Re: Majordomo performance issue

[Daniel Liston <dliston @ sonny . org>]

You may want to consider examining this section of the majordomo.cf file;

# You can force Majordomo to delay any processing if the system load is too
# high by uncommenting the following lines.  THIS ONLY WORKS if your "uptime"
# command (usually found in /usr/bin/uptime or /usr/bsd/uptime)
# returns a string like:
#   5:23pm  up  5:51,  9 users,  load average: 0.19, 0.25, 0.33
#
$max_loadavg = 10;                 # Choose the maximum allowed load
#
$uptime = `/usr/bin/uptime` if -x '/usr/bin/uptime';     # Get system uptime
#$uptime = `/usr/bsd/uptime` if -x '/usr/bsd/uptime';     # or uptime is over here.
#
($avg_1_minute, $avg_5_minutes, $avg_15_minutes) =
     $uptime =~ /average:\s+(\S+),\s+(\S+),\s+(\S+)/;
#
exit 75 if ($avg_15_minutes >= $max_loadavg);           # E_TEMPFAIL
#

Tune to your specific needs of course.  On my little 400Mhz single proc
intel box, I set $max_loadavg=2.

Dan Liston

Stewart Dean wrote:

> I'm trying to improve my MD mailing list throughput/latency.  What I see 
> when I run a 300-400 member list (on 4way IBM AIX box without much other 
> load)
> onesey-twoseys
> is that:
> 1 - TIME) it takes MD about 2-4 minutes to process the list (or that's 
> the time so far elapsed when in shows up in the mail syslog loggin), 
> then it takes another 10-20 minutes for sendmail to fire off all the 
> individual messages.
> 2 - CPU LOAD) while MD is processes there's a considerable load, say 
> 5-10% for the MD process, but after MD is done, the sendmail load 
> thereafter is negligible, hardly shows up at all.
> 
> Alas, there's a ^&%$#@$head that submits mailing for the 
> non-authorized.  This person has been requested to space things out, but 
> noooooooooooooooo, this person saves them up, then dumps them all in the 
> hoper at once.  Basic exercise in making impossible queues and 
> latency....75 such mailing within 2 hours, 135 within four.  What I have 
> seen then is:
> 1 - TIME) growing latency in the MD preprocess time....it begins at 2-4 
> minutes as above, then grows to 1 to 2 hours as the system is under 
> siege...finally, when the submissions stop slow down or stop, the 
> latency decreases back to 2-4 minutes.  Oddly enough, sendmail 
>  processing time is unchanged...even when the MD time has risen over an 
> hour, it only take SM 10-25 minutes
> 2 - CPU LOAD) The 5-10% load per mailing rapidly snowballs to a 80-100% 
> load.
> 
> Can anyone direct me on how to tune the MD processing?