Message-ID: <xr93a9xmwly7.fsf@gthelen.mtv.corp.google.com>
Date:	Wed, 22 Aug 2012 16:23:12 -0700
From:	Greg Thelen <gthelen@...gle.com>
To:	Glauber Costa <glommer@...allels.com>
Cc:	Michal Hocko <mhocko@...e.cz>, <linux-kernel@...r.kernel.org>,
	<linux-mm@...ck.org>, <cgroups@...r.kernel.org>,
	<devel@...nvz.org>, Johannes Weiner <hannes@...xchg.org>,
	Andrew Morton <akpm@...ux-foundation.org>,
	<kamezawa.hiroyu@...fujitsu.com>, Christoph Lameter <cl@...ux.com>,
	David Rientjes <rientjes@...gle.com>,
	Pekka Enberg <penberg@...nel.org>,
	Pekka Enberg <penberg@...helsinki.fi>,
	Suleiman Souhlal <suleiman@...gle.com>
Subject: Re: [PATCH v2 09/11] memcg: propagate kmem limiting information to children

On Wed, Aug 22 2012, Glauber Costa wrote:

>>>>
>>>> I am fine with either, I just need a clear sign from you guys so I don't
>>>> keep deimplementing and reimplementing this forever.
>>>
>>> I would be for make it simple now and go with additional features later
>>> when there is a demand for them. Maybe we will have runtimg switch for
>>> user memory accounting as well one day.
>>>
>>> But let's see what others think?
>> 
>> In my use case memcg will either be disable or (enabled and kmem
>> limiting enabled).
>> 
>> I'm not sure I follow the discussion about history.  Are we saying that
>> once a kmem limit is set then kmem will be accounted/charged to memcg.
>> Is this discussion about the static branches/etc that are autotuned the
>> first time is enabled?  
>
> No, the question is about when you unlimit a former kmem-limited memcg.
>
>> The first time its set there parts of the system
>> will be adjusted in such a way that may impose a performance overhead
>> (static branches, etc).  Thereafter the performance cannot be regained
>> without a reboot.  This makes sense to me.  Are we saying that
>> kmem.limit_in_bytes will have three states?
>
> It is not about performance, about interface.
>
> Michal says that once a particular memcg was kmem-limited, it will keep
> accounting pages, even if you make it unlimited. The limits won't be
> enforced, for sure - there is no limit, but pages will still be accounted.
>
> This simplifies the code galore, but I worry about the interface: A
> person looking at the current status of the files only, without
> knowledge of past history, can't tell if allocations will be tracked or not.

In the current patch set we've conflating enabling kmem accounting with
the kmem limit value (RESOURCE_MAX=disabled, all_other_values=enabled).

I see no problem with simpling the kernel code with the requirement that
once a particular memcg enables kmem accounting that it cannot be
disabled for that memcg.

The only question is the user space interface.  Two options spring to
mind:
a) Close to current code.  Once kmem.limit_in_bytes is set to
   non-RESOURCE_MAX, then kmem accounting is enabled and cannot be
   disabled.  Therefore the limit cannot be set to RESOURCE_MAX
   thereafter.  The largest value would be something like
   RESOURCE_MAX-PAGE_SIZE.  An admin wondering if kmem is enabled only
   has to cat kmem.limit_in_bytes - if it's less than RESOURCE_MAX, then
   kmem is enabled.

b) Or, if we could introduce a separate sticky kmem.enabled file.  Once
   set it could not be unset.  Kmem accounting would only be enabled if
   kmem.enabled=1.

I think (b) is clearer.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists