MIME-Version: 1.0
References: <20220107234138.1765668-1-joshdon@google.com> <Yd189wHB2LJcK1Pv@hirez.programming.kicks-ass.net>
 <CABk29NuGs_9uxgbv678W=BGGinZNiUHO5T57FHGbOG+HP-FT2g@mail.gmail.com> <Yd8429VrPCay9cn4@slm.duckdns.org>
In-Reply-To: <Yd8429VrPCay9cn4@slm.duckdns.org>
From:   Josh Don <joshdon@google.com>
Date:   Wed, 12 Jan 2022 17:55:05 -0800
Message-ID: <CABk29NtNhj2=6gHLSRp6XK4+BVtoSk3QgKdTbf4dKxTRdvgNtQ@mail.gmail.com>
Subject: Re: [PATCH 1/2] cgroup: add cpu.stat_percpu
To:     Tejun Heo <tj@kernel.org>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Zefan Li <lizefan.x@bytedance.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Ingo Molnar <mingo@redhat.com>,
        Juri Lelli <juri.lelli@redhat.com>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        cgroups@vger.kernel.org,
        linux-kernel <linux-kernel@vger.kernel.org>, haoluo@google.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Tejun,

On Wed, Jan 12, 2022 at 12:24 PM Tejun Heo <tj@kernel.org> wrote:
>
> Hello,
>
> On Tue, Jan 11, 2022 at 03:38:20PM -0800, Josh Don wrote:
> > Is the concern there just the extra overhead from making multiple
> > trips into this handler and re-allocating the buffer until it is large
> > enough to take all the output? In that case, we could pre-allocate
> > with a size of the right order of magnitude, similar to /proc/stat.
> >
> > Lack of per-cpu stats is a gap between cgroup v1 and v2, for which v2
> > can easily support this interface given that it already tracks the
> > stats percpu internally. I opted to dump them all in a single file
> > here, to match the consolidation that occurred from cpuacct->cpu.stat.
>
> Yeah, nack on this. That part was dropped intentionally. These text pseudo
> files aren't a great medium for this sort of (potentially large) data dump
> and they scale really badly with new fields which we may want to expose in
> the future. For detailed introspection, a better route would be using bpf
> and if that's inconvenient for some reason trying to make them more
> convenient.

Fair enough. Apart from needing to expose an rstat flush mechanism to
BPF, the main issues that come to mind are conveniently associating
with a particular cgroup, and allowing non-root to collect stats
independently. Hao's patch series
(https://lkml.kernel.org/r/%3C20220112192547.3054575-1-haoluo@google.com%3E)
makes this particularly convenient, but that's a topic for a separate
conversation :)