[Raw Msg Headers][Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: router eating sick amounts of core
> On a sun4/380 running sunos4.1.3, zmailer's router grows to about 80MB
> of ram and hovers around 83MB for the rest of it's life. Many times
> it segfaults and leaves a heavily (300k!!!) core file so there is no
> hope of me trying to throw a debugger on this puppy.
>
> I recall reading that there are some memory problems when expanding
> large mailing lists. This site runs about 70 or so majordomo lists,
> most have over 2,000 subscribers per list. I was hoping that zmailer
> would make my site humm, but it's just eating a lot of ram and doing
> worse than sendmail ever did.
The fix I deviced to that was mostly in the configuration
scripts. It works just fine, if:
- you don't use the old original listexpansion script
- instead you either feed individual addresses of the
recipients to the router, OR you feed it names of
(sub)lists, and use the new script:
$MAILSHARE/cf/aliases.cf:
----------------------------------------------------------------
a="$MAILVAR/lists/$lcuser"
nattr=()
[ $didexpand ] || [ -f "$a" ] &&
db add expansions "$key" list &&
priv=$(getpriv "664" $priv "$a" maillist) &&
nattr=($(newattribute $attr privilege $priv sender "$lcuser-owner")) &&
return $(listexpand -E "$user-owner" -p $priv -e "$lcuser-owner" -c "$a file expansion" $nattr "$a" "$user" < "$a")
----------------------------------------------------------------
If you use /etc/aliases to expand your sublists, there
THIS change does not affect, unfortunately..
I THOUGHT the Majordomo feeds individual addresses to the
mailer, but apparently it behaves differently from that of
the Listserv, and Listproc systems.
> In the meantime, I imagine that the router is caching the smtp
> information as it expands new list postings. Is there a way to config
> the router to not cache everything or to do less caching?
It caches some information per message resolving (loop-
preventation information), and some other for a bit longer
times (DNS lookups), but none of those are very long-living
databases (nor particularly large..)
> Or do people suspect memory leakage? I'm using a pretty trustworthy
> resolver library to link against, so I don't suspect that to be the
> problem.
I don't suspect it -- resolver is not (usually) the culprit..
> Oh, and this is with the latest rev. of zmailer by mea on ftp.funet.fi.
> (2.99.21 I believe).
>
> Later,
> David S. Miller
> davem@caip.rutgers.edu
/Matti Aarnio <mea@utu.fi> <mea@nic.funet.fi>