On Thu, Sep 04, 2025 at 01:21:47PM -0700, Kuniyuki Iwashima wrote: > On Wed, Sep 3, 2025 at 11:35 PM Johannes Weiner <hannes@xxxxxxxxxxx> wrote: > > > > On Wed, Sep 03, 2025 at 07:02:03PM +0000, Kuniyuki Iwashima wrote: > > > If all workloads were guaranteed to be controlled under memcg, the issue > > > could be worked around by setting tcp_mem[0~2] to UINT_MAX. > > > > > > In reality, this assumption does not always hold, and processes not > > > controlled by memcg lose the seatbelt and can consume memory up to > > > the global limit, becoming noisy neighbour. > > > > It's been repeatedly pointed out to you that this container > > configuration is not, and cannot be, supported. Processes not > > controlled by memcg have many avenues to become noisy neighbors in a > > multi-tenant system. > > > > So my NAK still applies. Please carry this forward in all future patch > > submissions even if your implementation changes. > > I see. > > I'm waiting for Shakeel's response as he agreed on decoupling > memcg and tcp_mem and suggested the bpf approach. Yes I agreed on decoupling memcg and tcp_mem but not for a weird configuration, so please stop using this motivatioan already. You can motivate the decoupling simply on performance. Why pay the cost of two orthogonal accounting mechanisms concurrently? Also you are not disabling memcg accounting, so we should be good from memcg side. Make this very clear in your commit message. I don't care how you plan to use this feature to enable your weird use-case but make sure this feature is beneficial to general Linux users.
