Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200109225646.22983-1-xiyou.wangcong@gmail.com>
 <20200110073822.GC29802@dhcp22.suse.cz> <CAM_iQpVN4MNhcK0TXvhmxsCdkVOqQ4gZBzkDHykLocPC6Va7LQ@mail.gmail.com>
 <20200121090048.GG29276@dhcp22.suse.cz> <CAM_iQpU0p7JLyQ4mQ==Kd7+0ugmricsEAp1ST2ShAZar2BLAWg@mail.gmail.com>
 <20200127144931.GM1183@dhcp22.suse.cz>
In-Reply-To: <20200127144931.GM1183@dhcp22.suse.cz>
From:   Cong Wang <xiyou.wangcong@gmail.com>
Date:   Mon, 27 Jan 2020 16:46:16 -0800
Message-ID: <CAM_iQpXGqGse1pybMW+dCi_vp05W9GTRWtS-3O2hdsY-hR=Yng@mail.gmail.com>
Subject: Re: [PATCH] mm: avoid blocking lock_page() in kcompactd
To:     Michal Hocko <mhocko@kernel.org>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-mm <linux-mm@kvack.org>, Mel Gorman <mgorman@suse.de>,
        Vlastimil Babka <vbabka@suse.cz>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Jan 27, 2020 at 6:49 AM Michal Hocko <mhocko@kernel.org> wrote:
>
> On Sun 26-01-20 11:53:55, Cong Wang wrote:
> > On Tue, Jan 21, 2020 at 1:00 AM Michal Hocko <mhocko@kernel.org> wrote:
> > >
> > > On Mon 20-01-20 14:48:05, Cong Wang wrote:
> > > > It got stuck somewhere along the call path of mem_cgroup_try_charge(),
> > > > and the trace events of mm_vmscan_lru_shrink_inactive() indicates this
> > > > too:
> > >
> > > So it seems that you are condending on the page lock. It is really
> > > unexpected that the reclaim would take that long though. Please try to
> > > enable more vmscan tracepoints to see where the time is spent.
> >
> > Sorry for the delay. I have been trying to collect more data in one shot.
> >
> > This is a a typical round of the loop after enabling all vmscan tracepoints:
> >
> >            <...>-455450 [007] .... 4048595.842992:
> > mm_vmscan_memcg_reclaim_begin: order=0 may_writepage=1
> > gfp_flags=GFP_NOFS|__GFP_HIGHMEM|__GFP_HARDWALL|__GFP_MOVABLE
> > classzone_idx=4
> >            <...>-455450 [007] .... 4048595.843012:
> > mm_vmscan_memcg_reclaim_end: nr_reclaimed=0
>
> This doesn't tell us much though. This reclaim round has taken close to
> no time. See timestamps.
>
> > The whole trace output is huge (33M), I can provide it on request.
>
> Focus on reclaim rounds that take a long time and see where it gets you.

I reviewed the tracing output with my eyes, they all took little time.
But of course I can't review all of them given the size is huge.

For me, it seems that the loop happens in its caller, something
like:

retry:
  mm_vmscan_memcg_reclaim_begin();
  ...
  mm_vmscan_memcg_reclaim_end();
  goto retry;

So, I think we should focus on try_charge()?

More interestingly, the margin of that memcg is 0:

$ sudo cat /sys/fs/cgroup/memory/system.slice/osqueryd.service/memory.usage_in_bytes
262144000
$ sudo cat /sys/fs/cgroup/memory/system.slice/osqueryd.service/memory.limit_in_bytes
262144000

Thanks!