Re: spamass-milter dies under heavy load

[Dan Nelson]

In the last episode (Sep 23), Matt Cramer said:
> On Fri, 20 Sep 2002, Dan Nelson wrote:
> 
> > If your HP-UX boxes are your mail servers and the Linux boxes are your
> > spamassassin servers, you want to put spamass-milter on the HP-UX box,
> > and edit the SPAMC define in config.h to "spamc -d linuxhost".
> > Theoretically, you could run `` SPAMC="spamc -d linuxhost" ./configure ''
> > but for some reason that doesn't work.  And I just figured out why :)
> > Next release you will be able to do that.
> 
> Well actually that didn't work either.  I had to go into the
> spamass-milter.cpp code and change where it lists arguments to the SPAMC
> define.  I just hard-coded it to my executable with the right flag for the
> remote host.  Putting the string "/usr/bin/spamc -d linuxhost" into
> config.h, config.status, and the Makefile did not work.  They caused me to
> get the following error in the log:

You're right.  I forgot SPAMD was exec()'d.  I guess a more general
solution would be to allow the user to specify spamc args when starting
the milter.  I wonder if there's a patch for that already.

> I am interested to see if this configuration (with spamc->spamd over
> the network rather than sendmail->spamass-milter) improves the
> ability to perform under load.

I'd guess it would.  The milter gets called once per header, and once
per NN bytes of body text (not sure the exact size).  Contrast that
with spamc, which sends the entire message to spamd, and receives the
entire response in one transaction.  Anything that reduces round-trips
over the network will speed you up.

 
-- 
        Dan Nelson
        address@hidden