Re: [tip:sched/core] sched/numa: Delay retrying placement for automatic NUMA balance after wake_affine()

[Srikar Dronamraju]

* Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> [2018-05-09 09:41:48]:

> On Mon, May 07, 2018 at 04:06:07AM -0700, Srikar Dronamraju wrote:
> > > @@ -1876,7 +1877,18 @@ static void numa_migrate_preferred(struct task_struct *p)
> > >
> > >  	/* Periodically retry migrating the task to the preferred node */
> > >  	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
> > > -	p->numa_migrate_retry = jiffies + interval;
> > > +	numa_migrate_retry = jiffies + interval;
> > > +
> > > +	/*
> > > +	 * Check that the new retry threshold is after the current one. If
> > > +	 * the retry is in the future, it implies that wake_affine has
> > > +	 * temporarily asked NUMA balancing to backoff from placement.
> > > +	 */
> > > +	if (numa_migrate_retry > p->numa_migrate_retry)
> > > +		return;
> > 
> > The above check looks wrong. This check will most likely to be true,
> > numa_migrate_preferred() itself is called either when jiffies >
> > p->numa_migrate_retry or if the task's numa_preferred_nid has changed.
> > 
> 
> You're right, without affine wakeups with a wakeup-intensive workload
> the path may never be hit and with the current code, it effectively acts
> as a broken throttling mechanism.

I haven't tried on an x86 box, but still trying to get my head around
that check. How does affine wakeups differ for this check.

Lets say p->numa_migrate_retry was set by wake_affine and task has
crossed that temporary period where we dont want the task to undergo
numa balancing.

Now the task is back at numa_migrate_preferred(); p->numa_migrate_retry
is lesser than jiffies (something like "current jiffies - 100"). It
would always return back from that check.

In the other scenario, where wake_affine set p->numa_migrate_preferred to a
bigger value, the task calls numa_migrate_preferred(),
numa_migrate_preferred  could be before p->numa_migrate_preferred. In
such a case, we should have stopped the task from migration.
However we overwrite p->numa_migrate_preferred and do the
task_numa_migrate(). Somehow this doesn't seem to achieve what the
commit intended.

Or did I misunderstand?

--
Thanks and Regards
Srikar