Re: [RFC PATCH] memcg: use root_mem_cgroup when css is inherited

[Zhaoyang Huang <huangzhaoyang@gmail.com>]

On Thu, Aug 25, 2022 at 4:50 PM Michal Hocko <mhocko@suse.com> wrote:
>
> On Thu 25-08-22 16:34:04, Zhaoyang Huang wrote:
> > On Thu, Aug 25, 2022 at 2:40 PM Michal Hocko <mhocko@suse.com> wrote:
> > >
> > > On Thu 25-08-22 08:43:52, Zhaoyang Huang wrote:
> > > > On Wed, Aug 24, 2022 at 6:27 PM Michal Hocko <mhocko@suse.com> wrote:
> > > > >
> > > > > On Wed 24-08-22 17:34:42, Zhaoyang Huang wrote:
> > > [...]
> > > > > > IMHO, charging the pages which out of explicitly memory
> > > > > > enabled group to root could solve all of the above constraints with no
> > > > > > harm.
> > > > >
> > > > > This would break the hierarchical property of the controller. So a
> > > > > strong no no. Consider the following example
> > > > >
> > > > >        root
> > > > >         |
> > > > >         A
> > > > > controllers="memory"
> > > > > memory.max = 1G
> > > > > subtree_control=""
> > > > > |      |      |
> > > > > A1     A2     A3
> > > > >
> > > > > althought A1,2,3 do not have their memory controller enabled explicitly
> > > > > they are still constrained by the A memcg limit. If you just charge to
> > > > > the root because it doesn't have memory controller enabled explicitly
> > > > > then you just evade that constrain. I hope you understand why that is a
> > > > > problem.
> > > > IMO, A1-A3 should be explicitly enabled via echo "+memory" >
> > > > A/subtree_control since memory.max has been set.
> > >
> > > You seem to be missing the point I've triedy to make here. It is not
> > > about how the respective subtree should or shouldn't be configured. It
> > > is about the hierarchical behavior. Configuration at a higher level should be
> > > enforced under subtree no matter how that subtree decides to
> > > enabled/disable controllers. Such subtree might have beeb delegated
> > > and configured differently yet the constrain should be still applied.
> > > See the point?
> > >
> > > What you seem to be proposing is similar to cgroup v1 use_hierarchy
> > > configuration. It has been decided that this is undesirable very early
> > > in the cgroup v2 development because it make delegation impossible
> > > (among other reasons).
> > Ok, I would like to know how AA3 achieve the goal of competing with A1
> > and A2 for cpu but keep memory out of control under current policy?
> >         root
> >          |
> >          A
> >  controllers="memory,cpu"
> >  memory.max = 1G
> >  subtree_control="memory,cpu"
> >  |      |            |
> >  A1     A2     A3 subtree_control="cpu"
> >                       |      |
> >                     AA3   AA4 controllers="cpu"
>
> I cannot really give you configuration you want without understanding
> what you are trying to achieve and why do you need it that way. Really,
> you can construct arbitrary hierarchies and only a very small subset of
> them actually makes sense. So far you have been very terse at your goals
> and intentions but rather demanding on the underlying mechanisms. This
> doesn't really makes the discussion productive.
>
> I hope you have at least understood that hierarchical property of the
> cgroup v2 is a must and it won't change. If you need a help to construct
> hierarchy for your specific workload I would recommend to clearly state
> your final goal and reasoning behind. Maybe you will get a more specific
> help that way. Good luck!
Sorry for any misunderstanding among the discussion. My purpose is
real and simple as I have stated from the very beginning that I would
like to have per-app cgroup hierarchy to charge memory to root if it
is not enabled explicitly for memory. The reason has also been stated
like reclaim and workingset regression in suren's report. I don't
think my proposal will do any harm to current v2's mechanism besides
asking for the admin echo "+memory" to their desire group.
> --
> Michal Hocko
> SUSE Labs