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: <d9802b6a-949b-b327-c4a6-3dbca485ec20@gmx.com> <20190820064620.5119-1-drake@endlessm.com>
 <4d998874-d02b-395f-1b81-7034db1a8fcd@redhazel.co.uk>
In-Reply-To: <4d998874-d02b-395f-1b81-7034db1a8fcd@redhazel.co.uk>
From:   Daniel Drake <drake@endlessm.com>
Date:   Fri, 23 Aug 2019 10:14:52 +0800
Message-ID: <CAD8Lp46jYbdtFW2i_HnR8f3GY6Ombne6ouYm2UAnmF9BmeVAFw@mail.gmail.com>
Subject: Re: Let's talk about the elephant in the room - the Linux kernel's
 inability to gracefully handle low memory pressure
To:     ndrw <ndrw.xf@redhazel.co.uk>
Cc:     aros@gmx.com, Linux Kernel <linux-kernel@vger.kernel.org>,
        Linux Upstreaming Team <linux@endlessm.com>,
        Bastien Nocera <hadess@hadess.net>,
        Johannes Weiner <hannes@cmpxchg.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Aug 23, 2019 at 9:54 AM ndrw <ndrw.xf@redhazel.co.uk> wrote:
> That's obviously a lot better than hard freezes but I wouldn't call such
> system lock-ups an excellent result. PSI-triggered OOM killer would have
> indeed been very useful as an emergency brake, and IMHO such mechanism
> should be built in the kernel and enabled by default. But in my
> experience it does a very poor job at detecting imminent freezes on
> systems without swap or with very fast swap (zram).

Perhaps you could share your precise test environment and the PSI
condition you are expecting to hit (that is not being hit). Except for
the single failure report mentioned, it's been working fine here in
all setups, including with zram which is shipped out of the box.

The nice thing about psi is that it's based on how much real-world
time the kernel is spending doing memory management. So it's very well
poised to handle differences in swap speed etc. You effectively just
set the threshold for how much time you view as excessive for the
kernel to be busy doing MM, and psi tells you when that's hit.

> > There's just one issue we've seen so far: a single report of psi reporting
> > memory pressure on a desktop system with 4GB RAM which is only running
> > the normal desktop components plus a single gmail tab in the web browser.
> > psi occasionally reports high memory pressure, so then psi-monitor steps in and
> > kills the browser tab, which seems erroneous.
>
> Is it Chrome/Chromium? If so, that's a known bug
> (https://bugs.chromium.org/p/chromium/issues/detail?id=333617)

The issue does not concern which process is being killed. The issue is
that in the single report we have of this, psi is apparently reporting
high memory pressure even though the system has plenty of free memory.

Daniel