6.1: kern.ipc.maxpipekva

[Dan Nelson]

In the last episode (Jun 17), Marc G. Fournier said:
> On Sat, 17 Jun 2006, Dan Nelson wrote:
> >In the last episode (Jun 17), Marc G. Fournier said:
> >>On Sat, 17 Jun 2006, Marc G. Fournier wrote:
> >>>Jun 17 16:00:03 pluto kernel: kern.ipc.maxpipekva exceeded; see tuning(7)
> >>>Jun 17 16:00:04 pluto kernel: kern.ipc.maxpipekva exceeded; see tuning(7)
> >>>
> >>>but I can't seem to find anything in tuning(7) about it ... so,
> >>>what is it and how do I monitor for it?
> >>
> >>More on this:
> >>
> >># sysctl -a | grep pipekva
> >>kern.ipc.maxpipekva: 16777216
> >>kern.ipc.pipekva: 15122432
> >>
> >>and I just rebooted the server ...
> >>
> >>so obviously I've been living on the edge ... not sure what to
> >>increase it to, since not sure what it affects, so will wait on
> >>responses ...
> >
> >Try also running "sysctl kern.ipc | grep pipe", which will also tell
> >you how many pipes are in use, plus some other counters.  The
> >comment at the top of sys/kern/sys_pipe.c explains how pipes are
> >given memory.
> 
> What uses all of these pipes?  right now, with 97 jails running:
> 
> kern.ipc.maxpipekva: 25165824
> kern.ipc.pipes: 7038
> kern.ipc.pipekva: 22179840
> kern.ipc.pipefragretry: 0
> kern.ipc.pipeallocfail: 0
> kern.ipc.piperesizefail: 0
> kern.ipc.piperesizeallowed: 1
> 
> That is an average of 7 pipes per process:
> 
> pluto# ps aux | wc -l
>     1326

"fstat | grep pipe" will tell you what processes have them open on what
fds.  pipes on fds 0, 1 and 2 are probably from shell pipelines.

-- 
	Dan Nelson
	dnelson at allantgroup.com