[lkml]   [2011]   [Oct]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [RFD] Isolated memory cgroups again
On 10/20/2011 05:33 AM, Michal Hocko wrote:
> Hi all,
> this is a request for discussion (I hope we can touch this during memcg
> meeting during the upcoming KS). I have brought this up earlier this
> year before LSF (
> The patch got much smaller since then due to excellent Johannes' memcg
> naturalization work (
> which this is based on.
> I realize that this will be controversial but I would like to hear
> whether this is strictly no-go or whether we can go that direction (the
> implementation might differ of course).
> The patch is still half baked but I guess it should be sufficient to
> show what I am trying to achieve.
> The basic idea is that memcgs would get a new attribute (isolated) which
> would control whether that group should be considered during global
> reclaim.

I'd like to hear a bit more of your use cases, but at first, I don't
like it. I think we should always, regardless of any knobs or
definitions, be able to globally select a task or set of tasks, and kill

We have a slightly similar need here (we'd have to find out how
similar...). We're working on it as well, but no patches yet (very
basic) Let me describe it so we can see if it fits.

The main concern is with OOM behaviour of tasks within a cgroup. We'd
like to be able to, in a per-cgroup basis:

* select how "important" a group is. OOM should try to kill less
important memory hogs first (but note: it's less important *memory
hogs*, not ordinary processes, and all of them are actually considered)
* select if a fat task within a group should be OOMed, or if the whole
group should go.
* assuming an hierarchical grouping, select if we should kill children
* assuming an hierarchical grouping, select if we should kill children
at all.

This is a broader work, but I am under the impression that you should
also be able to contemplate your needs (at least the OOM part) with such
mechanism, by setting arbitrarily high limits on certain cgroups.

Of course it might be the case that I am not yet fully understanding
your scenario. In this case, I'm all ears!

Thank you.

 \ /
  Last update: 2011-10-20 10:59    [W:0.087 / U:3.748 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site