On Tue, Jul 24, 2018 at 11:14:29AM -0700, Tejun Heo wrote: > Hello, Vivek. > > On Tue, Jul 24, 2018 at 02:02:04PM -0400, Vivek Goyal wrote: > ... > > > Now sane_behavior flag is gone and V2 does not allow configure limits > > > for root cgroup. So looks like we lost that capability now. Am I missing > > > something. Is there a way to still enable it. > ... > > So we effectively dropped that third mode where sane_behavior could be > > used on legacy hierarchy and get hierarchical throttling(Including > > throttling root cgroup). > > > > Is there any chance that throttling whole device could be supported. Or this > > is a use case we don't want to support going forward. > > So, it's just super weird to implement system-wide control in the root > cgroup cuz system-wide features don't have any reason to be in cgroup > at all. Unless it's sth specific to cgroups, the interface has to be > outside of cgroup so that it can be used whether cgroups is configured > or mounted (depending on dependency, I guess, the former might not be > possible). Hi Tejun, Hmm..., so basically create a separate per disk interface to configure io.max limit for the disk and use cgroup/blk-throttle infrastrcutre to do the actual throttling. So that defintely will require compling in cgroup. (Until and unless code can be refactor in such a way that we can get rid of cgroup dependency completely and still be able to just do IO throttling). > > For io.max, it's a bit weird cuz it is a cgroup only feature but I > think it'd probably be better to introduce a separate system level > interface for consistency's sake. Not sure where would be the best. > Somewhere in sys or proc, I guess, or even dynamic module param? > Would sth like that be too ugly? Something like the knobs available in /sys/block/<disk>/ or in /sys/block/<disk>/queue/. Atleast for io.max, conceptually it makes sense to put system wide per disk limit outside cgroup. I will try to spend some time on this and see if I can make something work. Thanks Vivek -- To unsubscribe from this list: send the line "unsubscribe cgroups" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
