How do you know that the processes represented in the 'w' column are in fact dead processes? This bring me back to my original question: What state is the process in? (i.e. how does it show up in a "ps -f" listing?)

Do you have a reproducible test case to demonstrate this?

Regards,
Brian



Mike DeMarco wrote:
From: Brian Ruthven - Sun UK <[email protected]>
To: Mike DeMarco <[email protected]>
Cc: [email protected]
Sent: Wed, October 14, 2009 10:58:06 AM
Subject: Re: [osol-discuss] swap purge


What state is the process in? How do you know it is swapped out? What problem 
is this causing?
If the parent crashed, then init should inherit the child as its own, and it 
AFAIK does reap dead children periodically.
If the process shows up as <defunct> because the parent is not reaping it, then 
there should be no memory associated with it any longer (actually, there is enough to 
give the hint a process is still there, but the address space should have already 
been reclaimed, and what's left of the process now uses the kernel's address space 
until the parent can be notified).

If you really want to force it into memory, then temporarily removing your swap 
device and re-adding it is one way of trying this, but may fail if you actually 
need the swap space. It's also a very heavy handed way of doing it  ;-)

Regards,
Brian


Whenever we have had systems go deep into swap some software will fail and the 
processes that were swapped out will die. When this happens the swapped out 
process will remain in swap space, there is no pid to get back on the stack to 
pull the process back out of the swap abyss. So the swap space remains consumed 
until the next reboot. The kernel must know that the process is in swap since 
it swapped it out and is keeping track of it as shown by the w column of 
vmstat. There has to be a way to message this iformation in the kernel and tell 
the kernel to forget about that swapped out process it is never coming back. 
This would help my by allowing me to cleanup the disk space that is being 
consumed by swap instead of allocating more disk spaceto swap that is of no use.

I have also found that unless the pid can pull back the swap space when swap -d 
is called the swap -d will fail since the kernel can not clear the swap file 
out fully to delete it.

--
Brian Ruthven
Solaris Revenue Product Engineering
Sun Microsystems UK
Sparc House, Guillemont Park, Camberley, GU17 9QG

_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to