466248 - Remove 'chrome' |--leak-threshold=8| from 'MacOSX Darwin 9.2.2 mozilla-central moz2-darwin8-slave0? dep unit test'

Status: VERIFIED FIXED

(Release Engineering :: General, defect, P2)

Description

[Serge Gautherie (:sgautherie)]

There are 2 issues:
1) The parameter is repeated "near-endlessly" on the command line:
   the later is 7k chars long in the log.
2) The |8| value override doesn't seem to be needed anymore:
   the |0| default will be just fine from now on.

{
python runtests.py --chrome --autorun --console-level=INFO --close-when-done --leak-threshold=8 [...] --leak-threshold=8
...
TEST-PASS | runtests-leaks | no leaks detected! (threshold set at 8 bytes)
...
TinderboxPrint: chrome<br/>1551/0/28
}

The Unix and Windows boxes don't have this override/bug.

(NB: This is not related to the MacOSX box renaming which was done today.)

Comment 1

[(not currently active) Ted Mielczarek]

I think this is in the buildbot configs, no? If so, this should be in mozilla.org : Release Engineering.

Comment 2

[Serge Gautherie (:sgautherie)]

I was just in the middle of checking the waterfall history ;-)

This was fixed (but I don't know how) between

http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1227613534.1227617843.13467.gz&fulltext=1
MacOSX Darwin 9.2.2 mozilla-central moz2-darwin8-slave01 dep unit test on 2008/11/25 03:45:34
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1227613534.1227618717.416.gz&fulltext=1
MacOSX Darwin 9.2.2 mozilla-central moz2-darwin8-slave02 dep unit test on 2008/11/25 03:45:34

and

http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1227622104.1227626426.20253.gz&fulltext=1
MacOSX Darwin 9.2.2 mozilla-central moz2-darwin8-slave01 dep unit test on 2008/11/25 06:08:24
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1227624129.1227629319.26856.gz&fulltext=1
MacOSX Darwin 9.2.2 mozilla-central moz2-darwin8-slave02 dep unit test on 2008/11/25 06:42:09

Comment 3

[Serge Gautherie (:sgautherie)]

This bug is back somehow, and now on all 3 platforms:

http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1230878897.1230882158.15568.gz&fulltext=1
Linux mozilla-central unit test on 2009/01/01 22:48:17
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1230879115.1230881482.13687.gz&fulltext=1
OS X 10.5.2 mozilla-central unit test on 2009/01/01 22:51:55
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1230273421.1230280193.29546.gz&fulltext=1
WINNT 5.2 mozilla-central unit test on 2008/12/25 22:37:01

Comment 4

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

[1] is supposedly where it comes in on Mac, but it's not obvious why it would be happening on linux and win32 too. Clearly there is a bug with MozillaMochichrome in buildbotcustom/unittest/steps.py [2] (probably also in MozillaMochitest & MozillaBrowserChromeTest, where the same code structure is used but no threshold is set). Could the __init__ be being called recursively ? There are 73 "--leak-threshold=8" in the logs.

[1] http://mxr.mozilla.org/seamonkey/source/tools/buildbotcustom/process/factory.py#1848
[2] http://mxr.mozilla.org/seamonkey/source/tools/buildbotcustom/unittest/steps.py#365

Comment 5

[Serge Gautherie (:sgautherie)]

Attachment: Remove (obsolete) MacOSX "--leak-threshold=8" [Checkin: Comment 8]

This line was moved to factory.py from unittest_master.py by bug 464692.
I didn't look further to find out why it was added there in the first place.

Anyway, there is currently no leak and this setting should not be hardcoded there.

Comment 6

[Serge Gautherie (:sgautherie)]

(In reply to comment #4)
> [1] is supposedly where it comes in on Mac, but it's not obvious why it would

Thanks for the pointer.

> be happening on linux and win32 too. Clearly there is a bug with

Could it be since the 3*2 individual boxes were replaced by "buildbots" ?
Questions would be:
*What fixed it on 2008/11/25 ?
*When/Why did it reappeared ?

> MozillaMochichrome in buildbotcustom/unittest/steps.py [2] {...}
> Could the __init__ be being called recursively ?

This would be interesting to find out !

Comment 7

[(not currently active) Ted Mielczarek]

Comment on attachment 355093 [details] [diff] [review]
Remove (obsolete) MacOSX "--leak-threshold=8"
[Checkin: Comment 8]

I can never remember what bits of our buildbot configs are in CVS vs. Hg anymore.

Comment 8

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

Comment on attachment 355093 [details] [diff] [review]
Remove (obsolete) MacOSX "--leak-threshold=8"
[Checkin: Comment 8]

Checking in factory.py;
/cvsroot/mozilla/tools/buildbotcustom/process/factory.py,v  <--  factory.py
new revision: 1.68; previous revision: 1.67
done

Shouldn't need downtime to reconfig for this, but if we have some this week we might as well tag along.

Comment 9

[Serge Gautherie (:sgautherie)]

(atm, Firefox and Firefox3.1 boxes have this bug.)

Comment 10

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

(In reply to comment #9)
> (atm, Firefox and Firefox3.1 boxes have this bug.)

Yes, which is why we need downtime to push this change. In case the reconfig doesn't go smoothly, we don't want to tank regular Firefox or 3.1 builds.

Comment 11

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

Hrmmm, an unrelated reconfig just happened and we lost l10n builds, but this change should be live now.

Comment 12

[Serge Gautherie (:sgautherie)]

V.Fixed, on current builds.

***

Do you know why this setting was applied multiple times and why it affected Linux/Windows now ?
Or should I open a separate bug to investigate that ?

Comment 13

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

(In reply to comment #12)
> Do you know why this setting was applied multiple times and why it affected
> Linux/Windows now ?
> Or should I open a separate bug to investigate that ?

Didn't look into it. A separate bug would be best.

Comment 14

[Serge Gautherie (:sgautherie)]

(In reply to comment #13)

I filed bug 473542.