Shared cpus are being removed from a container's cgroups, after Kubelet restart. This happens because after Kubelet restart the CPU manager reconciliation loop get triggered. 
It sends a container update request without the shared cpus (because CM manager is not aware of the shared cpus concept). We should add the shared cpus on CRI-O side again.

  Always  

    1. Create pod that request for shared cpus
    2. Check the pod cgroups - the shared cpus are in its cpuset
    3. Restart Kubelet on the node where the pod stays.
    4. Check the pod cgroups - the shared cpus are no longer in its cpuset while they should.

    The shared cpus are no longer in the container's cgroup cpuset after Kubelet restart

    The shared cpus should be under container's cgroup cpuset after Kubelet restart