.


-S


> 
>     -Sean
>