kernel: enable group scheduling in CFQ

[devimc]

With group scheduling we can specify a block IO weight
for all devices or a specific device. This change is to support
docker options --blkio-weight and --blkio-weight-device

fixes #156

Signed-off-by: Julio Montes julio.montes@intel.com

[devimc]

@jcvenegas @jodh-intel @kata-containers/packaging @Weichen81 @Pennyzct please take a look

[jcvenegas]

looks good , could you bump the file kernel/kata_config_version

[devimc]

@jcvenegas done, thanks

[mcastelino]

@devimc this confuses me a little bit. So we plan to enforce inside the VM vs setting this up on the host?

Enforcing inside the VM may help, but you really want to constrain it on the host side

[mcastelino]

@devimc expanding this a little more. Per spec

By default, all containers get the same proportion of block IO bandwidth (blkio). This proportion is 500. To modify this proportion, change the container’s blkio weight relative to the weighting of all other running containers using the --blkio-weight flag.

So this is proportional to the other containers running on the host.

Am I reading the spec correctly here.

[amshinde]

@mcastelino Yes, I think the weighs in the docker case applies to all containers running on the host. So for docker, it would make more sense to apply the constraints on the host side.
@devimc Have you looked at the equivalent option for kubernetes and how the blkio-weights are applied there?

[devimc]

@mcastelino from a compatible (runc) perspective I think the constraint should be applied in the VM, in that way the constraint is honoured and visible for the container (/sys/fs/cgroup/...). From a performance perspective I don't have idea, @grahamwhaley any thoughts ?

@amshinde afaik for kubernetes there are only two kind of resources: cpu and memory https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/#resource-types

[amshinde]

@devimc In case of docker, there is just one container running inside the VM. Its not quite useful to have this constraint applied to a single container.

[devimc]

@amshinde yes, I think the same
closing PR

[jcvenegas]

@mcastelino @amshinde I agree with @devimc we should apply inside of the VM when possible the main reason that we are different with the spec is that the VM is a pod for us.

[jcvenegas]

commented after refresh, that is a good point @amshinde.

[grahamwhaley]

Now we've discussed and agreed - can we check and add this info to the limitations doc if necessary please.