[PATCH] mm: vmscan: consistent update to pgrefill

[PATCH] mm: vmscan: consistent update to pgrefill

[Shakeel Butt]

The vmstat pgrefill is useful together with pgscan and pgsteal stats to
measure the reclaim efficiency. However vmstat's pgrefill is not updated
consistently at system level. It gets updated for both global and memcg
reclaim however pgscan and pgsteal are updated for only global reclaim.
So, update pgrefill only for global reclaim. If someone is interested in
the stats representing both system level as well as memcg level reclaim,
then consult the root memcg's memory.stat instead of /proc/vmstat.

Signed-off-by: Shakeel Butt <shakeelb@google.com>
---
 mm/vmscan.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 5215840ee217..4167b0cc1784 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -2030,7 +2030,8 @@ static void shrink_active_list(unsigned long nr_to_scan,
 
 	__mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, nr_taken);
 
-	__count_vm_events(PGREFILL, nr_scanned);
+	if (!cgroup_reclaim(sc))
+		__count_vm_events(PGREFILL, nr_scanned);
 	__count_memcg_events(lruvec_memcg(lruvec), PGREFILL, nr_scanned);
 
 	spin_unlock_irq(&pgdat->lru_lock);
-- 
2.27.0.383.g050319c2ae-goog

Re: [PATCH] mm: vmscan: consistent update to pgrefill

[Yafang Shao]

On Sat, Jul 11, 2020 at 9:15 AM Shakeel Butt <shakeelb@google.com> wrote:
>
> The vmstat pgrefill is useful together with pgscan and pgsteal stats to
> measure the reclaim efficiency. However vmstat's pgrefill is not updated
> consistently at system level. It gets updated for both global and memcg
> reclaim however pgscan and pgsteal are updated for only global reclaim.
> So, update pgrefill only for global reclaim. If someone is interested in
> the stats representing both system level as well as memcg level reclaim,
> then consult the root memcg's memory.stat instead of /proc/vmstat.
>
> Signed-off-by: Shakeel Butt <shakeelb@google.com>

Acked-by: Yafang Shao <laoar.shao@gmail.com>

> ---
>  mm/vmscan.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 5215840ee217..4167b0cc1784 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -2030,7 +2030,8 @@ static void shrink_active_list(unsigned long nr_to_scan,
>
>         __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, nr_taken);
>
> -       __count_vm_events(PGREFILL, nr_scanned);
> +       if (!cgroup_reclaim(sc))
> +               __count_vm_events(PGREFILL, nr_scanned);
>         __count_memcg_events(lruvec_memcg(lruvec), PGREFILL, nr_scanned);
>
>         spin_unlock_irq(&pgdat->lru_lock);
> --
> 2.27.0.383.g050319c2ae-goog
>


-- 
Thanks
Yafang

Re: [PATCH] mm: vmscan: consistent update to pgrefill

[Roman Gushchin]

On Fri, Jul 10, 2020 at 06:14:59PM -0700, Shakeel Butt wrote:
> The vmstat pgrefill is useful together with pgscan and pgsteal stats to
> measure the reclaim efficiency. However vmstat's pgrefill is not updated
> consistently at system level. It gets updated for both global and memcg
> reclaim however pgscan and pgsteal are updated for only global reclaim.
> So, update pgrefill only for global reclaim. If someone is interested in
> the stats representing both system level as well as memcg level reclaim,
> then consult the root memcg's memory.stat instead of /proc/vmstat.
> 
> Signed-off-by: Shakeel Butt <shakeelb@google.com>

So you went into the opposite direction from the "previous version"
( https://lkml.org/lkml/2020/5/7/1464 ) ?

Acked-by: Roman Gushchin <guro@fb.com>

Thanks!

Re: [PATCH] mm: vmscan: consistent update to pgrefill

[Shakeel Butt]

On Fri, Jul 10, 2020 at 7:32 PM Roman Gushchin <guro@fb.com> wrote:
>
> On Fri, Jul 10, 2020 at 06:14:59PM -0700, Shakeel Butt wrote:
> > The vmstat pgrefill is useful together with pgscan and pgsteal stats to
> > measure the reclaim efficiency. However vmstat's pgrefill is not updated
> > consistently at system level. It gets updated for both global and memcg
> > reclaim however pgscan and pgsteal are updated for only global reclaim.
> > So, update pgrefill only for global reclaim. If someone is interested in
> > the stats representing both system level as well as memcg level reclaim,
> > then consult the root memcg's memory.stat instead of /proc/vmstat.
> >
> > Signed-off-by: Shakeel Butt <shakeelb@google.com>
>
> So you went into the opposite direction from the "previous version"
> ( https://lkml.org/lkml/2020/5/7/1464 ) ?
>

Yes because we already had those stats in the root memcg and exposing
root memcg's memory.stat resolved the issue.

> Acked-by: Roman Gushchin <guro@fb.com>

Thanks a lot.

Re: [PATCH] mm: vmscan: consistent update to pgrefill

[Chris Down]

Shakeel Butt writes:
>The vmstat pgrefill is useful together with pgscan and pgsteal stats to
>measure the reclaim efficiency. However vmstat's pgrefill is not updated
>consistently at system level. It gets updated for both global and memcg
>reclaim however pgscan and pgsteal are updated for only global reclaim.
>So, update pgrefill only for global reclaim. If someone is interested in
>the stats representing both system level as well as memcg level reclaim,
>then consult the root memcg's memory.stat instead of /proc/vmstat.
>
>Signed-off-by: Shakeel Butt <shakeelb@google.com>

Acked-by: Chris Down <chris@chrisdown.name>