🚀 go-pugleaf

Philip Yzarn de Louraille (yzarn@lhdsy1.lahabra.chevron.com) wrote:

+ >On Ultrix/RISC 4.2A we often find processes that 
+ >can't be killed (by root, "kill -9") - short of a reboot.

+ >Is this expected behaviour under some circumstances?

+ Yes, if the process has been swapped out to disk and is not recalled.
+ The problem is why isn't the process recalled?

	This occurs because the process has gone into a device wait for
some reason. It doesn't receive the signals until it comes out of the
kernel, which (for instance) is why the pagedaemon and swapper can't
be killed - they always have these little "D"'s beside them if you do a
"ps aux".
	Philip has pointed out one possible cause - there are a number of
ways of doing this, usually to do with NFS and servers being down,
broken automounters, something waiting for the close on a pipe (under some
systems, not sure about Ultrix).
	There are a huge number of fun things you can do with the mmap()
call under SunOS, (un?)fortunatly none of these seem to work with OSF/1.
	There are ways of getting rid of processes like this, however it
involves grubbing around inside the kernel - it is NOT advised and you'll
probably get the kernel to panic if you miss one small thing. Only do this
if you were going to bring the machine down anyway.

	Andrew.

/*
 * Andrew Rutherford                 andrewr@itd.adelaide.edu.au
 * +61 8 228 5669            Real Programmers always confuse Christmas and
 * Room 1060, Adelaide Uni      Halloween because OCT 31 == DEC 25 !
 */