Re: [PATCH v3 4/5] sched/fair: Task based throttle time accounting

[Aaron Lu <ziqianlu@bytedance.com>]

Hi Michal,

Thanks for taking a look.

On Tue, Aug 26, 2025 at 04:10:37PM +0200, Michal Koutný wrote:
> Hello.
> 
> On Tue, Aug 19, 2025 at 05:34:27PM +0800, Aaron Lu <ziqianlu@bytedance.com> wrote:
> > Got it, does the below added words make this clear?
> > 
> >     With task based throttle model, the previous way to check cfs_rq's
> >     nr_queued to decide if throttled time should be accounted doesn't work
> >     as expected, e.g. when a cfs_rq which has a single task is throttled,
> >     that task could later block in kernel mode instead of being dequeued on
> >     limbo list and account this as throttled time is not accurate.
> > 
> >     Rework throttle time accounting for a cfs_rq as follows:
> >     - start accounting when the first task gets throttled in its hierarchy;
> >     - stop accounting on unthrottle.
> > 
> >     Note that there will be a time gap between when a cfs_rq is throttled
> >     and when a task in its hierarchy is actually throttled. This accounting
> >     mechanism only started accounting in the latter case.
> 
> Do I understand it correctly that this rework doesn't change the
> cumulative amount of throttled_time in cpu.stat.local but the value gets
> updated only later?
> 
> I'd say such little shifts are OK [1]. What should be avoided is
> changing the semantics so that throttled_time time would scale with the
> number of tasks inside the cgroup (assuming a single cfs_rq, i.e. number
> of tasks on the cfs_rq).

As Valetin explained, throttle_time does not scale with the number of
tasks inside the cgroup.

> [1] Maybe not even shifts -- in that case of a cfs_rq with a task, it
> can manage to run in kernel almost for the whole period, so it gets
> dequeued on return to userspace only to be re-enqueued when its cfs_rq
> is unthrottled. It apparently escaped throttling, so the reported
> throttled_time would be rightfully lower.

Right, in this case, the throttle_time would be very small.