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, 28 May 2020 22:14:41 +0100
From:   Chris Down <chris@chrisdown.name>
To:     Johannes Weiner <hannes@cmpxchg.org>
Cc:     Shakeel Butt <shakeelb@google.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Tejun Heo <tj@kernel.org>, Michal Hocko <mhocko@kernel.org>,
        Linux MM <linux-mm@kvack.org>,
        Cgroups <cgroups@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Kernel Team <kernel-team@fb.com>
Subject: Re: [PATCH] mm, memcg: reclaim more aggressively before high
 allocator throttling
Message-ID: <20200528211441.GA3410@chrisdown.name>
References: <20200520143712.GA749486@chrisdown.name>
 <CALvZod7rSeAKXKq_V0SggZWn4aL8pYWJiej4NdRd8MmuwUzPEw@mail.gmail.com>
 <20200528194831.GA2017@chrisdown.name>
 <20200528202944.GA76514@cmpxchg.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Disposition: inline
In-Reply-To: <20200528202944.GA76514@cmpxchg.org>
User-Agent: Mutt/1.14.2 (2020-05-25)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Johannes Weiner writes:
>> I don't feel strongly either way, but current->memcg_nr_pages_over_high can
>> be very large for large allocations.
>>
>> That said, maybe we should just reclaim `max(SWAP_CLUSTER_MAX, current -
>> high)` for each loop? I agree that with this design it looks like perhaps we
>> don't need it any more.
>>
>> Johannes, what do you think?
>
>How about this:
>
>Reclaim memcg_nr_pages_over_high in the first iteration, then switch
>to SWAP_CLUSTER_MAX in the retries.
>
>This acknowledges that while the page allocator and memory.max reclaim
>every time an allocation is made, memory.high is currently batched and
>can have larger targets. We want the allocating thread to reclaim at
>least the batch size, but beyond that only what's necessary to prevent
>premature OOM or failing containment.
>
>Add a comment stating as much.
>
>Once we reclaim memory.high synchronously instead of batched, this
>exceptional handling is no longer needed and can be deleted again.
>
>Does that sound reasonable?

That sounds good to me, thanks. I'll change that in v2 soonish and update the 
changelog.