[katello-devel] Organization deletion bug, orchestration, testing

[Ivan Necas]

----- Original Message -----
> On Fri, Dec 14, 2012 at 03:06:33AM -0500, Ivan Necas wrote:
> > With JRuby, reusing this tools might be feasible, right?
> 
> It does not depend on which platform are we running Katello on,
> integration with such a service is usually external - via an API like
> web-service, REST service or messaging. Engine is a standalone
> process
> usually.
> 
> I don't thing this is acceptable from the management point of view.
> But
> I think it is feasible to refactor this.
> 
> > We also need better testing for edge cases. E.g. what happens when
> > I delete
> > an organization and restart all the services in the mean time. How
> > can I recover?
> > Trying to solve this issues can point us the the weak points (which
> > will be probably
> > very similar to those described in this thread.)
> > 
> > I agree neither black box testing nor unit tests will help us much
> > here. Unit test because
> > of the isolation, black box because everything might seem working
> > at the moment and
> > knowledge of the code base can help finding the issue.
> > 
> > 
> > Stories:
> > 
> > As a integrator, I want to have the edge cases automated.
> > As a integrator, I want to provide standardized ways for recovering
> > from various non-standard situations.
> > 
> > When talking about keeping consistency between the systems, the
> > first obvious step, as Mirek pointed out, is to finally introduce
> > foreign keys
> > to keep the consistency in our own database. This should be #1
> > priority, if we don't want to spend the rest of our lives fixing
> > bugs like this
> > one.
> 
> Well, with decent integration approach you don't really need this -
> data
> changes should be one-way only. If any process changing our database
> gets stuck, you get notified and you can easily make it finish (e.g.
> after you restart services involved).
> 
> The main advantage of this approach is you (as a caller) are not
> responsible error recovery anymore. Then you don't need to focus so
> much
> on testing edge (error) cases.

And who is responsible for that then?

-- Ivan

> 
> But if we stick with our current "web server process rules them all"
> approach, then we would need that. Many more tests. I'd start with
> implementing integration oriented tests (that should come with PulpV2
> and minitest/webmock/facorygirl/vcr merge).
> 
> 
> --
> Later,
> 
>  Lukas "lzap" Zapletal
>  #katello #systemengine
> 
> _______________________________________________
> katello-devel mailing list
> katello-devel at redhat.com
> https://www.redhat.com/mailman/listinfo/katello-devel
>