BETA3: network processes hang in keglimit (unkillable)

[Pieter de Goeje]

On Friday 11 September 2009, Pyun YongHyeon wrote:
> On Fri, Sep 11, 2009 at 07:24:10PM +0200, Pieter de Goeje wrote:
> > I just experienced a complete loss of network, where every network
> > related process would sleep on "keglimit". I was unable to recover from
> > this situation. I tried killing the processes with SIGKILL, but it didn't
> > work, and I tried to bring the network interface (em) down and up again,
> > which also didn't work. A reboot "solved" the problem. The machine was
> > handling some major traffic, both TCP and UDP.
> >
> > Apparently the system was waiting on some resource to free up, but what
> > could it be? How can this situation be avoided in the future?
> >
> > uname: 8.0-BETA3 FreeBSD 8.0-BETA3 #1: Sun Aug 30 01:23:36 CEST 2009
> > amd64
>
> Both em(4) and igb(4) had mbuf leak bug which was recently fixed by
> Jack. Check mbuf statistics with "netstat -m" and see whether
> you've reached mbuf resource limit(see 4K jumbo cluster counter).
> The leak can be easily seen on UDP traffic, especially NFS over UDP.

Ok, that matches my usage (NFS over UDP), so I will monitor mbuf usage and 
test Jack's fixes when they're MFCed (or sooner if I can reproduce this).

Thank you for your reply,

Pieter de Goeje