Re: [PATCH 02/14] resource limits: aggregate task highwater marks to cgroup level

From: Topi Miettinen <toiwoton@gmail.com>
To: Tejun Heo <tj@kernel.org>
Cc: linux-kernel@vger.kernel.org, Jonathan Corbet <corbet@lwn.net>,
	Li Zefan <lizefan@huawei.com>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Markus Elfring <elfring@users.sourceforge.net>,
	"David S. Miller" <davem@davemloft.net>,
	Nicolas Dichtel <nicolas.dichtel@6wind.com>,
	"open list:DOCUMENTATION" <linux-doc@vger.kernel.org>,
	"open list:CONTROL GROUP (CGROUP)" <cgroups@vger.kernel.org>
Subject: Re: [PATCH 02/14] resource limits: aggregate task highwater marks to cgroup level
Date: Tue, 19 Jul 2016 16:57:10 +0000	[thread overview]
Message-ID: <08d85be0-5ac7-d7ea-0fec-ceb3970f264b@gmail.com> (raw)
In-Reply-To: <20160718225249.GM3078@mtj.duckdns.org>

On 07/18/16 22:52, Tejun Heo wrote:
> On Fri, Jul 15, 2016 at 05:15:41PM +0000, Topi Miettinen wrote:
>> There are clear semantics for the limits themselves, either they apply
>> per task or per user. It makes sense to gather values according to these
>> semantics. Then with systemd or other tools you can use the valuse to
>> set the limits for a service regardless if the limit applies per task or
>> per user and it works according to each limit's semantics.
> 
> What does it mean to collect the maximum of the high watermarks of
> multiple users or the high water marks along process hierarchy which
> is spread across multiple cgroups?  These are non-sensical numbers.
> If you want to collect high watermarks per-cgroup, the numbers have to
> be per-cgroup - how many fds are being used in that particular cgroup
> and what's the high watermark of that number and so on.  You can't
> just take maximum from process hierarchy or user watermarks.

Then there would need to be new limit checks at cgroup level. Would you
see problems with that approach?

-Topi