[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <85bca797-6ddd-4541-ade5-a75864870472@bytedance.com>
Date: Tue, 30 Jul 2024 14:06:07 +0800
From: Chuyi Zhou <zhouchuyi@...edance.com>
To: Michal Koutný <mkoutny@...e.com>
Cc: tj@...nel.org, hannes@...xchg.org, surenb@...gle.com,
peterz@...radead.org, chengming.zhou@...ux.dev, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] psi: Inherit parent cgroup psi enable state
Hello Michal,
在 2024/7/29 19:37, Michal Koutný 写道:
> Hello Chuyi.
>
> On Mon, Jul 29, 2024 at 11:41:00AM GMT, Chuyi Zhou <zhouchuyi@...edance.com> wrote:
>> This patch tries to solve this issue. When a child cgroup is created, it
>> would inherit the psi enabled state of the parent in group_init().
>> Once the enable state is found to be false in the css_populate_dir(), the
>> {cpu, io, memory}.pressure files will be hidden using cgroup_file_show().
>
> I'd consider such an inheritance cgroupv1-ism. Notably,
> mmemory.swappiness also had inheritance and it caused some headaches
> because of its susceptibility to races (cgroup creation vs parent
> configuration).
Agree.
>
> What about adding an option (mount option?) to determine the default of
> per-cgroup pressure accounting?
> (Based on reading about your use case in other mail.) With that you
> could only enable it for the "online" groups of interest.
>
Yes. We can add a another mount option to disable the default PSI
accounting and enable the pod we interest.
Thanks.
Powered by blists - more mailing lists