Re: [RFC] [PATCH] Cgroup based OOM killer controller

[Evgeniy Polyakov <zbr@ioremap.net>]

Hi David.

On Tue, Jan 27, 2009 at 01:37:55AM -0800, David Rientjes (rientjes@google.com) wrote:
> > /dev/mem_notify is a great idea, but please do not limit existing
> > oom-killer in its ability to do the job and do not rely on application's
> > ability to send a SIGKILL which will not kill tasks in unkillable state
> > contrary to oom-killer.
> > 
> 
> You're missing the point, /dev/mem_notify would notify userspace of lowmem 
> situations and allow it to respond appropriately in any number of ways 
> before an oom condition exists.

Yes, I know.

> When the system (or cpuset, memory controller, etc) is oom, userspace can 
> choose to defer to the oom killer so that it may kill a task that would 
> most likely lead to future memory freeing with access to memory reserves.
> 
> There is no additional oom killer limitation imposed here, nor can the oom 
> killer kill a task hung in D state any better than userspace.

Well, oom-killer can, since it drops unkillable state from the process
mask, that may be not enough though, but it tries more than userspace.

My main point was to haev a way to monitor memory usage and that any
process could tune own behaviour according to that information. Which is
not realated to the system oom-killer at all. Thus /dev/mem_notify is
interested first (and only the first) as a memory usage notification
interface and not a way to invoke any kind of 'soft' oom-killer.
Application can do whatever it wants of course including killing itself
or the neighbours, but this should not be forced as a usage policy.

-- 
	Evgeniy Polyakov