1034674 - use a property other than speed to differentiate between slaves in slavealloc

Status: RESOLVED WONTFIX

(Release Engineering :: General, defect)

Description

[Kim Moir [:kmoir] ET]

In bug 1034055, we needed to specify a different instance type for 2.3 reftests etc al to run in AWS.  Some 2.3 tests run on c3.xlarge, other run on the smaller type.  The meant we had to differentiate the slaves in slavealloc via speed which rail mentioned is not optimal.  So he asked me to open a bug and he would look at it.

See https://bugzilla.mozilla.org/show_bug.cgi?id=1034055#c15 for details

Comment 1

[Rail Aliiev [:rail]]

Attachment: use_moztype-tools.diff

this looks working "on my laptop". Still need to figure out how to migrate the DB.

Comment 2

[Dustin J. Mitchell [:dustin] (he/him)]

Comment on attachment 8457458 [details] [diff] [review]
use_moztype-tools.diff

I'm always wary of naming something with a 'moz' prefix.  What's a "moztype"?  It sounds like you're adding this because the name "speed" doesn't seem right to you, but "moztype" doesn't mean anything more.

When we invented "speed", the idea was to differentiate system types -- I think at the time we had hardware and vmware builders, and the latter were a lot slower.  We didn't want one master to get saddled with all the slow builders.

So it seems like speed *is* a pretty good match to instance type?  If not, then at least "moztype" didn't help me figure out what you meant :)

Comment 3

[Rail Aliiev [:rail]]

yeah, I thought about using speed, but the idea i not specifying AWS instance types, instead, usbing so called moz-type which may correspond to multiple AWS instance types (like in http://hg.mozilla.org/build/cloud-tools/file/6939b4d5aaf7/configs/watch_pending.cfg#l50).

I have no idea for the name, but will keep it in mind.

Comment 4

[Dustin J. Mitchell [:dustin] (he/him)]

You're referring to 'build-linux64'?  Isn't that redundant to trustlevel, bitlength, and os?

Comment 5

[Rail Aliiev [:rail]]

for build-linux64 vs try-linux64 it is enough, but not for tst-linux64 vs tst-emulator64, see http://hg.mozilla.org/build/cloud-tools/file/6939b4d5aaf7/cloudtools/slavealloc.py#l45

Comment 6

[Dustin J. Mitchell [:dustin] (he/him)]

maybe that's just a different purpose, aside from build and test?  At the end of the day, it doesn't really matter, as pool is just determined by distinct tuples of attributes.  Adding a new moztype attribute, using purpose, speed, whatever.. same effect.  It's just what makes the most sense, I think

Comment 7

[Dustin J. Mitchell [:dustin] (he/him)]

*just a matter of choosing what makes the most sense

Comment 8

[Rail Aliiev [:rail]]

Not working on this actively.

Comment 9

[Chris Cooper [:coop] (he/him)]

WONTFIX in favor of bug 1135818.