RunAway milters

[Chuck Yerkes]

Platform:
OpenBSD 3.3/SPARC.

Running an unpatched current download (0.1.3a).
Straight build, no issues.

Every couple hours, I get one that's running away eating
CPU.  A ktrace on it show this forever (10's of thousands/second):

 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2
 29721 spamass-milter PSIG  SIGSEGV SIG_DFL code 2 addr=0x8b8c3dd4 trapno=2


Of course, when I run it in debugging mode, filling my logs, it
won't runaway.  Grrrr.

Worth noting, I believe I had more of these when I had the milter timeouts
set such that more scans took longer than the timeout.  I bumped the
timeouts to 5 minutes from 1 on this fairly slow and loaded machine
(old SPARC 10).

Has anyone seen this behavior?  I've got a cron job catching processes
with more than 10 minutes of process time and killing them.  That's the
wrong answer, but it works.


(I'm still wishing sendmail were consistent and had a "check_$MILTERNAME"
rule that would let me choose not to milter things in rulesets.)


Thanks for tips/pointers.