On Mon, Sep 13, 2021 at 7:24 PM brookxu <brookxu.cn@xxxxxxxxx> wrote: > > Thanks for your time. > > Vipin Sharma wrote on 2021/9/14 12:51 上午: > > On Sun, Sep 12, 2021 at 10:01 PM brookxu <brookxu.cn@xxxxxxxxx> wrote: > >> > >> From: Chunguang Xu <brookxu@xxxxxxxxxxx> > >> > >> Introduce misc.events and misc.events.local to make it easier for > > > > I thought Tejun only gave go ahead for misc.events and not for > > misc.events.local. > > > > Maybe I missed something. I think events.local is somewhat useful. For > example, the events of node A is large. If we need to determine whether > it is caused by the max of node A, if there is no events.local, then we > need to traverse the events of the child nodes and compare them with > node A. This is a bit complicated. If there is events.local, we can do > it very easily. Should we keep the events.local interface？ Tejun mentioned in his previous email that he prefers the hierarchical one. https://lore.kernel.org/lkml/YTuX6Cpv1kg+DHmJ@xxxxxxxxxxxxxxx/ I agree with you that it's easier to identify the constraint cgroup with the local file. However, there is one downside also, which is if a cgroup gets deleted then that local information is lost, we will need a hierarchical reporting to observe the resource constraint. I will be fine with both files but if I have to choose one I am now more inclined towards hierarchical (events). Thanks Vipin
