Tracking down "kern.ipc.maxpipekva exceeded"

[Lowell Gilbert]

Ewald Jenisch <a at jenisch.at> writes:

> Hi,
> 
> For some time one of my machines in plaged be becoming completely
> unresponsive after different amounts of time (several hours up to
> several days).
> 
> Symptoms:
> Machine is PINGable, but no access over the network is possible
> (neither ssh-login nor http-access). 
> 
> /var/log/messages:
> May 11 12:11:00 io kernel: kern.ipc.maxpipekva exceeded; see tuning(7)
> maxproc limit exceeded by uid 0, please see tuning(7) and login.conf(5).
> (tons of both)
> and finally 
> May 11 12:42:03 io kernel: maxproc
> May 11 12:42:03 io kernel: limit exceede
> May 11 12:42:03 io kernel: d by uid
> May 11 12:42:03 io kernel: 0, ple
> May 11 12:42:03 io kernel: se see tun
> May 11 12:42:03 io kernel: ng(7) and l
> May 11 12:42:03 io kernel: ogin.conf
> May 11 12:42:03 io kernel: (5).
> May 11 12:42:03 io kernel: 
> 
> >From that time on I can't log in any more, not even from the console.
> 
> When rebooting the box via ctl-alt-del, before shutdown, it fails to
> sync buffers on shutdown (counts down to 1 and stays there) and
> finally gives up - requiring an fsck on reboot.
> 
> Please note that the box does *not* crash completely, nor do I get a
> kernel panic though.
> 
> The system is running "5.4-STABLE FreeBSD 5.4-STABLE #0: Sat Apr 16
> 10:19:26 CEST 2005" but this problem has been there since I first
> installed 5.3 on the box.
> 
> So my primary question is:
> 
> How do I track down the cause of the problem, i.e. which
> program/process is responsible for "kern.ipc.maxpipekva" to be
> exceeded?

I would suggest keeping an eye on kern.ipc.pipekva and trying to
correlate any changes to the activity on the system at the time.