On 28/09/12 09:28 PM, Boris Zbarsky wrote:
On 9/28/12 8:32 PM, Chris Pearce wrote:
This is indeed unfortunate. However I'd prefer to add more capacity to
our test infrastructure, rather than discourage developers from properly
testing before landing.
I think the concern is the definition of "properly".
If Try is hogging resources needed by Inbound, we should lower the
priority of Try.
Try is hogging resources needed by _Try_. The last try push I did,
asking for one test on one platform, waited for over a day before the
test even started running.
This is why we've been pushing hard on turning off tests that are
perma-red/orange. The test machines are shared between all branches. If
you reduce the test load from inbound and other branches, then more
capacity is available for try. This effort has made some progress. In
August 1,410 XP test jobs on try waited longer than 24 hours to start.
In September (so far!) 0 jobs have waited longer than 24 hours.
In a world of infinite compute resources, we should run all the tests.
In our world of finite resources, we have to make sure we're getting
good information for the tests we run.
One proposal that's been made elsewhere
(https://bugzilla.mozilla.org/show_bug.cgi?id=791385) is to have a soft
limit of one active push per developer on try. If you try and push a 2nd
time before your previous jobs are all finished, you will be asked to
cancel your previous jobs. There would be some kind of manual override
that would allow you to push additional patches.
Take a look at http://people.mozilla.org/~catlee/try_pushers.html. This
is a report of people who pushed to try within 60 minutes of their
previous push since September 1. In some cases patches are clearly
unrelated, and in other cases people have cancelled their previous job
(woohoo!). There are many other cases where it appears that people are
pushing very similar code to try without having cancelled their previous
jobs.
http://people.mozilla.org/~catlee/highscores/highscores.html is a report
of where our time on Try is going.
Cheers,
Chris
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
