Re: [PATCH 6/7] [RFC] Support multiply-bindable cgroup subsystems

[Li Zefan <lizf-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org>]

CC: Balbir Singh

Paul Menage wrote:
> On Tue, Mar 17, 2009 at 7:39 PM, Li Zefan <lizf-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org> wrote:
>>> - always display one line for each subsystem; if a subsystem is
>>> multiply-bindable then the "hierarchy id" and "num cgroups" columns
>>> may be empty or have multiple (comma or slash-separated?) values
>>>
>> Then "hierarchy id" is no long a single number..
> 
> Correct.
> 
>>> - for a multiply-bindable subsystem, have a header line to indicate
>>> that the subsystem exists, and then a separate line for each bound
>>> instance of the subsystem.
>>>
>> I think it's better to show every subsystems including debug_subsys in
>> /proc/cgroups, and show exactly n lines of debug_susys if we have n
>> hierarcies with debug_subsys binded, but no header line.
> 
> But that gives a contradiction when n == 0 - we can't show exactly 0
> lines for an unbound multi subsys, and still show every subsystem.
> 

I don't see what's wrong with this behavior:

multi subsys sits in rootnode if it's unbound, and is removed from
rootnode if it's binded at least in one hierarchy.

> Does libcgroup actually parse /proc/cgroup? If not, maybe we should

Balbir may answer this. :)

> just break the format now and replace it with something more
> extensible for future changes.
> 

Even /proc/cgroup is ok, I don't think we can break this based on assumption
that on one is making use of /proc/cgroups. :(

>> And some cgroup test programs, like controller tests in ltp and some
>> test programs that I wrote, but they are ok for this change.
> 
> Right, changes to test programs aren't a compatibility issue.
>