On 2013-04-03 10:59 PM, Jeff Hammel wrote:
So I'm not sure I understand:
> 1. This will incur a significant increase in our infra resource usage
since all of these patches have to do a full try run. We simply cannot
afford that in today's world where we're struggling against wait times
and infrastructure overload.
If we push to e.g. inbound, we still run all the tests + builds. My
understanding was the whole point was to avoid the double run (running
all/most tests/builds on try and then landing on e.g. inbound where
they're all run again). Is the concern the overhead pushing to try?
autoland or no, that is a problem indeed. If not, I don't really
understand the concern here.
Autoland needs to first push to try, and then possibly push to
inbound/central. Without autoland, lots of people avoid using the try
server when they make a judgement call on a patch being safe enough.
With Autoland, that will stop, hence there will be an infra load increase.
2., all of our damn intermittent, is indeed a concern :( And of course
not just for this reason: tons of human and often machine time is lost
to them.
Sure.
I can imagine this being addressed for autoland, but will
leave ctalbert or others to comment as to plans there since I'm sure
they've thought of everything I could propose right now (and then some).
Obviously the best solution is to kill all the orange, but ... yeah
Back when I was fighting this war with the help of others, we got the
orange factor down to about 1 orange per push, which was fantastic. Our
experience has shown that this is definitely doable, it just needs
somebody to own the effort and work on it.
Cheers,
Ehsan
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
