Open Stack

I started to look at a cinder bug [1] to migrate our cgroup usage from 
v1 to v2 and I'm wondering: is anyone using this feature to throttle 
cinder's local I/O per cinder-volume instance?  Docker since 20.10 and 
kubernetes since version 1.19 support cgroup v2.  Since cinder services 
are deployed within containers, I think it makes sense to consider 
implementing I/O throttling at the container-level as opposed to 
application-specific code within a particular service (cinder in this 
case, but I believe glance also leverages cgroups).  That would mean 
instead of updating our implementation to use the v2 interface, we could 
drop our cgroup code entirely in favor of a more generic solution that 
could be applied to any container that performs local I/O.  Would anyone 
be opposed to this idea?  Thoughts?

1: https://bugs.launchpad.net/cinder/+bug/1942203

-- 
Jon