On Fri, Apr 19, 2002 at 12:24:45PM -0400, der Mouse wrote:
> >> That doesn't sound plausible to me.  When the queues back up, log
> >> messages usually get dropped, rather than blocking the application.
> > Yes, kernel messages get dropped, but if syslogd makes a write(2) to
> > /dev/console, is it smart enough to time this out and or fail
> > gracefully if the console is "stuck"?
> 
> syslogd may hang.  But that's no reason for (say) telnetd to hang, or
> fail to start.  Even if telnetd is trying to log something, it's just
> doing a send() to an AF_LOCAL socket (/dev/log).  Even processes that
> specify LOG_CONS shouldn't hang until the /dev/log buffer fills up, and
> I'm not sure about even then.

Yes, and that's the problem: when syslog is hung on write(/dev/console),
/dev/log buffer is full after some time. And a lot of things use
syslog, even if their message don't end up in a file with the normal
config (LOG_DEBUG isn't logged for example).

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--