973900 - Please schedule desktop jit-tests on all trunk trees and make them ride the trains

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Dan Minor [:dminor]]

It looks like all of the consistent oranges for the desktop jit-tests on Cedar are now fixed and we're down to intermittents.

It is difficult to tell whether or not we hit more intermittents on the test machines than we do on the build machines while running on Cedar, due to the low volume of pushes.

I think we should turn these on for desktop builds on trunk trees, hidden for now, and see how things look.

We might need to resolve Bug 969986 first.

Comment 1

[Phil Ringnalda (:philor)]

And three other unfiled blockers, besides getting all Linux to run on try and include them in trychooser:

* decide whether to run them on all Mac and Windows platforms (probably not)
* some parallel permaorange on 10.6 opt and WinXP opt+debug
* the Linux slaves are so brutally slow that for linux64 asan and debug, this'll tie with the very slowest of browser-chrome runs for being the long pole for completing tests on a push

Comment 2

[Kim Moir [:kmoir] ET]

dminor:  Can you address philor's questions in comment #1?  Should I indeed enable them on all trunk trees and ride the trains for all platforms or are there some platforms where they should be excluded for now?

Comment 3

[Dan Minor [:dminor]]

I'm not sure if I'm the right person to decide this. I think the minimum would be to cover platforms that currently run them as part of make check (Linux, Windows XP, Mac OS X 10.8).

I have seen different results on Windows 7 and 8 than on WinXP on Cedar, but I'm not sure if we actually need that additional test coverage.

Ted, Terrence, any preferences here?

Comment 4

[(not currently active) Ted Mielczarek]

Let's schedule them on all platforms for now. If we decide we don't need them we can always selectively disable them later.

Comment 5

[(not currently active) Ted Mielczarek]

(although honestly I'd take philor's advice over mine)

Comment 6

[Terrence Cole [:terrence]]

Please do turn these on everywhere. I expect that this will expose real issues that are currently hidden because we are not already running them on all of our tier-1 platforms. We will fix anything that crops up.

Comment 7

[Kim Moir [:kmoir] ET]

Attachment: bug973900.patch

Will attach builder diffs too

Comment 8

[Kim Moir [:kmoir] ET]

Attachment: mac builder diff

Comment 9

[Kim Moir [:kmoir] ET]

Attachment: linux builder diff

Comment 10

[Kim Moir [:kmoir] ET]

Attachment: windows builder diff

Comment 11

[Rail Aliiev [:rail]]

Comment on attachment 8400152 [details] [diff] [review]
bug973900.patch

Review of attachment 8400152 [details] [diff] [review]:
-----------------------------------------------------------------

lgtm

::: mozilla-tests/config.py
@@ +1927,5 @@
> +                    if 'fedora' in slave_platform:
> +                        continue  # Don't use rev3 mini's with this stuff
> +                    BRANCHES[name]['platforms'][platform][slave_platform]['opt_unittest_suites'] += jittests[:]
> +                    BRANCHES[name]['platforms'][platform][slave_platform]['debug_unittest_suites'] += jittests[:]
> +     

A nit: can you remove the trailing spaces

Comment 12

[John Hopkins (:jhopkins)]

in production.

Comment 13

[Kim Moir [:kmoir] ET]

verified on tbpl