[syslog-ng] syslog-ng fails to drop file handles and reopen on HUP signal

[Balazs Scheidler]

On Thu, 2011-11-17 at 10:36 -0800, Dave Rawks wrote:
> Since 3.3.2 I've been seeing this weird behavior where syslog-ng fails
> to reload cleanly on a HUP. This has been particularly hard to
> troubleshoot since it appears as though when the reload fails it is
> causing both sshd and login to block; which I suspect is because
> /dev/log has disappeared. Luckily I do have nagios nrpe setup to allow
> me to remotely restart syslog-ng which alleviates the inability to
> login, but also makes it very hard to collect any troubleshooting
> information.
> 
> Anyone else seeing any failures like this?
> 
> I also sometimes see a a kernel error like one of the following on reload:
> Nov 17 10:30:02 app-bogus kernel: [ 3788.489852] syslog-ng[28520]:
> segfault at 7fd054022350 ip 00007fd054022350 sp 00007fffc93404e8 error 15
> 
> OR
> 
> Nov 17 10:32:35 app-bogus kernel: [ 3941.605005] syslog-ng[29805]
> general protection ip:7fd05f0bcc4e sp:7fd05ac71240 error:0 in
> libsyslog-ng-3.3.2.so[7fd05f092000+7b000]
> 
> I was really excited about the features this version of syslog-ng has
> brought with it, but I'm finding it WAY to unstable to use in production.
> 
> Is there any information I could provide to help resolve these possibly
> related issues?

Yup, Patrick has also found this issue, and I posted a fix in a separate
thread. One of the last fixes going into 3.3.2 was flawed, and I didn't
notice. 

I'm working on 3.3.3 with just that fix.

In fact, it is being uploaded to our webserver (syncs to the public
every 3 hours), grab it while it's hot.

For the impatient, the patch is available at github.

-- 
Bazsi