[QA] Recently failing browser tests

[Greg Grossmeier]

(let's use mailing lists instead of manually cc'ing people, please)

Timo/Jon: Do you have tasks in phabricator that show what you've tried
doing/debugging already? Would hate to re-do it all.

Greg


<quote name="Timo Tijhof" date="2015-05-28" time="11:57:05 +0200">
> I've seen this and similar issues on beta cluster for almost a year now and am reasonably certain that there is no bug in ResourceLoader related to this. We might uncover an indirect cause in MediaWiki core later, but the immediate cause is most likely in Beta cluster. This because I've observed the relevant issues with non-ResourceLoader requests as well. And because we've not been able to reproduce this in production.
> 
> I don't know enough about Beta cluster or our http caching to know how to help. I've already exhausted possibly causes I can think of on my own – last month. However I'm happy to join a hangout session at some point to consult and help pinpoint the problem from a MediaWiki perspective, we should have a prod opsen (knowledgable about Varnish), and a beta devop present there as well. 
> 
> -- Timo
> 
> > On 28 May 2015, at 11:37, Jon Robson <jrobson at wikimedia.org> wrote:
> > 
> > Firstly apologises for all the browser test spam recently. This is an update on what is going on and to give you confidence it is being taken seriously.
> > 
> > I've managed to trace the issue to some kind of caching problem in mediawiki core but I'm a bit out of my depth trying to work it out.
> > 
> > Basically under certain circumstances our code is broken and this is why the tests are failing.
> > 
> > The issue is that we are serving old JavaScript along with new JavaScript and these files are naturally incompatible causing the errors as they are trying to access old frontend APIs that no longer exist.
> > 
> > How ResourceLoader deals with caching changed last month and I suspect there is a bug in that code. I suspect Timo would be able to help me chase down the route cause further.
> > 
> > In the meantime I'm not sure if there is a way to stop the email spam till this gets resolved. One idea I had was that when a test build fails it would be great if the email alerts stopped and developers become responsible for restarting them. Greg/Antoine/Dan is this possible? There's little benefit in continuing to run them every day and emailing about faikures until this test is fixed.
> > 
> > Any help on https://phabricator.wikimedia.org/T100542 much appreciated. I'm out of my depth as this issue seems to fall outside the good ship readership and more in the ResourceLoader and quality assurance sea.

-- 
| Greg Grossmeier            GPG: B2FA 27B1 F7EB D327 6B8E |
| identi.ca: @greg                A18D 1138 8E47 FAC8 1C7D |