Open Stack

On 9 October 2015 at 01:10, Ihar Hrachyshka <ihrachys at redhat.com> wrote:
> Hi all,
>
> just a heads up that today fixtures 1.4.0 broke neutron py34 gate [1] and we needed to patch some logging code to overcome it [2]. The failures were triggered by a patch that started to raise logging exceptions for incorrect format strings (which is fine), but it also started to raise exceptions from stdlib logging code, and apparently till 3.5 it had a bug for the case when someone uses LOG.exception() in context where no exception was actually raised.
>
> More details about why I think the issue is in python interpreter are available in [2] commit message.
>
> I agree that using LOG.exception in such context is wrong, but still I wanted to notify other about potential issue, and a way to fix it.
>
> [1]: https://launchpad.net/bugs/1504053

Yeah - so the situation was that Ironic noticed they had bad logging
happen in prod and it wasn't caught in test. Tracking that down
pointed at the way logging eats all errors, and a patch from John
Villalovos to change fixtures to expose those errors.

Nova has had a local thing to detect bad strings for a while, but this
was a systemic fix - we're sorry about the firedrill :/.

The constraints stuff for unit tests, which we're in the last stages
of poc - basically we need the Neutron patches for constraints enabled
docs and flake8 runs, and then we can enable it in project-config...
at which point we can point at that as the template and encourage
wider adoption. Constraints jobs would have allowed back-pressure on
picking up the new release (e.g. if we had neutron unit test jobs
voting on requirements changes).

-Rob


-- 
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Converged Cloud