Re: user defined OOM policies

[Michal Hocko]

On Mon 25-11-13 17:29:20, David Rientjes wrote:
> On Wed, 20 Nov 2013, Luigi Semenzato wrote:
> 
> > Yes, I agree that we can't always prevent OOM situations, and in fact
> > we tolerate OOM kills, although they have a worse impact on the users
> > than controlled freeing does.
> > 
> 
> If the controlled freeing is able to actually free memory in time before 
> hitting an oom condition, it should work pretty well.  That ability is 
> seems to be highly dependent on sane thresholds for indvidual applications 
> and I'm afraid we can never positively ensure that we wakeup and are able 
> to free memory in time to avoid the oom condition.
> 
> > Well OK here it goes.  I hate to be a party-pooper, but the notion of
> > a user-level OOM-handler scares me a bit for various reasons.
> > 
> > 1. Our custom notifier sends low-memory warnings well ahead of memory
> > depletion.  If we don't have enough time to free memory then, what can
> > the last-minute OOM handler do?
> > 
> 
> The userspace oom handler doesn't necessarily guarantee that you can do 
> memory freeing, our usecase wants to do a priority-based oom killing that 
> is different from the kernel oom killer based on rss.  To do that, you 
> only really need to read certain proc files and you can do killing based 
> on uptime, for example.  You can also do a hierarchical traversal of 
> memcgs based on a priority.
> 
> We already have hooks in the kernel oom killer, things like 
> /proc/sys/vm/oom_kill_allocating_task

How would you implement oom_kill_allocating_task in userspace? You do
not have any context on who is currently allocating or would you rely on
reading /proc/*/stack to grep for allocation functions?

> and /proc/sys/vm/panic_on_oom that 
[...]
-- 
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/