> I am running Redhat AS 3.0 and SA 2.63 with sendmail and spamass-milter.
> I will start by mentioning the following:
>
> - I am processing about 10,000 msgs / hour
> - This is a dual proc 3.2GHz Zeon with 2G of RAM.  The system is not
> swapping at all (based on vmstat)
> - I have already fixed the UTF8 issue
> - I am not running any external checks (have run with spamd -L and still
> have the same issue)
> - I have disabled the Bayes stuff (from the local.cf file)
> - I am not running the auto white list stuff (from the local.cf file)
>
> Now for the fun part...  I am hitting loads of 10+ on the server on a
> regular basis which makes sendmail back off (I know how to fix the
> sendmail issue so please do not comment on that).
>
> I have tried everything I can think of to take the load off the system
> but nothing is working for me.
>
> When I do a TOP on the system, this is what I see...
>
>  18:57:50  up 8 days,  3:29,  2 users,  load average: 9.37, 9.35, 9.38
> 124 processes: 108 sleeping, 16 running, 0 zombie, 0 stopped
> CPU states:  cpu    user    nice  system    irq  softirq  iowait    idle
>            total   74.4%    0.0%   22.9%   0.0%     0.0%    0.0%    2.4%
>            cpu00   76.1%    0.0%   21.9%   0.0%     0.1%    0.0%    1.7%
>            cpu01   70.2%    0.0%   26.0%   0.0%     0.0%    0.1%    3.4%
>            cpu02   76.7%    0.0%   21.1%   0.0%     0.1%    0.0%    1.9%
>            cpu03   74.6%    0.0%   22.4%   0.0%     0.0%    0.1%    2.6%
> Mem:  2061652k av, 1496032k used,  565620k free,       0k shrd,   84660k
> buff
>                     905668k actv,  258248k in_d,    8904k in_c
> Swap: 2097112k av,       0k used, 2097112k free                  439292k
> cached
>
>   PID USER     PRI  NI  SIZE  RSS SHARE STAT %CPU %MEM   TIME CPU
> COMMAND
> 17262 mail      25   0  107M 107M  2400 R     7.9  5.3   0:01   2 spamd
> 17298 mail      25   0  108M 108M  2400 R     4.7  5.3   0:00   1 spamd
> 17284 mail      25   0  108M 108M  2392 R     4.5  5.3   0:00   3 spamd
> 17286 mail      25   0  108M 108M  2392 R     4.3  5.3   0:00   3 spamd
> 17301 mail      25   0  108M 108M  2412 R     3.5  5.3   0:00   3 spamd
> 17295 mail      25   0  107M 107M  2392 R     3.4  5.3   0:00   3 spamd
> 17333 mail      25   0  108M 108M  2392 R     3.4  5.3   0:00   2 spamd
> 17319 mail      25   0  108M 108M  2416 R     3.3  5.3   0:00   0 spamd
> 17310 mail      25   0  108M 108M  2416 R     3.0  5.3   0:00   1 spamd
> 17314 mail      25   0  108M 108M  2416 R     2.9  5.3   0:00   0 spamd
> 17330 mail      25   0  107M 107M  2412 R     2.9  5.3   0:00   0 spamd
>  7944 mail      23   0  107M 107M  2376 S     1.2  5.3   5:03   2 spamd
>
> Does anyone have any suggestions on what I can try next?  Any help is
> greatly appriciated...

Hmm,

I don't know what is causing your problem specifically, but I see one
thing that is VERY wrong. Spamd should not be using 108M of ram, which is
what seems to be happening in your case. Normally Spamd will use about 18M
of ram per process. Find out why that is happening (I can only suggest a
process of elimination) and you should be well on the way to finding what
the heck is going on....a freshly started spamd should only be using about
18M of ram.

A couple of others have had a similar problem with excess memory usage
before although I don't recall what the reason was - extra rulesets with
errors in them perhaps. Maybe they will see your message and respond.

Also consider that 10,000 message per hour is a BIG load for a single
machine of the specs you listed when running spamassassin, and depending
on how you integrate spamassassin, and you may simply be loading the
machine too much.

You don't say how you're calling spamassassin but since you're using spamd
and sendmail, I'll put two and two together and guess that you're calling
spamc from procmail ? This is what I'm doing too, and we're starting to
notice serious load issues now that the amount of spam is going up...

Spec's here are P4 2.4Ghz, 1GB ram, single CPU, and by comparision we're
only doing about 50,000 messages per DAY, and I'm seeing load average
problems (>10) frequenly.

According to some on this list, sendmail + procmail + spamc/spamd simply
doesn't scale, and I'd have to grudgingly agree with this, and I'm looking
into my options to improve this... The problem seems to be 50/50
procmail/spamc overhead, and sendmail's poor local delivery scheduling and
queuing system. (Which tends try and scan a lot of things at once without
any proper throttling of maximum local simultanous deliverys)

The options are to basically call spamassassin directly from sendmail
using something like MimeDefang, Mailscanner or Amavisd, which apparently
performs a lot better, but you lose a lot of per-customer control of spam
settings, (a problem for us) since it calls SA *before* a message is
seperated out to each recipient,

Or to use a different mailserver altogether that supports better
integration with spamassassin, and I've been taking a look at Exim
recently, but havn't got very far yet.

So in summary, you do have something screwy going on there with the ram
usage, but even if you resolve that, if you are using a
sendmail/procmail/spamc combination you can expect to be getting
performance problems with the number of messages you're processing, so you
need to look seriously at a different integration approach for
spamassassin or perhaps dumping sendmail...(both of which I'm exploring
myself right now....)

Regards,
Simon



Reply via email to