445622 - many builds on moz2 unit test tinderboxes are clobbers

Status: RESOLVED FIXED

(Release Engineering :: General, defect, P3)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I'm looking at the cycle times on "MacOSX Darwin 9.2.2 mozilla-central qm-moz2mini01 dep unit test" and many of them seem unnecessarily long because the objdir seems to be getting clobbered.  For example, the 07:34:19 cycle today should have been quick, but it looks like the objdir got clobbered.  (The 06:53:09 cycle was a depend build.)  The only new changeset in it was http://hg.mozilla.org/mozilla-central/index.cgi/rev/9e360506ae23 which should not have triggered any huge rebuild.

I think were were doing mostly/only depend builds back on http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox3.0

This bug could be similar to bug 433756, which was a bug with the build tinderboxes.

Comment 1

[[e-mail] (:bhearsum)]

There's a patch in our tree that hasn't landed in upstream code yet that fixes a bug where HgPoller triggered builds will always clobber...I wonder if the unit test slaves are using that code..Lukas or Rob, did the insntalled-Buildbot on the slaves change recently?

Comment 2

[Rob Campbell [:rc] (:robcee)]

not to my knowledge though lukas has the more recent picture.

Comment 3

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

In triage meeting today, Lukas confirmed she did not update/change any of the HgPoller code for unittests. 

Ben: can you point us to where this bug (or bugfix) is checked in, so we can compare and verify if unittests are running with this fix or not? 

Comment 4

[[e-mail] (:bhearsum)]

buildbot/slave/commands.py should be version 1.4. That's when the fix was checked in. A 'cvs stat' on commands.py will tell you the version.

http://bonsai.mozilla.org/cvslog.cgi?file=mozilla/tools/buildbot/buildbot/slave/commands.py&rev=HEAD&mark=1.4

Comment 5

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

I happened to (partly) check on this today. The two linux machines for mozilla-central have copies of slave/commands.py containing "has already been run" (from rev1.4 of that file); that's moz2-linux-slave07,08. The two mac machines don't contain that phrase; that's moz2-darwin8-slave01,02. The windows machines were busy so I couldn't look in on them.

Comment 6

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

The mac slaves got fixed in bug 464662, we just need to check windows.

Comment 7

[Lukas Blakk [:lsblakk] use ?needinfo]

Does this still apply now that we've consolidated? Nick?

Comment 8

[[e-mail] (:bhearsum)]

Haven't had this in forever.