Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230615034830.1361853-1-hezhongkun.hzk@bytedance.com> <CAOUHufZBUEm1P7gm0nFkPSFkNg2fPbs3v3qhL-R9m5yFNoW2YA@mail.gmail.com>
In-Reply-To: <CAOUHufZBUEm1P7gm0nFkPSFkNg2fPbs3v3qhL-R9m5yFNoW2YA@mail.gmail.com>
From:   Fabian Deutsch <fdeutsch@redhat.com>
Date:   Thu, 15 Jun 2023 11:32:50 +0200
Message-ID: <CA+PVUaQ40isbN14=gJTM7Bo_h1cRF=7AQUbE2g+PYNUMwdZLog@mail.gmail.com>
Subject: Re: [RFC PATCH 1/3] zram: charge the compressed RAM to the page's memcgroup
To:     Yu Zhao <yuzhao@google.com>
Cc:     Zhongkun He <hezhongkun.hzk@bytedance.com>, minchan@kernel.org,
        senozhatsky@chromium.org, mhocko@suse.com, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org,
        Andrea Arcangeli <aarcange@redhat.com>,
        David Hildenbrand <david@redhat.com>,
        Yosry Ahmed <yosryahmed@google.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Jun 15, 2023 at 6:59=E2=80=AFAM Yu Zhao <yuzhao@google.com> wrote:
>
> On Wed, Jun 14, 2023 at 9:48=E2=80=AFPM Zhongkun He
> <hezhongkun.hzk@bytedance.com> wrote:
> >
> > The compressed RAM is currently charged to kernel, not to
> > any memory cgroup, which is not satisfy our usage scenario.
> > if the memory of a task is limited by memcgroup, it will
> > swap out the memory to zram swap device when the memory
> > is insufficient. In that case, the memory limit will have
> > no effect.
> >
> > So, it should makes sense to charge the compressed RAM to
> > the page's memory cgroup.

While looking at this in the past weeks, I believe that there are two
distinct problems:
1. Direct zram usage by process within a cg ie. a process writing to a
zram device
2. Indirect zram usage by a process within a cg via swap (described above)

Both of them probably require different solutions.
In order to fix #1, accounting a zram device should be accounted
towards a cgroup. IMHO this is something that should be fixed.

Yu Zhao and Yosry are probably much more familiar with the solution to #2.
WRT per-cgrou-swapfile, to me this is addressing #2, but I agree with
Yu Zhao, that there are probably better solutions to this.

Lastly, this patchset, while it will possibly not address the swap
issue (#2) completely, is it satisfying the needs of #1?

- fabian

> We used to do this a long time ago, but we had per-memcg swapfiles [1[
> to prevent compressed pages from different memcgs from sharing the
> same zspage.
>
> Does this patchset alone suffer from the same problem, i.e., memcgs
> sharing zspages?
>
> [1] https://lwn.net/Articles/592923/
>