Re: Sudden instability

[Todd Lyons]

Dan Nelson wanted us to know:

>> Up until yesterday, the mail system was very stable (3 load balanced
>> SMTP's handling about 50K messages each per day with clamav and
>> spamassassin).  Yesterday I made one config change to spamd, telling
>> it not to convert marked spams to an attachment (report_safe 0).  SA
>> marks it at 5.0 and the milter rejects it at 7.0.  These are RedHat
>> 9.0 boxen.
>It could be that there's a new spam sender out there with unusual
>header or HELO information that happens to crash the milter.  The
>report_safe stuff shouldn't make any difference.  I believe the redhat

Turns out that it did.  I reverted the change on mx1 and mx3, leaving
mx2 as the only machine with the report_safe 0 setting.  Over the past
24 hours, only mx2 has crashed (about 5 times since this time
yesterday).

>RPMs include all the crashing fixes made to CVS since 0.2.0, so if it's

I don't think I used the RedHat RPM, I rolled my own.  I can't find my
srpm or my spec file any more, I think I built it on a box that we've
since reimaged to a Gentoo box.

I'm going to try a CVS snapshot on a new machine (a Gentoo box) since I
need the virtuser support.

>crashing it's a new bug.  A stack trace might help, or if the machine
>is fast enough, run the milter for a while under valgrind and see if it
>generates any errors.  I only get around 7k messages/day here, and
>valgrind doesn't cause enough load to be a problem on a 500Mhz box.

I might be able to do that.  I've inherited a few other projects
recently but I'll spend some time on it as I can.  I've never used
valgrind before, and I'll have to rebuild the rpms in order to get a
version capable of a stack trace, so my progress on giving you reliable
information will be slow.
-- 
Regards...              Todd
They that can give up essential liberty to obtain a little temporary 
safety deserve neither liberty nor safety.       --Benjamin Franklin
Linux kernel 2.6.3-8mdkenterprise   2 users,  load average: 0.04, 0.04, 0.00