Date:   Wed, 24 Nov 2021 21:03:20 +0900
From:   Alexey Avramov <hakavlad@inbox.lv>
To:     Mel Gorman <mgorman@techsingularity.net>
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org, mhocko@suse.com,
        vbabka@suse.cz, neilb@suse.de, akpm@linux-foundation.org,
        corbet@lwn.net, riel@surriel.com, hannes@cmpxchg.org,
        david@fromorbit.com, willy@infradead.org, hdanton@sina.com,
        penguin-kernel@i-love.sakura.ne.jp, oleksandr@natalenko.name,
        kernel@xanmod.org, michael@michaellarabel.com, aros@gmx.com,
        hakavlad@gmail.com
Subject: Re: mm: 5.16 regression: reclaim_throttle leads to stall in
 near-OOM conditions
Message-ID: <20211124210320.591dd883@mail.inbox.lv>
In-Reply-To: <20211124115007.GG3366@techsingularity.net>
References: <20211124011954.7cab9bb4@mail.inbox.lv>
        <20211124103550.GE3366@techsingularity.net>
        <20211124195449.33f31e7f@mail.inbox.lv>
        <20211124115007.GG3366@techsingularity.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

>but can you test this?

I have already tested different parameters, and 
found that even zero timeout is unsatisfactory.

The introduction of reclaim_throttle() itself 
dramatically worsens the stall in near-OOM.

More info: 
https://lore.kernel.org/lkml/20211124011954.7cab9bb4@mail.inbox.lv/

What test else I should perform?

Okay, on one side of the scale is the use of the CPU, on the other 
side is the risk of stall.

My dissatisfaction is caused by the fact that the scale has now 
tipped sharply in favor of stall.
Although even before this change, users complained about the inability 
to wait for OOM:
https://lore.kernel.org/lkml/d9802b6a-949b-b327-c4a6-3dbca485ec20@gmx.com/