More 4.50.15 woes on FreeBSD

[TCIS List Acct]

TCIS List Acct wrote:

> Ok, I've tracked down part of the problem (from the 4.48.4-2 release 
> notes):
> 
> - Rearranged SpamAssassin spam.assassin.prefs.conf file, it is now read 
> by SpamAssassin via a link called "mailscanner.cf" in the site_rules 
> directory.  It is no longer read directly by MailScanner, it is just 
> read by Spam-Assassin during its normal initialisation process.
> 
> This really hosed those of us not using an RPM or install.sh script (and 
> those of us dumb enough not to read the relnotes before doing this :)) 
> -- it caused my SA prefs not to be read in, and thus caused SA to use 
> its default settings (Bayes on, DNS BL's on, etc) which caused the 
> majority of the performance problems.
> 

Unfortunately the above was just a temporary fix -- the problem has cropped up 
again.  Specifically, if I let 4.50.15 run for a few hours, it appears to leak 
memory and/or zombie perl processes to such an extent that the box runs out of 
RAM and swap space.  This occurs with the new SpamAssassin cache turned off as 
well, so that isn't the issue.  Here is a snippet of "top" when it is occurring:

last pid: 97100;  load averages: 16.75, 20.74, 26.46 
                     up 63+05:29:19  21:18:11
297 processes: 53 running, 191 sleeping, 53 zombie 

CPU states: 22.6% user,  0.0% nice, 13.4% system,  0.3% interrupt, 63.7% idle 

Mem: 1529M Active, 183M Inact, 287M Wired, 7412K Cache, 199M Buf, 3004K Free 

Swap: 2048M Total, 960M Used, 1088M Free, 46% Inuse, 452K In, 4804K Out

I have the number of child processes set at 10, but there are 100+ MailScanner 
processes according to a ps -aux | grep MailScanner

$ ps -aux | grep MailScanner | wc -l
      100

Note that downgrading to 4.47.4 on the same box (all I do is change the symlink 
to where /opt/MailScanner points to) immediately solves the issue and the box 
runs normally from that point on.  The MailScanner.conf files for both versions 
are using near identical parameters (same number of a/v scanners, same DNSBL's 
being used, same spam.assassin.prefs.conf, etc) so I am certain it is something 
that changed code-wise between 4.47.4 and 4.50.15.

Julian, I'll be happy to troubleshoot further if you give me some pointers on 
what to try next.

-- 

-----------------------------------------
Mike Bacher / listacct at tulsaconnect.com
TCIS - TulsaConnect Internet Services
http://www.tulsaconnect.com
-----------------------------------------