On Tue, Dec 15, 2020 at 6:47 AM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> On Tue, Dec 15, 2020 at 01:22:33PM +1100, Dave Chinner wrote:
> > On Mon, Dec 14, 2020 at 02:37:18PM -0800, Yang Shi wrote:
> > > Currently the number of deferred objects are per shrinker, but some slabs, for example,
> > > vfs inode/dentry cache are per memcg, this would result in poor isolation among memcgs.
> > >
> > > The deferred objects typically are generated by __GFP_NOFS allocations, one memcg with
> > > excessive __GFP_NOFS allocations may blow up deferred objects, then other innocent memcgs
> > > may suffer from over shrink, excessive reclaim latency, etc.
> > >
> > > For example, two workloads run in memcgA and memcgB respectively, workload in B is vfs
> > > heavy workload. Workload in A generates excessive deferred objects, then B's vfs cache
> > > might be hit heavily (drop half of caches) by B's limit reclaim or global reclaim.
> > >
> > > We observed this hit in our production environment which was running vfs heavy workload
> > > shown as the below tracing log:
> > >
> > > <...>-409454 [016] .... 28286961.747146: mm_shrink_slab_start: super_cache_scan+0x0/0x1a0 ffff9a83046f3458:
> > > nid: 1 objects to shrink 3641681686040 gfp_flags GFP_HIGHUSER_MOVABLE|__GFP_ZERO pgs_scanned 1 lru_pgs 15721
> > > cache items 246404277 delta 31345 total_scan 123202138
> > > <...>-409454 [022] .... 28287105.928018: mm_shrink_slab_end: super_cache_scan+0x0/0x1a0 ffff9a83046f3458:
> > > nid: 1 unused scan count 3641681686040 new scan count 3641798379189 total_scan 602
> > > last shrinker return val 123186855
> > >
> > > The vfs cache and page cache ration was 10:1 on this machine, and half of caches were dropped.
> > > This also resulted in significant amount of page caches were dropped due to inodes eviction.
> > >
> > > Make nr_deferred per memcg for memcg aware shrinkers would solve the unfairness and bring
> > > better isolation.
> > >
> > > When memcg is not enabled (!CONFIG_MEMCG or memcg disabled), the shrinker's nr_deferred
> > > would be used. And non memcg aware shrinkers use shrinker's nr_deferred all the time.
> > >
> > > Signed-off-by: Yang Shi <shy828301@xxxxxxxxx>
> > > ---
> > > include/linux/memcontrol.h | 9 +++
> > > mm/memcontrol.c | 110 ++++++++++++++++++++++++++++++++++++-
> > > mm/vmscan.c | 4 ++
> > > 3 files changed, 120 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > > index 922a7f600465..1b343b268359 100644
> > > --- a/include/linux/memcontrol.h
> > > +++ b/include/linux/memcontrol.h
> > > @@ -92,6 +92,13 @@ struct lruvec_stat {
> > > long count[NR_VM_NODE_STAT_ITEMS];
> > > };
> > >
> > > +
> > > +/* Shrinker::id indexed nr_deferred of memcg-aware shrinkers. */
> > > +struct memcg_shrinker_deferred {
> > > + struct rcu_head rcu;
> > > + atomic_long_t nr_deferred[];
> > > +};
> >
> > So you're effectively copy and pasting the memcg_shrinker_map
> > infrastructure and doubling the number of allocations/frees required
> > to set up/tear down a memcg? Why not add it to the struct
> > memcg_shrinker_map like this:
> >
> > struct memcg_shrinker_map {
> > struct rcu_head rcu;
> > unsigned long *map;
> > atomic_long_t *nr_deferred;
> > };
> >
> > And when you dynamically allocate the structure, set the map and
> > nr_deferred pointers to the correct offset in the allocated range.
> >
> > Then this patch is really only changes to the size of the chunk
> > being allocated, setting up the pointers and copying the relevant
> > data from the old to new.
>
> Fully agreed.
Thanks folks. Such idea has been discussed with Roman in the earlier
emails. I agree this would make the code neater. Will do it in v3.
>
> In the longer-term, it may be nice to further expand this and make
> this the generalized intersection between cgroup, node and shrinkers.
>
> There is large overlap with list_lru e.g. - with data of identical
> scope and lifetime, but duplicative callbacks and management. If we
> folded list_lru_memcg into the above data structure, we could also
> generalize and reuse the existing callbacks.
Yes, agree we should look further to combine and deduplicate all the
pieces for the long run.
