Re: [PATCH 19/27] mm: Move vmscan writes and file write accounting to the node

[Mel Gorman]

On Thu, Jun 23, 2016 at 04:06:09PM +0200, Vlastimil Babka wrote:
> On 06/23/2016 03:57 PM, Mel Gorman wrote:
> >On Wed, Jun 22, 2016 at 04:40:39PM +0200, Michal Hocko wrote:
> >>On Tue 21-06-16 15:15:58, Mel Gorman wrote:
> >>>As reclaim is now node-based, it follows that page write activity
> >>>due to page reclaim should also be accounted for on the node. For
> >>>consistency, also account page writes and page dirtying on a per-node
> >>>basis.
> >>>
> >>>After this patch, there are a few remaining zone counters that may
> >>>appear strange but are fine. NUMA stats are still per-zone as this is a
> >>>user-space interface that tools consume. NR_MLOCK, NR_SLAB_*, NR_PAGETABLE,
> >>>NR_KERNEL_STACK and NR_BOUNCE are all allocations that potentially pin
> >>>low memory and cannot trivially be reclaimed on demand. This information
> >>>is still useful for debugging a page allocation failure warning.
> >>
> >>As I've said in other patch. I think we will need to provide
> >>/proc/nodeinfo to fill the gap.
> >>
> >
> >I added a patch on top that prints the node stats in zoneinfo but only
> >once for the first populated zone in a node. Doing this or creating a
> >new file are both potentially surprising but extending zoneinfo means
> >there is a greater chance that a user will spot the change.
> 
> BTW, there should already be /sys/devices/system/node/nodeX/vmstat providing
> the per-node stats, right?
> 
> Changing zoneinfo so that some zones have some stats that others don't seems
> to me like it can break some scripts...
> 

I suspect a lot of scripts that read zoneinfo just blindly record it.
Similarly, there is no guarantee that a smart script knows to look in
the per-node vmstat files either. This is a question of "wait see what
breaks".

-- 
Mel Gorman
SUSE Labs