hardware: Don't consider overhead CPUs for unpinned instances

The 'cpuset_reserved' argument for the '_numa_fit_instance_cell'
function is only ever non-zero if the 'isolate' CPU thread policy is in
effect, and this thread policy can only be used if using the 'dedicated'
CPU policy is used. As such, we don't need to worry about overhead if
either the 'share' CPU thread policy or the 'shared' CPU policy is in
effect.

Part of blueprint use-pcpu-and-vcpu-in-one-instance

Change-Id: Ib863b6a8d3bd3d2848fabd777fc1ac3387a7dd42
Signed-off-by: Stephen Finucane <[email protected]>

Author: stephenfin

nova/virt/hardware.py

@@ -1127,8 +1127,6 @@ def _numa_fit_instance_cell(host_cell, instance_cell, limit_cell=None,
     # NOTE(stephenfin): As with memory, do not allow an instance to overcommit
     # against itself on any NUMA cell
     if instance_cell.cpu_policy == fields.CPUAllocationPolicy.DEDICATED:
-        # TODO(stephenfin): Is 'cpuset_reserved' present if consuming emulator
-        # threads from shared CPU pools? If so, we don't want to add this here
         required_cpus = len(instance_cell.cpuset) + cpuset_reserved
         if required_cpus > len(host_cell.pcpuset):
             LOG.debug('Not enough host cell CPUs to fit instance cell; '
@@ -1140,14 +1138,12 @@ def _numa_fit_instance_cell(host_cell, instance_cell, limit_cell=None,
                       })
             return
     else:
-        required_cpus = len(instance_cell.cpuset) + cpuset_reserved
+        required_cpus = len(instance_cell.cpuset)
         if required_cpus > len(host_cell.cpuset):
             LOG.debug('Not enough host cell CPUs to fit instance cell; '
-                      'required: %(required)d + %(cpuset_reserved)d as '
-                      'overhead, actual: %(actual)d', {
+                      'required: %(required)d, actual: %(actual)d', {
                           'required': len(instance_cell.cpuset),
                           'actual': len(host_cell.cpuset),
-                          'cpuset_reserved': cpuset_reserved
                       })
             return