On Tue, Jun 12, 2007 at 03:34:06PM +0200, Til Obes wrote:
> Robert Felber schrieb:
> >Despair:
> >use (bz|z)grep "policyd-weight\[" /var/log/* | \
> > grep -v "(decided|weighted)"
> >Unfortunately there is no strict standard how to log stuff, and each OS logs
> >a little bit different (der Teufel ist ein Eichhoernchen). So grep all your
> >logs which may exist in /var/log (and subdirectories) for policyd-weight
> >messages and grep out (-v) "decided" and "weighted" messages.
>
> Ok i found something:
> Jun 5 11:20:59 mailhost postfix/policyd-weight[13463]: child: spawned
> Jun 5 11:21:03 mailhost postfix/policyd-weight[13463]: warning: ignoring
> garbage: TEST
> Jun 5 11:21:04 mailhost postfix/policyd-weight[13463]: fatal_exit:
> unrecognized request type: ''
> Jun 5 11:21:04 mailhost postfix/policyd-weight[13463]: child: err: fatal:
> unrecognized request type: '' at
> /usr/local/bin/policyd-weight line 690, <GEN7> line 2.
> Jun 5 11:21:04 mailhost postfix/policyd-weight[13303]: child 13463 exited
This is normal and not the cause.
Currently I can think only of some situations where a "crash" without
logging could happen:
- Signals like KILL, INT, TERM, QUIT, PIPE (and other sigs which cause
termination)
are not logged. In any case - this wouldn't be normal behavior anyways.
If those signals arive at the master, then someone/something else must
trigger it.
SIGSEGV should appear in the kernel log.
You could grep for all master PIDs which every appeared in your logs:
On a FreeBSD with zsh and gnu-bzgrep I do it so:
p=`bzgrep "daemonized" /var/log/mail/maillog* | \
perl -pe 'while(<>){s/.*?\[(\d+)\].*/$1/s; push(@o, $_)} \
print "\\\(\\\[".join("\\\]\\\|\\\[",@o)."\\\]\\\)" '`; \
bzgrep -r -i -e "policyd-weight$p" /var/log/*
(This probably won't work with your system, I've tested it with zsh and
bash, unfortunately they don't behave the same everywhere).
You will find such stuff then, too:
Jun 12 18:29:14 <console.info> fpsvr1z150 kernel: Jun 12 18:29:14 <kern.info>
fpsvr1z150 kernel: pid 50216 (perl5.8.8), uid 0: exited on signal 11
(This was a manually crafted signal)
As you see, no "policyd-weight" relation, but I have found this via the
PID, as this was a master process. Works not for other signals though,
but you might want to give it a try.
also (to continue the secnarios)
- kernel buggy?
- syslog broken?
- perl broken?
After all, I would like to know whether this is in some form reproduceable.
I am on my way home now. Expect in some hours a version which tries to log
each and every signal caught. As a workaround you could also run policyd-weight
in master.cf mode.
--
Robert Felber (PGP: 896CF30B)
Munich, Germany
____________________________________________________________
Policyd-weight Mailinglist - http://www.policyd-weight.org/
