[request-tracker-maintainers] Bug#649864: Bug#649864: request-tracker4: modperl apache2 config leads to memory full

[Olivier Berger]

On Thu, 24 Nov 2011 13:11:23 +0000, Dominic Hargreaves <dom at earth.li> wrote:
> On Thu, Nov 24, 2011 at 12:57:30PM +0100, Olivier Berger wrote:
> > On a 2Gb VM, I'm running a rebuilt package of RT 4 on a squeeze + backports system.
> > 
> > I configured the apache config to use mod_perl.
> > 
> > The memory used by the apache process is reported like this in top :
> >   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+   
> > 18864 www-data  20   0  827m 345m 4428 S  0.0 17.2   0:08.15 apache2
> > 18836 www-data  20   0  828m 345m 4388 S  0.0 17.2   0:08.64 apache2
> > 18828 root      20   0  517m 322m 6124 S  0.0 16.1   0:03.40 apache2
> > 18834 www-data  20   0  516m 316m  692 S  0.0 15.8   0:00.00 apache2
> > 27804 mysql     20   0  173m  39m 7568 S  0.0  2.0   0:27.58 mysqld
> > 
> > I fear there's a problem with the memory management of RT, or RT with mod_perl.
> 
> Based on those figures alone, I wouldn't say that there was a problem.
> You might find that people who have direct experience of running RT4 with
> mod_perl can provide more feedback though (I haven't).
> 
> > I used to run RT 3.8 from squeeze before, and had experienced problems with memory full and actions not being executed because of memory exhaustion reported in the logs.
> > 
> > I tried to upgrade in order to see if things would be improving with the next version.
> > 
> > Is this a known problem ?
> 
> It's not clear yet that there is a problem.
> 

I can't believe one needs 2 Gb to run a web app that manages 3 dozens
tickets.

I've now switched to mod_fcgid and have now :

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND                                                                                                                                                                                                           
20202 www-data  20   0  218m 103m 4172 S  0.0  5.2   0:03.65 rt-server.fcgi
20206 www-data  20   0  217m 102m 4172 S  0.0  5.1   0:02.49 rt-server.fcgi
20207 www-data  20   0  216m 102m 4172 S  0.0  5.1   0:02.68 rt-server.fcgi
20145 www-data  20   0  318m 9332 2712 S  0.0  0.5   0:00.67 apache2
20139 root      20   0 98.4m 8932 4616 S  0.0  0.4   0:00.16 apache2
20144 www-data  20   0  317m 8104 2060 S  0.0  0.4   0:00.64 apache2
20143 www-data  20   0 97.7m 4848  636 S  0.0  0.2   0:00.06 apache2
20142 www-data  20   0 98848 4820  608 S  0.0  0.2   0:00.00 apache2

I'm not sure this is ideal, but looks a little bit better to me.

What's the usual practice, and may there be recommendations to be
provided for users of a not so much tuned RT install for Debian ?

I'm pretty sure there may be tons of customizations done to tune the
RDBMS, the Web server and RT itself if memory consumption is a problem,
as for any complex app, but if it's packaged for Debian, it would be
great to have reasonable defaults that can be used in production on a
"regular" VM (maybe I'm a bit too old, but dedicating 1 Gb RAM to a Web app
used by a dozen users seems comfortable to me in general)...

Thanks for you help.

Best regards,
-- 
Olivier BERGER 
http://www-public.it-sudparis.eu/~berger_o/ - OpenPGP-Id: 2048R/5819D7E8
Ingenieur Recherche - Dept INF
Institut TELECOM, SudParis (http://www.it-sudparis.eu/), Evry (France)