On Mon, Jul 7, 2025 at 11:45 PM Daniel Sedlak <daniel.sedlak@xxxxxxxxx> wrote: > > Hi Eric, > Thank you for your feedback. > > On 7/7/25 2:48 PM, Eric Dumazet wrote: > > On Mon, Jul 7, 2025 at 3:55 AM Daniel Sedlak <daniel.sedlak@xxxxxxxxx> wrote: > >> > >> Currently, we have two memory pressure counters for TCP sockets [1], > >> which we manipulate only when the memory pressure is signalled through > >> the proto struct [2]. > >> > >> However, the memory pressure can also be signaled through the cgroup > >> memory subsystem, which we do not reflect in the netstat counters. > >> > >> This patch adds a new counter to account for memory pressure signaled by > >> the memory cgroup. > > > > OK, but please amend the changelog to describe how to look at the > > per-cgroup information. > > Sure, I will explain it more in v2. I was not sure how much of a > "storytelling" is appropriate in the commit message. > > > > I am sure that having some details on how to find the faulty cgroup > > would also help. > > Right now, we have a rather fragile bpftrace script for that, but we > have a WIP patch for memory management, which will expose which cgroup > is having "difficulties", but that is still ongoing work. > > Or do you have any suggestions on how we can incorporate this > information about "this particular cgroup is under pressure" into the > net subsystem? Maybe a log line? Perhaps an additional trace point ? bpftrace would require to ship a separate program... Ideally we could trace the cgroup path, or at least the pid.
