806787 - Per-checkin pgo builds are missing pgo from the buildername on aurora, beta & esr10

Status: RESOLVED DUPLICATE of bug 700924

(Release Engineering :: General, defect)

Description

[Ed Morley [:emorley]]

Since at least 11th Sept (TBPL data purge means I can't go back further), the tests run on opt builds for {linux,linux64,Win7} are being reported as PGO.

eg:
https://tbpl.mozilla.org/php/getParsedLog.php?id=16576371&tree=Mozilla-Aurora


09:34:41 - nthomas: uh, non pgo build is doing a pgo sendchange
09:35:04 - nthomas: so if you look at 97c6c326435e (tip of aurora), the tests show up as linux pgo when there was no pgo build
09:40:43 - nthomas: it comes from http://hg.mozilla.org/build/buildbot-configs/file/default/mozilla-tests/config.py#l975, by way of http://hg.mozilla.org/build/buildbotcustom/file/default/process/factory.py#l2589

Comment 1

[Ed Morley [:emorley]]

09:42:07 - nthomas: and it is a PGO build, just doesn't say so in the name
09:42:16 - nthomas: good luck untangling that mess
09:43:02 - edmorley: nthomas: oh so the build is pgo? so the tests are reporting correctly, the builds are not?
09:43:07 - msucan [[e-mail]] entered the room.
09:43:17 - jfkthame [[e-mail]] entered the room.
09:43:19 - nthomas: seems that way to me
09:43:52 - nthomas: esr10 will be the same

Comment 2

[Nick Thomas [:nthomas] (UTC+12)]

Since the tests are present, just not on the same line, I'm reducing the severity here.

Comment 3

[Ed Morley [:emorley]]

I'm not sure I agree really.

We don't know exactly what we're testing (given the uncertainty in comment 1), and even if we were, we're only running half the tests (either just pgo or just opt).

Comment 4

[[e-mail] (:bhearsum)]

As far as I can tell, these branches are supposed to have per-checkin PGO:
https://github.com/mozilla/build-buildbot-configs/blob/master/mozilla/config.py#L1312
https://github.com/mozilla/build-buildbot-configs/blob/master/mozilla/config.py#L1370
https://github.com/mozilla/build-buildbot-configs/blob/master/mozilla/config.py#L1233

And they've been set that way for a long time (at least 6 months). Looking back on FTP confirms this:
https://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-aurora-linux-pgo/1348744810/ has a bunch of test logs
https://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-aurora-linux/1348779055/ has the build parts

And its sendchange is PGO:
python /builds/slave/m-aurora-lnx/tools/buildfarm/utils/retry.py -s 5 -t 1800 -r 5 --stdout-regexp 'change sent successfully' buildbot sendchange --master buildbot-master36.build.mozilla.org:9301 --username sendchange --branch mozilla-aurora-linux-pgo-talos --revision 6db116c397b1 --comments 'Bug 788436. In the mac plugin code if the event model changes we may need to add/remove our scroll position listener. r=josh a=akeybl' --property buildid:20120927135055 --property pgo_build:True --property builduid:9fa2ce4d20454715ad740f36e95a0940 http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-aurora-linux/1348779055/firefox-17.0a2.en-US.linux-i686.tar.bz2

Any chance this is a result of a TBPL change? I see https://hg.mozilla.org/users/mstange_themasta.com/tinderboxpushlog/rev/caa1ba87ec98 making lots of noteworthy changes.

Comment 5

[Ed Morley [:emorley]]

(In reply to Ben Hearsum [:bhearsum] from comment #4)
> As far as I can tell, these branches are supposed to have per-checkin PGO:

Oh, so we're not supposed to be doing opt builds, just pgo?

If so, then the issue in comment 0 is in fact intentional.

Comment 6

[[e-mail] (:bhearsum)]

I would say that the opt builds are supposed to be PGO, rather than saying we should be doing PGO instead of opt -- but yes, it's intentional.

Comment 7

[Armen [:armenzg]]

What bhearsum says is correct. I always advocated that the name of the builds should have been explicit that they are pgo.

Comment 8

[[e-mail] (:bhearsum)]

Removing "regression" as a keyboard, because it's always been this way...

Comment 9

[Phil Ringnalda (:philor)]

Been so long I can't even remember why I filed bug 700924 phrased the opposite way - I guess because the assumption used to be that if it was Linux or Windows, it was PGO.