[rt-users] Charts displaying only "sometimes" (GD problem?)

Kevin Falcone falcone at bestpractical.com
Sat Nov 26 19:57:50 EST 2011

On Thu, Nov 24, 2011 at 12:03:37AM +0000, Scotto Alberto wrote:
> Eg. given chart A:
> - if I login with user1 it shows, but with user2 it does not;
> - given a user, if I select <bar,queue> it doesn't show, but <bar, CreatorNickname> works
> - given a user and a certain pair <type_of_chart, domain_on_x_axis>, it works if I turn to English from user preferencies; it doesn't if I select Italian
> - and so on...

Are these all repeatable, or does it sometimes work and sometimes

> I'm using RT 4.0.2 (never upgraded, it's a fresh install) on Red Hat Enterprise Linux AS release 4 (Nahant Update 2) with Apache/2.0.52, mod_perl ??, GD::Graph 1.44, glibc-2.3.4-2.13
> 1a. as soon as my browser asks for the image and RT doesn't generate it, RT prints in /opt/rt4/var/log/apache2.error the following 2 lines:
> *** glibc detected *** free(): invalid pointer: 0x0dd16764 *** [Wed Nov 23 16:38:23 2011] [notice] child pid 10963 exit signal Aborted (6), possible coredump in /tmp/apache_core_dumps
> 1b. There's something more. Here's what apache's core dump says (resulting from the commands gdb /usr/sbin/httpd /tmp/apache_core_dumps/core.22613 + bt full):
> #0  0x0096c7a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2 No symbol table info available.
> #1  0x006a07d5 in raise () from /lib/tls/libc.so.6 No symbol table info available.
> #2  0x006a2149 in abort () from /lib/tls/libc.so.6 No symbol table info available.
> #3  0x006d440a in __libc_message () from /lib/tls/libc.so.6 No symbol table info available.
> #4  0x006dab3f in _int_free () from /lib/tls/libc.so.6 No symbol table info available.
> #5  0x006daeba in free () from /lib/tls/libc.so.6 No symbol table info available.
> #6  0x00cb2a57 in gdFree () from /usr/lib/libgd.so.2 No symbol table info available.
> #7  0x00be9c36 in XS_GD__Image_png (my_perl=0x9b85a98, cv=0xa875550)
>     at GD.xs:944
>         data = (void *) 0xcec9a44
>         size = 3545
>         level = Variable "level" is not available.
> 2. after enabling logging by adding Set($LogToFile, "debug") to RT_Site_Config.pm, I noticed the following message in rt.log:
> [Wed Nov 23 10:56:18 2011] [debug]: You've enabled GraphViz, but we couldn't load the module: Can't locate GraphViz.pm in @INC (@INC contains: [...] at /opt/rt4/sbin/../lib/RT/Config.pm line 542. (/opt/rt4/sbin/../lib/RT/Config.pm:543)

This log is unrelated to what you're seeing.
You're much more likely to get something useful from apache's logs
than RT's logs.

> - clean mason cache (more than once)
> - clean browser's cache (more than once)

These shouldn't be relevant.

> - reinstalled GD::Image (with cpan 'force install')

Why did you need to force install?
Does GD::Image pass tests under the cpan shell?
The top-level package should actually be GD, RT requires
that GD, GD::Graph and GD::Text are all installed and working properly
to generate charts.

> - added Set($DisableGD, undef) to RT_Site_Config.pm, as well as Set($DisableGD, 0)

These are just the defaults so shouldn't matter.

> According to clue#1b it's GD that breaks everything ("at GD.xs:944").
> Can you confirm my interpretation is correct? If so, maybe I could go
> for a reinstall of GD..? Otherwise I could upgrade glibc, to what
> version?

> Nevertheless, I *feel* like it's a cache problem, server-side. Not client-
> side because cleaning browser's cache doesn't help. I don't know if
> there exists a server-side cache in RT other than mason cache(which I
> already cleaned), however it's my own best explanation. In fact, if it
> were glibc or GD, I should never ever get any charts then! While
> "sometimes" I do!

It's much more likely that you're exposing some sort of bug in GD than
having a cache problem.  RT doesn't cache charts, it generates them on
the fly.

Are you getting other segfaults on the machine?  I'd be looking at the
version of libgd itself and the GD perl modules.

If you can provide a replication recipe that one of us could reproduce
in a debugging environment, that would be great to see as a bug

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <http://lists.bestpractical.com/pipermail/rt-users/attachments/20111126/25526496/attachment.sig>

More information about the rt-users mailing list