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

[Mel Gorman]

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.
> 

Sorry for the delay getting back -- viral infections combined with a public
day off is slowing me.

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. However, I've confirmed that "fixing"
it has mixed results with many regressions on x86 for both 2 and 4 socket
boxes. I need time to think about it and see if this can be fixed without
introducing another regression. I'll also check if a plain revert is the
way to go for a short-term fix and then revisit it.

Thanks Srikar.

-- 
Mel Gorman
SUSE Labs