Re: [PATCH v2 3/4] sched:Fix task_numa_migrate to always update preferred node

[Rik van Riel <riel@redhat.com>]

On 06/16/2015 07:56 AM, Srikar Dronamraju wrote:

> @@ -1519,16 +1519,9 @@ static int task_numa_migrate(struct task_struct *p)
>  	 * and is migrating into one of the workload's active nodes, remember
>  	 * this node as the task's preferred numa node, so the workload can
>  	 * settle down.
> -	 * A task that migrated to a second choice node will be better off
> -	 * trying for a better one later. Do not set the preferred node here.
>  	 */
>  	if (p->numa_group) {
> -		if (env.best_cpu == -1)
> -			nid = env.src_nid;
> -		else
> -			nid = env.dst_nid;
> -
> -		if (node_isset(nid, p->numa_group->active_nodes))
> +		if (env.dst_nid != p->numa_preferred_nid)
>  			sched_setnuma(p, env.dst_nid);
>  	}

Looking at the original code again, it looks like my code
has a potential bug (or at least downside), too.

We set p->numa_group->active_nodes depending on which nodes
a group triggers many NUMA faults happen from (the CPUs the
tasks in the group were running on when they had NUMA faults).

This means if a workload is not yet converged, the active_nodes
mask may be much larger than desired, and we can end up setting
p->numa_preferred_nid to a node that is currently in the active_nodes
mask, but really shouldn't be...

I have no ideas on how to improve that situation, though :)