no respond to KILL -9 or any signal

[Kris Kennaway]

On Tue, May 08, 2007 at 08:39:57PM +0200, Niels Dettenbach wrote:
> Dear List readers and FreeBSD coders,
> 
> since 5.2 i experienced in timespans of some days or weeks the effect on our 
> internet server machine, that processes got hanging and are not killable by 
> kill -9 or any other signal.
> 
> Most times it seems that perl based processes are affected like spamd 
> (Spamassassin), amavisd-new but one time my exim got hanged too.
> 
> ps afux says the proces are in state "DL" - wich doesn't change after a "kill" 
> or "kill -9 " - i didn't found any way to get the procs killed or cleaned 
> out.
> 
> So i have to reboot the machine (with hard reset!) each time to get the 
> hanging service running again - that's really not nice for a productive 
> system!
> 
> The Machine started with a 5.0 installation years ago. The effect arises since 
> somewhere around 5.1/5.2 - during my 6.0 installation not over several 
> monthes - since updates / with new fresh install on 6.2 it comes back. 
> Ideas - it comes from a hardware bug - i can't follow. We have checked the 
> hardware different times and the problem was gone over monthes during a 
> (untouched) 6.0 installation. With Linux the box the problem is not there but 
> im not like to switch to linux - i like FreeBSD in its way for my productive 
> systems.
> 
> We tried different installations and kernel configs and a lot of new world 
> build and port updates. It looks like a problem somewhere with any kind of IO 
> interactions but im not shure...
> 
> i use:
> ===
> - FreeBSD ffm-dr1.syndicat.com 6.2-RELEASE-p3 FreeBSD 6.2-RELEASE-p3 #0:
> - on a Intel Dual XEON 
> - Intel SE Server Board
> - 2GB RAM 
> - with Intel (former ICP-Vortex) U320 SCSI Controller (RAID1) - (iir driver)
> - IDE disc drive (backup only)
> - 2x NIC
> - with and without jails
> 
> env:
> ===
> CPUTYPE=p4
> CPUTYPE?=p4
> 
> Perl is 5.8.8 - i've tried perl from ports as the default FreeBSD perl.
> 
> (add. data available on request)
> 
> The problem is urgent for us so any useful info or help is very welcome!...

When this happens you need to break to DDB and obtain backtraces of
the hung processes.  Other debugging options and commands are also
possibly required; see the developers handbook.

Kris