Killing a process when kill -9 doesn't work?

[Darrin Chandler]

On Thu, Jul 05, 2007 at 08:09:54AM -0700, Dan Lund wrote:
> On 7/5/07, Matt Graham <danceswithcrows at usa.net> wrote:
> > On Thursday 05 July 2007 10:27, after a long battle with technology,
> > Shawn Badger wrote:
> > > how do you kill something when kill -9  doesn't work?
> >
> > You don't, generally.  SIGKILL will kill anything that isn't waiting on
> > a syscall to return.  If something is waiting on a syscall to return
> > for more than about 0.5 seconds, you've got a hardware problem, a
> > kernel bug, or a dead NFS server.  (There's currently a live thread on
> > comp.os.linux.misc about SIGKILL, oddly enough.)
> >
>
> No, it does happen with the rpm application rarely, and almost always
> its so honked up that a SIGKILL won't terminate it.
> I've seen it once or twice myself, and it almost always has to do with
> a corrupted rpm database.
> As far as killing it, that's something that I'd love to figure out....
> Though I've only ran into this on new installs when I'm applying
> patches.  I've always just figured itd take less man hours to
> re-install.

No, he's got it right. There's nothing that -9 won't kill unless the
kernel looks at it and says "oh, I'm not killing that!" The most common
form of this is a zombie process, which is a dead process that the
kernel keeps around waiting for the zombie's child to finish. Let the
child exit on it's own, or kill the child process, and then kernel will
(sooner or later) clean up the zombie itself.

-- 
Darrin Chandler            |  Phoenix BSD User Group  |  MetaBUG
dwchandler at stilyagin.com   |  http://phxbug.org/      |  http://metabug.org/
http://www.stilyagin.com/  |  Daemons in the Desert   |  Global BUG Federation