622205 - Memory leak(s) in Tinderbox Valgrind runs

Status: RESOLVED DUPLICATE of bug 631811

(Core :: General, defect)

Description

[Nicholas Nethercote [inactive]]

From http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1293727008.1293727433.12328.gz&fulltext=1:

==29991== LEAK SUMMARY:
==29991==    definitely lost: 37,693 bytes in 1,641 blocks.
==29991==      possibly lost: 451,527 bytes in 264 blocks.
==29991==    still reachable: 1,953,516 bytes in 9,337 blocks.
==29991==         suppressed: 0 bytes in 0 blocks.

This should cause the Valgrind tests to go red, but it doesn't because of bug 620828 (which requires only a one-line fix).

This leak has been present for at least a week or two.

I've filed this under the "general" component because without bug 620828 there's no info on where the leak(s) are happening, unless someone manually runs that test under Valgrind.

Comment 1

[Nicholas Nethercote [inactive]]

This should block 2.0 without question.  Memory leaks are unacceptable.

Comment 2

[Julian Seward [:jseward]]

(In reply to comment #1)
> This should block 2.0 without question.  Memory leaks are unacceptable.

True; but it might not be that cut-n-dried.  Many of the leaks I see
on x64-linux builds are leaks in innumerable gnome and wotnot low level
libraries, that we can't do anything about.  For sure, we need
--leak-check=full to get anything useful out of this, though.

Comment 3

[Benjamin Smedberg]

njn, can you run these locally and figure out what's actually leaking. Is there a definite regression range? Please renominate when we know the actual severity of the problem.

Comment 4

[Nicholas Nethercote [inactive]]

Can someone (philor?) help me understand how to run the test myself?  AFAICT the command being run (from http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1293727008.1293727433.12328.gz&fulltext=1) is this:

args: ['/usr/bin/valgrind', '--error-exitcode=1', '--smc-check=all', '--gen-suppressions=all', '--suppressions=/builds/moz2_slave/valgrind-linux/build/scripts/scripts/valgrind/i686-redhat-linux-gnu.sup', '/builds/moz2_slave/valgrind-linux/build/objdir/dist/firefox/firefox-bin', '-no-remote', '-profile', '/builds/moz2_slave/valgrind-linux/build/objdir/_profile/pgo/pgoprofile/', 'http://localhost:8888/index.html']

I don't know what the contents of that index.html file are.  Also, getting my hands on that suppression file (/builds/moz2_slave/valgrind-linux/build/objdir/dist/firefox/firefox-bin) would be helpful, how do I do that?

Comment 5

[Phil Ringnalda (:philor)]

Easier to see what's being run from http://hg.mozilla.org/build/tools/file/d5dca516aa62/scripts/valgrind/valgrind.sh - build with --enable-valgrind and --disable-jemalloc, then from the objdir run python _profile/pgo/profileserver.py --debugger=valgrind --debugger-args="--error-exitcode=1 --smc-check=all --gen-suppressions=all --suppressions=i686-redhat-linux-gnu.sup" having first copied over http://hg.mozilla.org/build/tools/file/d5dca516aa62/scripts/valgrind/i686-redhat-linux-gnu.sup.

Well, first-first having installed the VM from ftp://ftp.mozilla.org/pub/mozilla/VMs/CentOS5-ReferencePlatform.tar.bz2 so you have something that the suppression file will properly suppress.

Comment 6

[Phil Ringnalda (:philor)]

Ick, actually that VM isn't going to be all that close to real, since that predates when releng started using puppet to update various packages that were just too old to still work.

Comment 7

[Nicholas Nethercote [inactive]]

This is subsumed by bug 631811.