Processes that won't die

[Devin Carraway]

On Mon, Nov 12, 2001 at 03:17:52PM -0800, E Frank Ball wrote:
> I'm using Kernel 2.2.12, but I'll consider updating (up2date errored out
> on me).  rmb is known for blocking kill signals in "protracted and

A SIGKILL can't be caught or ignored by a process -- that's more or less the
purpose.  SIGINT, TERM, HUP and friends can be, but the only thing that can
delay a signal 9 is if the process is blocked, generally during I/O (NFS tends
to incur this sort of thing.)  I/O shouldn't block for long, unless you have a
hardware issue (disk on a faulting bus, NFS on a noisy or disconnected
network, etc) going on.

Quite a few classes of kernel bugs could trap processes in D state.  Bring
the box to 2.2.20 and see if you can recreate it.  There are several
significant security vulnerbilities in 2.2.12 anyway.

> So if something is in a "D" state I'm hosed?
> (I'll look next time I get things good and stuck).

Any process that hits the disk enters D state (blocked) for a while.  That's
normal, but staying there in the absence of hardware causes isn't.  Check
dmesg to see if the kernel's reporting anything of that nature.


-- 
Devin  \ aqua(at)devin.com, 1024D/E9ABFCD2;  http://www.devin.com
Carraway \ IRC: Requiem  GCS/CC/L s-:--- !a !tv C++++$ ULB+++$ O+@ P L+++