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

[Topi Miettinen <toiwoton@gmail.com>]

On 07/15/16 14:10, Tejun Heo wrote:
> Hello, Topi.
> 
> On Fri, Jul 15, 2016 at 01:35:49PM +0300, Topi Miettinen wrote:
>> Collect resource usage highwater marks of a task to cgroup
>> statistics when the task exits.
> 
> I'm not sure how this makes sense.  The limits are enforced and
> collected per user or along the process hierarchy which can be very
> different from cgroup organization.  What does collecting high
> watermarks from orthogonal structure, sometimes even combining
> per-user numbers from different users, even mean?  These are numbers
> without clear semantics.

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.

cgroups are used to aggregate values from a group of tasks, which still
are related to one service. Because with systemd the services also are
given a cgroup context, the values will completely make sense there too.

It could be useful to introduce a new set of limits that apply only
cgroup level. It would not remove the need to aggregate values from a
group of tasks.

-Topi

> 
> Thanks.
>