[16298] in Athena Bugs
Re: sun4 8.2.9: syslogd is six feet under?
daemon@ATHENA.MIT.EDU (Greg Hudson)
Fri Sep 4 15:45:18 1998
To: Dan Winship <danw@MIT.EDU>
Cc: Greg Hudson <ghudson@MIT.EDU>, Jonathon Weiss <jweiss@MIT.EDU>,
John Hawkinson <jhawk@MIT.EDU>, bugs@MIT.EDU
In-Reply-To: Your message of "Fri, 04 Sep 1998 15:27:52 EDT."
<199809041927.PAA10096@zork.mit.edu>
Date: Fri, 04 Sep 1998 15:45:10 EDT
From: Greg Hudson <ghudson@MIT.EDU>
> You mean you don't think that's the cause or you don't think it's
> there? (When you HUP it, it re-execs itself, if you then HUPped it
> again after it has restarted but before it has had a chance to set
> up its handlers, it should die, right?)
Oh, that is interesting. I didn't look at the HUP code. Presumably
execv() will reset the signal mask set by pthread_sigmask().
It's interesting that the machines have no syslog.pid files (unless
jhawk is confused and actually did look for /etc/syslogd.pid instead
of /etc/syslog.pid). If Dan's race is at fault, then newsyslogd would
have to be waiting five seconds between the unlink of PidFile and the
execv of syslogd to send the signal (there's a sleep(5) in between).
Believable on a dialup, but not really on a workstation.
I guess the next step is to get truss output from newyslog from when
this event happens, so we can figure out what the stimulus is.
