[Devel] Re: [PATCH v2 0/5] per-cgroup /proc/stat statistics

[Glauber Costa]

On 04/10/2012 02:25 AM, Glauber Costa wrote:
> Hi,
>
> This patch aims at allowing userspace to recreate the most important
> contents of /proc/stat per-cgroup. It exports the data needed for it
> from the guts of the scheduler, and then anyone can parse it and
> present it to a container in a meaningful way. Again, the kernel won't
> get involved in this directly.
>
> Part of it will come from the cpu cgroup. Another part, from the cpuacct
> cgroup. Data is exported in cgroup files stat_percpu. They are just like
> the normal stat files, but with a cpuXXX value before the actual data
> field. As so, they are also extensible. So if anyone wants to give a
> shot at values I am currently ignoring (as iowait) in the future, we
> at least won't have a format problem.
>
> Let me know what you think.
>
> Glauber Costa (5):
>    measure exec_clock for rt sched entities
>    account guest time per-cgroup as well.
>    record nr_switches per task_group
>    expose fine-grained per-cpu data for cpuacct stats
>    expose per-taskgroup schedstats in cgroup
>
>   kernel/sched/core.c  |  214 ++++++++++++++++++++++++++++++++++++++++++++++++--
>   kernel/sched/fair.c  |   27 ++++++-
>   kernel/sched/rt.c    |    5 +
>   kernel/sched/sched.h |    6 ++
>   4 files changed, 245 insertions(+), 7 deletions(-)
>
Paul and the other scheduler folks:

Do you have a saying on this?

Thanks