Developers of Sympa

[Olivier Salaün <address@concealed>]

Hi Thomas,

Your Sympa server should be able to deliver messages a bit quicker ;-) 
As an example, our Sympa server gets rid of a messages for 16.000 
subscribers in 481 seconds.

Note that sympa-distribute processes end up forking, with the child 
process executing the sendmail program and dying. I suppose that the 
traces you get relate to the sendmail program. Therefore you should 
investigate in this direction ; Sympa being probably not guilty  O:-)

Thomas Berry a écrit :
> Sympa version 5.3.2 on Red Hat EL4 update 4
>
>  This is part of my continuing saga dealing with a distribution of 
> 7600 subscribed members that is taking 45-60 minutes for Sympa to 
> complete.   Another list of 440 messages took 95 minutes to complete, 
> while smaller lists take only a few seconds to complete.
>
>  Today, I setup Sympa so that its call to sendmail would queue the 
> messages locally.  I setup queue runner sendmail process to deliver 
> the messages from the local queue. (Sendmail is configured with a 
> smart host so that it delivers to a cluster of MTAs on our domain).  
> The interesting part of this is that Sympa only took 20 seconds to 
> complete this distribution in our development environment running the 
> same Sympa and sendmail configuration, Sympa and OS release, and a 
> list with the same list of members exported from the mailing list in 
> production.
>
>  During the distribution, top shows the load average at about .15; 
> over 600MB out of 2GB of free memory; and a very low IO wait.
>
>  Note that the sympa-distribute process that writes the message file 
> takes several minutes for each message to complete: "write(11, " 
> \nScience activities concluded w"..., 2265) = 2265".  Also of note, 
> once Sympa finished generating the messages for this distribution, the 
> messages that were queued up behind it distributed rather quickly with 
> no signs of delay.  I've also done SQL and LDAP tests which show no 
> signs of delays (From what I know about Sympa, interactions with LDAP 
> and SQL have completed before the distribution process starts 
> generating messages).
>
>  I've asked this before, but I'm getting desperate trying to find out 
> why the production system isn't performing anywhere near as 
> efficiently as the development server (both environments use the 
> hardware).
>
>
>  Here's a sample of strace output from the sympa-distribute process 
> while it was constructing the messages for the distribution.  Maybe 
> someone can glean some information from it.
>
> --- SIGCHLD (Child exited) @ 0 (0) ---
> select(0, NULL, NULL, NULL, {0, 297000}) = 0 (Timeout)
> write(11, "Received: from mail.jpl.nasa.gov"..., 4096) = 4096
> write(11, " \nScience activities concluded w"..., 2265) = 2265
> close(11)                               = 0
> pipe([10, 11])                          = 0
> ioctl(10, SNDCTL_TMR_TIMEBASE or TCGETS, 0xbffbaf58) = -1 EINVAL 
> (Invalid argument)
> _llseek(10, 0, 0xbffbaf90, SEEK_CUR)    = -1 ESPIPE (Illegal seek)
> ioctl(11, SNDCTL_TMR_TIMEBASE or TCGETS, 0xbffbaf58) = -1 EINVAL 
> (Invalid argument)
> _llseek(11, 0, 0xbffbaf90, SEEK_CUR)    = -1 ESPIPE (Illegal seek)
> fcntl64(10, F_SETFD, FD_CLOEXEC)        = 0
> fcntl64(11, F_SETFD, FD_CLOEXEC)        = 0
> clone(child_stack=0, 
> flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD, 
> child_tidptr=0xb7fa8708) = 23853
> --- SIGCHLD (Child exited) @ 0 (0) ---
> close(10)                               = 0
> select(0, NULL, NULL, NULL, {0, 299999}) = 0 (Timeout)
> write(11, "Received: from mail.jpl.nasa.gov"..., 4096) = 4096
> write(11, " \nScience activities concluded w"..., 2265
>
>
>
> Here's the sympa log event that shows the distribution completed:
>
> Sep  4 14:58:46 list sympa(message)[20483]: main::DoMessage() Message 
> for cassini from address@concealed accepted (2790 seconds, 59 
> sessions, 7695 subscribers), 
> message-id=<p06230916c4e5fa93c13c@[137.78.239.39]> , size=6114