Re: [PATCH RFC] mm/memcg: recalculate chargeable space after waiting migrating charges

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu 12-07-12 19:51:25, Wanpeng Li wrote:
> On Thu, Jul 12, 2012 at 01:08:38PM +0200, Michal Hocko wrote:
> >On Thu 12-07-12 18:39:21, Wanpeng Li wrote:
> >> From: Wanpeng Li <liwp@xxxxxxxxxxxxxxxxxx>
> >> 
> >> Function mem_cgroup_do_charge will call mem_cgroup_reclaim,
> >> there are two break points in mem_cgroup_reclaim:
> >> if (total && (flag & MEM_CGROUP_RECLAIM_SHIRINK))
> >> 	break;
> >> if (mem_cgroup_margin(memcg))
> >> 	break;
> >> so mem_cgroup_reclaim can't guarantee reclaim enough pages(nr_pages) 
> >> which is requested from mem_cgroup_do_charge, if mem_cgroup_margin
> >> (mem_over_limit) >= nr_pages is not true, the process will go to
> >> mem_cgroup_wait_acct_move to wait doubly charge counted caused by
> >> task move. 
> >
> >I am sorry but I have no idea what you are trying to say. The
> >mem_cgroup_wait_acct_move just makes sure that we are waiting until
> >charge is moved (which can potentially free some charges) rather than
> >OOM which should be the last resort so it makes sense to retry them
> >charge.
> >
> >> But this time still can't guarantee enough pages(nr_pages) is
> >> ready, directly return CHARGE_RETRY is incorret. 
> >
> >So you think it is better to oom? Why? What prevents you from a race
> >that your mem_cgroup_margin returns true but another CPU consumes those
> >charges right after that. See? The check is pointless. It doesn't
> 
> Hmm, if there are a race as you mentioned it can't guarantee enough pages 
> is ready. 

And there is no point in guaranteeing anything which I tried to tell you
by the example... The only thing that matters is whether we get the charge
on the next attempt and if not whether we are able to reclaim something.
See?

> But it also means that available memory is too low if this
> race happen. If available charges still less than nr_pages
> after mem_cgroup_wait_acct_move(which can potentially
> free some charges) return, the CHAGE_RETRY will trigged,
> and then mem_cgroup_do_charge=>meory_cgroup_reclaim
> =>mem_cgroup_wait_acct_move, if available charges still less than
> nr_pages in this round, CHAGE_RETRY.....

> To avoid this infinite retry when available memory 

I do not see a realistic scenario which would cause this to be infinite loop
withou OOM jumping in.
We would have to hit the wait for move after each reclaim and the move would
have to keep the the usage constant (move is really fast without moving
charges).
So what you are trying to address (if I understand it at all) is to fix
an almost impossible to trigger issue with a bogus change which doesn't
help at all because it is racy as well.

> in this memcg is very low, go to OOM if mem_cgroup_margin(mem_over_limit) 
> < nr_pages is a better way I think. Because the codes have already try
> its best to reclaim some pages. :-)


> 
[...]
-- 
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9    
Czech Republic

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>


[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]