[Pkg-openldap-devel] Bug#689025: Re: Bug#689025: slapd: chokes on unresponsive syslogd

[Ryan Tandy]

tags 689025 + moreinfo
thanks

Hi Dominik,

On 28/09/12 01:27 PM, Dominik George wrote:
>> A full GDB backtrace of all threads would be useful for examining this issue
>> any further.
>
> kI will have to build a test case for that. Obviously, I cannot break
> production deliberately for that ;).

Did you ever get a chance to set up a test case for this? I'm trying to 
reproduce it tonight, and not having a lot of luck. I'm in a wheezy 
chroot, with rsyslog logging to mysql over tcp (made sure with 
lsof/netstat), and slapd logging at stats level.

First, while rsyslog does block once its queue fills, I don't see slapd 
totally hanging, only becoming very slow to respond. Attaching with gdb 
(sadly ltrace in wheezy fails silently on PIE binaries) I see several 
threads in the middle of syslog() calls, which I guess are eventually 
timing out.

Next, and probably this is because I didn't actually manage to reproduce 
your hang, but when the database server becomes available again, for me 
slapd does recover every time, after rsyslog has noticed and resumed 
logging.

Without more information, at least some details about a setup that 
reproduces the problem, I don't know how else to help with this. Ideally 
we need a backtrace showing, for example, whether your slapd is stuck in 
syslog() or in slapd itself.

thanks,
Ryan