Re: Immortal processes [was: mc not loading and not dying (sid)]
On Sunday 08 January 2006 07:19 pm, Jan C. Nordholz wrote:
> On Sun, Jan 08, 2006 at 04:06:01PM -0800, Andrew Sackville-West wrote:
> > On Sun, 08 Jan 2006 15:32:11 +0200
> > Linas Zvirblis <firstname.lastname@example.org> wrote:
> > > By the way, there is no such thing as unkillable, "killall -9
> > > mc" and "killall -9 acroread" should get rid of them.
> > I have lately encountered processes that don't respond to -9. I think it
> > was an automount process. Regardless, what's up with that? Is there any
> > other way to kill processes that don't respond to -9 (aside from of
> > course, the dreaded reboot...)?
> Yes, such a thing exists: Processes that get locked up in
> kernel I/O. They will show up in 'ps' output with state 'D',
> and won't respond to _anything_ until the kernel function
> returns - and if the hardware itself is locked up, all you
> can do is reboot. Scsi tapes are infamous for doing such
> things, and NFS is, depending on the mount options used,
> very good at doing the same.
What are options to use in NFS to avoid that? I've had problems before where
I had to reboot the NFS server and NFS on the clients would not let me kill
it or remount the shares. I'd have to reboot the client to be able to
remount the shares. (And rebooting always seems like such a pain -- I guess
when you hardly ever do something, it's a big deal when you have to!)