RE: Stability progress?

[Nate Schindler]

Allright.  I just installed valgrind.
I'll see if I can duplicate my issue with the tool running.

Do you have a preferred list of command line switches for valgrind?

I also set "ulimit -c unlimited", which hadn't been done yet.

-----Original Message-----
From: Dan Nelson [mailto:address@hidden
Sent: Tuesday, April 26, 2005 11:55 AM
To: Nate Schindler
Cc: address@hidden
Subject: Re: Stability progress?


In the last episode (Apr 26), Nate Schindler said:
> The "dying under heavy load" problem became a lot worse in my
> installation after upgrading to 0.3.0.  I was wondering if any
> changes have been planned to help correct this, or if the cause is
> even known yet?
>
> One of my inside users sends a broadcast e-mail to about 1000 people
> every couple of weeks, but this could just as easily happen from the
> outside. What I end up with are a whole bunch of hung spamc
> processes, and spam passes through until I kill everything off and
> bring it all back up again.
> 
> If it helps, I'm running:
> RedHat 7.2
> kernel 2.4.20-28.7
> Sendmail 8.13.3
> SpamAssassin 3.0.2 (with standard plugins installed and enabled)
> spamass-milter 0.3.0
> clamav-milter 0.83
> 
> If there's any debugging information you'd like to to gather, I'd be
> happy to do so.

If it's crashing, a stack trace is the best way to track the problem
down.  Either run gdb on a coredump, or try running the milter under
valgrind and see if it traps the error.  Every array or list should be
dynamic, so even 1000 recipients shouldn't pose a problem.  The only
places there are possible buffer overflow are if your OS doesn't
support a safe printf (asprintf or snprintf) and:

- If an incoming recipient address is over 1024 chars and -x is used
- If the -b or -B recipients are over 1024 chars long
- If a logging message is over 1024 chars long

RedHat 7.1 has snprintf so the worst case is that long addresses will
get truncated at 1024 chars but not cause an overflow.

-- 
        Dan Nelson
        address@hidden