Vm restart was simply calling systemctl directly and that was wiping out the cpu burst limit from a cgroup settings. Those settings are not persisted anywhere and have to be re-set every time the service restarts. This was handled for the host restart, but not for the VM restart.
To fix, I added a new method in the vm_setup and call it from setup-vm master script. The nexus will call that script.
82ce73988c