On Fri, Dec 6, 2013 at 3:01 PM, Antoine Musso <amusso(a)wikimedia.org> wrote:
Is there any reason to use the Barack Obama article for testing? The
reason that article has grown popular is that I used it to stress test
the Cite extension something like 3 or 4 years ago. At that time that
was an article with probably the most {{cite}} and it highlighted we
needed a better template system (short story: we have LUA now).
One of the big reasons we've used the Obama article to test against is
because it's long and complex. Historically, we've had issues that would
only surface in particular cases which could often be replicated on highly
complex articles, but not necessarily on really simple articles. Is there
another article that we could use that meets the criteria of complex and
long, but will not cause the 500 errors like Obama article? If so, then
let's try that instead. I suspect we do not actually know the answer to
this question and worry that we can try finding other workarounds for this
problem that will ultimately pile up and cause even more headaches.
The one thing I do know is that we need to do something, and do it soon. If
our tests keep failing because of an architectural issue rather an issue in
our code, the tests lose their utility - we stop paying attention to
failures when they are false positives (or false positive-ish).
Given what I've heard form Antoine and folks in ops, this issue is not
likely to be resolved in the near term. Who is the 'owner' of betalabs
anyway - ops? platform? QA?
*Chris*, what do you think we should do to resolve this asap?
To conclude, I will attempt to investigate that issue
on Monday.
*Antoine*, did you have any luck?
A possibility would be to have a sleep.php that would
take a sleeping time
value in parameter. Can then attempt to find out whether the timeout is
respected, aka query:
sleep?60
sleep?180
..
I think we should either use another article or really fix the issue. Yet
another hack feel like it will just introduce further fragility and
possibly cause issues with test validity.
--
Arthur Richards
Software Engineer, Mobile
[[User:Awjrichards]]
IRC: awjr
+1-415-839-6885 x6687