On Thu, Sep 9, 2021 at 7:37 AM Michal Koutn√Ω [off-list ref] wrote:

On Wed, Sep 08, 2021 at 01:24:36PM +0800, brookxu [off-list ref] wrote:

So there could be two type of events (not referring to the v1-specific
failcnt):
- max - number of times the cgroup's misc.max was hit,
- fail - number of times operation failed (rejected) in the cgroup.

The former would tell you which limit is probably set too low, the
latter would capture which cgroup workload is affected. (The difference
would become apparent with >1 level deep hierarchies.)

We are adding two files in this patch series, misc.events and
misc.events.local. I think "fail" should go in misc.events.local and
its name should be changed to "max". So, both files will have "max"
entry but the one in misc.events will refer to the failures in the
current cgroup and its children, and the one in misc.events.local will
refer to the originating cgroup.

Regards,
Michal

[1] https://lore.kernel.org/lkml/20191202191100.GF16681-LpCCV3molIbIZ9tKgghJQ/d9D2ou9A/h@public.gmane.orgcebook.com/