bayes expire tokens

[Brady Tucker]

Chris,

I'm going to be very verbose in all of this.. Because this problem haunted
me and at least one other guy here for a very long time...and we finally
worked on it off list together.... I think everybody else here thought we
were crazy ;)   I ran for a year+ without this problem... Then starting
having this problem sporadically.. Then it became an continually occurring
annoyance with 100's of them/day.....

I continued to have these problems with various SA/MS versions on 4 servers
even after following the advice of many people in multiple threads last
year.... Even going so far as to try mounting my /root/.spamassassin dir in
tmpfs with some hourly backup scripts and a script that repopulates it from
backup on boot) which helped but didn't fix the problem.  I have 2 gigs of
ram in my main MailScanner machine and played with anything from 3 to 10
children and 5 to 200 messages per scan.

I've disabled the rebuild in SA,  done the same as you have in
MailScanner.conf with the wait/build/reload parameters etc... To no avail at
the time I was having problems.

Here's what finally fixed it for me... I run the bayes expire script from a
cron job at least 3 times a day.  If I Run it twice and I get a few lock
files... Run it once/day and I get a crapload.... Run it three times/day and
its smooth.   I therefore do it every 8 hours.  I'm assuming that the bayes
DB gets so big that doing the expire job only once/day it takes to long to
finish and it gives up and tries again repeatedly ?  (even when running in
tmpfs)????  I don't know why.. But it works.. And worked on all 4 servers...
The expiry only takes about 1-3 minutes when run 3 times a day... Was taking
upwards of 7 to 10 when once/day... Can't believe it was the problem...
But.. There you have it.  Somebody smarter than me can either tell me I'm
crazy... Or explain why.

I've setup boxes for clients that are *low* (less than say 5,000
message/day) usage, or *real* servers (dual/quad proc's etc doing other jobs
as well) and don't have these symptoms and never have.  I don't bother with
the tmfps bayes or tmpfs on mailscanner/incoming or any of the other
optimizations suggested in the MAQ for those machines and they run fine.

Good luck...  Let me know how you come out,

                                                 Brady A. Tucker
                                                 batucker at icnet.net
                                                 Internet Complete!
                                                 w w w . i c n e t . n e t


My basic HW/SW Configs:
  My 4 systems are averaging 80-100,000 msg/day each
    all but one act as a gateway for a 5th machine,
    the 4th acts as a gateway but delivers some locally
  FC3 on all but one which still has RH9 on it
  Each machine is SCSI Raid V, various HW/Drive speed, 1 machine is SATA
  Procs from AMD Barton 2500's to AMD64 3500+
  1 or 2 gigs in each machine
  Running SA/MS/DCC/Pyzor/Razor/ClamAV/McAfee (latest or close to latest on
each)
     few rbls in SA disabled.. But mostly vanilla install
  Running MailWatch 0.5.1 w/sql logging
  ( MySQL on each machine for MailWatch with some CF scripts
    I've written for searching/managing them by techs here )

------------------------ MailScanner list ------------------------
To unsubscribe, email jiscmail at jiscmail.ac.uk with the words:
'leave mailscanner' in the body of the email.
Before posting, read the MAQ (http://www.mailscanner.biz/maq/) and
the archives (http://www.jiscmail.ac.uk/lists/mailscanner.html).

Support MailScanner development - buy the book off the website!