Single instance Postfix install

Eric Sandquist esandquist at IHMS.NET
Fri Jan 23 17:27:06 GMT 2004


An update as to where we are now... Much improved, comments through out the
company, I guess I get to keep my job... Whew... :)

HW: Compaq Proliant 1600, Dual P2 500, 1024meg Ram, 5x9.1g Raid 5...

I should have created a 2x9.1 striped, and a 3x9.1 raid 5... but hind
site...

Mandrake 8.2, Enterprise Kernel 2.4.18.6mdk
Postfix 2.0.10 from tar.gz
SpamAssassin 2.63 from tar.gz
MailScanner 2.54-5 from RPM
ClamAv-0.65 from tar.gz
Using DNS Caching... to minimize lookup times.. :)


Currently have set at 5 processes per CPU...
Finally got the header rules correct in post fix to dump all mail to the
HOLD for processing by MailScanner, modified /etc/init.d/MailScanner for
single instance of postfix...
Using tmfs for /var/spool/MailScanner/incoming
no RBL/DCC/or Razor...
Reduced /home partition and allocated 1g to /tmp ( previously this was not
on it's own partition and the "/" partiotion only had 24meg allocated to
it... an obvous problem...

(Acronis.com has a wonderful partition tool that allowed us to boot from a
CD and adjust partition sizes and create new ones for ext3 and ReiserFS
partision type, works on Microsoft[stool] servers as well, fast too...)

Max message size is set for 10meg...  Realestate brokers love to email
excessively large documents, I may need to raise this (by management
directive), but I hope not...

Still looking for anything that will make it more efficient... and minimize
memory usage...

Processed about 40,000msg in last 24hrs...

We were running a dual postfix installation, response time of the server was
over 15minute to pass an email... Now under 2minutes.. :), memory was maxed
and we had a constant swapfile of 90megs or larger with 512meg RAM at the
time...  When we upgraded the RAM to 1024meg, the system consumed all of it
over a 40minute period...  Seems as though there is a memory leak in one of
apps for email, it didn't do this if email is off...  Also, stopping and/or
restarting email didn't return all the memory that appeared to be in use...

I haven't rebooted the machine to clear all the memory since converting to a
single instance of Postfix...    Server load is at 3.4, used to be over 6 on
a regular basis and climbing...  Memory is at 96% with 14meg in Swap.  Seems
to be stable there...  We'll see if it goes down when I get a chance to
reboot.

I have not gotten the bayes filters set to autolearn yet under postfix...
and with some of the dictionary spam that is coming through I've a little
nervous about having anything automated...

I was intending on adding a hidden email link to our main webpage that would
be something spam-spiders would grab and autofeed everything that came into
it to sa-learn... Good idea? Bad idea?

Does anyone know of any open issues with gzip or other "normal" tools that
are commonly used in scanning email that may have memory leaks?  I've been
scanning my archives, I keep justabout everything that comes into me except
spam, and seem to remember something about a malformed zip issue, but can't
seem to find the reference...

Anyway... so far so good... much better than yesterday when I dinged LILO
when installing the enterprise kernel and the machine wouldn't come back up
..... Blood drains from face, suddenly dizzy and nautious, is this a good
time to go home?...



More information about the MailScanner mailing list