On Sat, 19 Aug 2006 18:06:59 -0400 "Daryl C. W. O'Shea" <[EMAIL PROTECTED]> wrote:
> Josh Trutwin wrote: > > On Fri, 18 Aug 2006 12:16:51 -0400 > > "Daryl C. W. O'Shea" <[EMAIL PROTECTED]> wrote: > > > >> Josh Trutwin wrote: > >>> I've recently had a server experience some really slow spam > >>> processing - I'm not sure what's going on but I notice a lot > >>> of timeouts in the mail log: > >>> > >>> Aug 18 09:20:21 www spamd[27673]: timeout with empty $@ > >>> at /usr/local/share/perl/5.8.4/Mail/SpamAssassin/Timeout.pm > >>> line 182, <GEN16> line 1126. Aug 18 09:22:02 www spamd[27674]: > >>> timeout with empty $@ > >>> Any suggestions? > >>> > >>> Debian linux - spamd 3.0.4 with pyzor/dcc/razor > >>> > >>> spamd running with: > >>> > >>> /usr/bin/spamd -d -D -q -x -H /etc/razor --max-children=12 > >>> --socketpath=/var/spool/spamassassin/spamd.sock -u spamd > >> Unless you've got at least 600 or more MB of free RAM just for > >> spamd's use, you've got too many children and are swap > >> thrashing. Back of the --max-children number. > > > > I was getting the same results with less values - the box has 1 > > GB > > - more is on the way though. I disabled network tests with -L > > and things work great again so something along that line is the > > culprit. > > Using -L processes messages faster, thus requiring less children > to handle the load. This error is caused by a system not being > able to restore the child's config fast enough after it's done > processing a message. It's always due to one of two things... > high load, or high load caused by swap thrashing. > > If you're using a lot of add-on rulesets your children may be > taking up even more memory than budgeted above. See what they're > using and confirm that you're not going to hit swap. If the > problem still persists I'd like to hear about it. Still having problems - even with -L. Server has 1 GB of memory, more is on the way I hope. Anyway - I have the following rules: fastconcepts:/etc/mail/spamassassin# ls 10_misc.cf 88_FVGT_headers.cf 70_sare_adult.cf 88_FVGT_rawbody.cf 70_sare_bayes_poison_nxm.cf 88_FVGT_subject.cf 70_sare_evilnum0.cf 88_FVGT_uri.cf 70_sare_genlsubj.cf 99_FVGT_DomainDigits.cf 70_sare_header.cf 99_FVGT_meta.cf 70_sare_highrisk.cf 99_sare_fraud_post25x.cf 70_sare_html.cf RulesDuJour 70_sare_obfu.cf antidrug.cf 70_sare_oem.cf backhair.cf 70_sare_random.cf bogus-virus-warnings.cf 70_sare_ratware.cf chickenpox.cf 70_sare_specific.cf coaching.cf 70_sare_spoof.cf init.pre 70_sare_stocks.cf local.cf 70_sare_unsub.cf mr_wiggly.cf 70_sare_uri.cf random.cf 70_sare_uri0.cf tripwire.cf 70_sare_whitelist_rcvd.cf tsa-list.cf 70_sare_whitelist_spf.cf useless.cf 72_sare_bml_post25x.cf v310.pre 72_sare_redirect_post3.0.0.cf v312.pre 88_FVGT_body.cf weeds.cf I created coaching.cf and tsa-list.cf - they are basically one (well two technically) liners. If I need to cut back I'm not sure where to start - the SARE rules are the ones listed on rulesemporium.com - the FVGT are on http://www.exit0.us/index.php?pagename=RulesDuJourRuleSets. Of the rest, only bogus-virus-warnings.cf is of any substantial size (and still 1/3 the size of the largest SARE rule). Should v310.pre be removed if v312.pre is also there? Thanks, Josh