[syslog-ng]sysnlog-ng lock whole machine ...

Nate Campi syslog-ng@lists.balabit.hu
Wed, 17 Mar 2004 08:25:57 -0800


On Wed, Mar 17, 2004 at 02:32:49PM +0100, Pawel Dziekonski wrote:
> 
> 1st: please confirm how do I log kernel messages (assuming klogd is NOT
> running):

I use this         pipe("/proc/kmsg"); 

> 2nd: the machine "hangs" from time to time - cron is hanging, login is
> not spawning shell, sshd too. i have pinpointed it so i'm pretty sure
> that all those processes hang on socket connection to /dev/log. killing
> syslog-ng cures that.

I am also a debian user, I don't know if this problem is specific to
debian or not, but I used to have lockups all the time when two things
would happen at once: syslog log rotation and postfix reloads. The HUP
of syslog-ng when rotating logs and postfix writing to /dev/log would
hang up the whole machine, cron processes would pile up and eventually
you couldn't even log into the machine, presumably because syscalls for
logging would hang. If I already had a shell and killed postfix most of
the processes would un-freeze, but cron would never recover. I always
had to reboot or power cycle if init got screwed.

I honestly can't remember what I did, I think I made sure syslog-ng was
HUP'd only once, at the end of rotating all the logs, and maybe reorder
any cron.daily or cron.weekly jobs that HUP postfix until well after log
rotation.

Also make remove postfix and syslog-ng start/stop/reload from
/etc/ppp/ip-up.d/ and /etc/ppp/ip-down.d/ if you use pppd, unless you
really need them there. I've had those cause the same freezes, when use
ppp for VPNs.
-- 
Nate

Corporations are not evil. That kind of anthropomorphism is inappropriate.
Corporations are too stupid to be evil, only people can be that.  -jwz