Re: [patch 4/7 -mm] oom: badness heuristic rewrite

[David Rientjes]

On Wed, 17 Feb 2010, Minchan Kim wrote:

> >> Okay. I can think it of slight penalization in this patch.
> >> But in current OOM logic, we try to kill child instead of forkbomb
> >> itself. My concern was that.
> >
> > We still do with my rewrite, that is handled in oom_kill_process(). ÂThe
> > forkbomb penalization takes place in badness().
> 
> 
> I thought this patch is closely related to [patch  2/7].
> I can move this discussion to [patch 2/7] if you want.
> Another guys already pointed out why we care child.
> 

We have _always_ tried to kill a child of the selected task first if it 
has a seperate address space, patch 2 doesn't change that.  It simply 
tries to kill the child with the highest badness() score.

> I said this scenario is BUGGY forkbomb process. It will fork + exec continuously
> if it isn't killed. How does user intervene to fix the system?
> System was almost hang due to unresponsive.
> 

The user would need to kill the parent if it should be killed.  The 
unresponsiveness in this example, however, is not a question of the oom 
killer but rather the scheduler to provide interactivity to the user in 
forkbomb scenarios.  The oom killer should not create a policy that 
unfairly biases tasks that fork a large number of tasks, however, to 
provide interactivity since that task may be a vital system resource.

> For extreme example,
> User is writing some important document by OpenOffice and
> he decided to execute hackbench 1000000 process 1000000.
> 
> Could user save his important office data without halt if we kill
> child continuously?
> I think this scenario can be happened enough if the user didn't know
> parameter of hackbench.
> 

So what exactly are you proposing we do in the oom killer to distinguish 
between a user's mistake and a vital system resource?  I'm personally much 
more concerned with protecting system daemons that provide a service under 
heavyload than protecting against forkbombs in the oom killer.