memcg creates an unkillable task in 3.11-rc2

Michal Hocko mhocko at suse.cz
Mon Sep 9 08:31:47 UTC 2013


On Fri 06-09-13 11:09:21, Eric W. Biederman wrote:
> Michal Hocko <mhocko at suse.cz> writes:
> 
> > It seems that this one fell though the cracks?
> 
> Not completely, but it happened just as I was doing my initial triage of
> memcg problems and I haven't quite made it back to this.

OK. I am primarily asking because I am not sure I understood whether the
proposed patch helped or not. It should as per my last email but it is
possible that I have missed something...

> I have an even nastier memcg hang (without yet an easy reproducer).
> During mkdir ext3 can add a page to the page cache with the ext3 journal
> transaction lock held.  Normally that isn't a problem but freezing there
> stops all writes to that filesystem, and the world stops.
> 
> It looks like the only way to avoid that kind of scenario is to move the
> the memcg sleep to the edge of userspace, like we do with signals and a
> few other things so we can be guaranteed not to increase lock hold
> times, when it is avoidable.  I think I saw some similar comments about
> the slab limiting.

Johannes has patches that move memcg oom out of any locks already: 
https://lkml.org/lkml/2013/8/3/81. There is some development going on
there and I guess he will post v3 soon but that sounds it might help
with what you describe above.

Thanks!
-- 
Michal Hocko
SUSE Labs


More information about the Containers mailing list