The mailing list for listmasters using Sympa

[Mark Sallee <address@concealed>]

Olivier,
We changed a couple of things that appear to have helped the system
performance.
1. Moved the virtual machine hosting sympa to another system with more
RAM (4GB)
2. Corrected a few addresses with strange characters in our Oracle
database (comma-separated addresses on one line, multiple '@' signs in
one address).
3. Added the line 'distribution_mode fork' to sympa.conf so we get two
sympa.pl processes running

The processes are currently using memory as follows: using 'top' while idle
  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
 3456 sympa     16   0 37432  28m 3332 S    0  0.7   0:00.30 sympa.pl
 3457 sympa     16   0 39536  30m 3328 S    0  0.8   0:00.37 sympa.pl
 3460 sympa     16   0 33604  23m 2940 S    0  0.6   0:00.01 archived.pl
 3463 sympa     16   0 32868  22m 2276 S    0  0.6   0:00.07 bounced.pl
which appears to be fine.

But when I send one message to a sympa list, the CPU and memory use goes
up dramatically: these next lines were captured one after another, as
fast as I could copy and paste from the terminal.
PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
3457 sympa     25   0 1249m 1.2g 3328 R  100 30.2   0:07.20 sympa.pl
3457 sympa     25   0 2190m 2.1g 3328 R  100 54.1   0:13.21 sympa.pl
3456 sympa     16   0 37432  28m 3332 S    0  0.7   0:00.34 sympa.pl

One of our 2 sympa processes has died and task_manager is not running:
ps -ef | grep .pl
sympa     3456     1  0 Oct28 ?        00:00:00 /usr/bin/perl
/home/sympa/bin/sympa.pl
sympa     3460     1  0 Oct28 ?        00:00:00 /usr/bin/perl
/home/sympa/bin/archived.pl
sympa     3463     1  0 Oct28 ?        00:00:00 /usr/bin/perl
/home/sympa/bin/bounced.pl

/etc/init.d/sympa status
Status of Sympa subsystem:
Status file for subsystem found.
sympa (pid 3456) is active...
archived (pid 3460) is active...
bounced (pid 3463) is active...
task_manager died, pid file remains.




I can't answer your question right now on how much outgoing e-mail
traffic we have, but we will find out after we start sending to some
large lists. In this stage of testing, we haven't really begun to use
the resources yet and we will 'ramp' up gradually. In this case I'm only
trying to send out one message to a list with one subscriber. I have
also tested with a list of 100 test accounts and it acts about the same.

Thank you.

Mark Sallee



Olivier Salaün wrote:
> Hi Mark,
>
> First I should mention that we were never reported such significant
> problems with memory shortage on a Sympa server.
>
> However we are aware that Sympa processes suffer from memory leaks,
> but we haven't determined yet what perl module is responsible for them.
> It would greatly help if you provide additional informations about
> your sympa service including :
>
>    * the memory used by each Sympa process (sympa.pl, task_manager.pl,
>      bounced.pl)
>    * the average outgoing traffic of your server (#mail X #members)
>
> Feel free to send these informations offlist if your prefer.
>
> As an example:
>
>    * we are running our own ML service on a server with 8GB RAM.
>    * we have lists with 20.000 members
>    * the outgoing traffic is around 1.000.000 messages a day
>
>
> I don't think you'd have better performances with an older Sympa
> release ; especially since we've optimized a few things in the 5.4.x
> versions on the web interface.
>
>