Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 19 Aug 2021 16:38:59 -0400
From:   Johannes Weiner <hannes@cmpxchg.org>
To:     Michal Hocko <mhocko@suse.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        Leon Yang <lnyng@fb.com>, Chris Down <chris@chrisdown.name>,
        Roman Gushchin <guro@fb.com>, linux-mm@kvack.org,
        cgroups@vger.kernel.org, linux-kernel@vger.kernel.org,
        kernel-team@fb.com
Subject: Re: [PATCH] mm: memcontrol: fix occasional OOMs due to proportional
 memory.low reclaim
Message-ID: <YR7BY2Z0cXvW/uTO@cmpxchg.org>
References: <20210817180506.220056-1-hannes@cmpxchg.org>
 <YR5yUolPN+hSsUgJ@dhcp22.suse.cz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YR5yUolPN+hSsUgJ@dhcp22.suse.cz>
Precedence: bulk

On Thu, Aug 19, 2021 at 05:01:38PM +0200, Michal Hocko wrote:
> On Tue 17-08-21 14:05:06, Johannes Weiner wrote:
> > We've noticed occasional OOM killing when memory.low settings are in
> > effect for cgroups. This is unexpected and undesirable as memory.low
> > is supposed to express non-OOMing memory priorities between cgroups.
> > 
> > The reason for this is proportional memory.low reclaim. When cgroups
> > are below their memory.low threshold, reclaim passes them over in the
> > first round, and then retries if it couldn't find pages anywhere else.
> > But when cgroups are slighly above their memory.low setting, page scan
> > force is scaled down and diminished in proportion to the overage, to
> > the point where it can cause reclaim to fail as well - only in that
> > case we currently don't retry, and instead trigger OOM.
> > 
> > To fix this, hook proportional reclaim into the same retry logic we
> > have in place for when cgroups are skipped entirely. This way if
> > reclaim fails and some cgroups were scanned with dimished pressure,
> > we'll try another full-force cycle before giving up and OOMing.
> > 
> > Reported-by: Leon Yang <lnyng@fb.com>
> > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
> 
> Acked-by: Michal Hocko <mhocko@suse.com>

Thanks

> 
> Although I have to say that the code is quite tricky and it deserves
> more comments. See below.
> 
> [...]
> > @@ -2576,6 +2578,15 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
> >  			 * hard protection.
> >  			 */
> >  			unsigned long cgroup_size = mem_cgroup_size(memcg);
> > +			unsigned long protection;
> > +
> > +			/* memory.low scaling, make sure we retry before OOM */
> > +			if (!sc->memcg_low_reclaim && low > min) {
> > +				protection = low;
> > +				sc->memcg_low_skipped = 1;
> > +			} else {
> > +				protection = min;
> > +			}
> 
> Just by looking at this in isolation one could be really curious how
> does this not break the low memory protection altogether.

You're right, it's a bit too terse.

> The logic is spread over 3 different places.
> 
> Would something like the following be more understandable?
> 
> 			/*
> 			 * Low limit protected memcgs are already excluded at
> 			 * a higher level (shrink_node_memcgs) but scaling
> 			 * down the reclaim target can result in hard to
> 			 * reclaim and premature OOM. We do not have a full
> 			 * picture here so we cannot really judge this
> 			 * sutuation here but pro-actively flag this scenario
> 			 * and let do_try_to_free_pages to retry if
> 			 * there is no progress.
> 			 */

I've been drafting around with this, but it seems to say the same
thing as the comment I put into struct scan_control already:

	/*
	 * Cgroup memory below memory.low is protected as long as we
	 * don't threaten to OOM. If any cgroup is reclaimed at
	 * reduced force or passed over entirely due to its memory.low
	 * setting (memcg_low_skipped), and nothing is reclaimed as a
	 * result, then go back back for one more cycle that reclaims
	 * the protected memory (memcg_low_reclaim) to avert OOM.
	 */

How about a brief version of this with a pointer to the original?

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 701106e1829c..c32d686719d5 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -2580,7 +2580,12 @@ static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
 			unsigned long cgroup_size = mem_cgroup_size(memcg);
 			unsigned long protection;
 
-			/* memory.low scaling, make sure we retry before OOM */
+			/*
+			 * Soft protection must not cause reclaim failure. Let
+			 * the upper level know if we skipped pages during the
+			 * first pass, so it can retry if necessary. See the
+			 * struct scan_control definition of those flags.
+			 */
 			if (!sc->memcg_low_reclaim && low > min) {
 				protection = low;
 				sc->memcg_low_skipped = 1;
@@ -2853,16 +2858,16 @@ static void shrink_node_memcgs(pg_data_t *pgdat, struct scan_control *sc)
 
 		if (mem_cgroup_below_min(memcg)) {
 			/*
-			 * Hard protection.
-			 * If there is no reclaimable memory, OOM.
+			 * Hard protection. Always respected. If there is not
+			 * enough reclaimable memory elsewhere, it's an OOM.
 			 */
 			continue;
 		} else if (mem_cgroup_below_low(memcg)) {
 			/*
-			 * Soft protection.
-			 * Respect the protection only as long as
-			 * there is an unprotected supply
-			 * of reclaimable memory from other cgroups.
+			 * Soft protection must not cause reclaim failure. Let
+			 * the upper level know if we skipped pages during the
+			 * first pass, so it can retry if necessary. See the
+			 * struct scan_control definition of those flags.
 			 */
 			if (!sc->memcg_low_reclaim) {
 				sc->memcg_low_skipped = 1;