vsphere: virt-v2v transfers disks sequentially

when scheduling migrations to accommodate the controller_max_vm_inflight
setting for vSphere, we didn't take into account that disks are
transferred sequentially by virt-v2v (as opposed to CDI that transfers
them in parallel by different pods). This lead to performance
degregeration since could have triggered more migrations, in case of
migrations of multi-disk VMs, simultaneously without exceeding the value
of controller_max_vm_inflight. This is fixed by setting the cost of each
VM for which the disks are transferred by virt-v2v to 1.

Signed-off-by: Arik Hadas <[email protected]>

pkg/controller/plan/scheduler/vsphere/scheduler.go

@@ -167,9 +167,18 @@ func (r *Scheduler) buildPending() (err error) {
 		}
 
 		if !vmStatus.MarkedStarted() && !vmStatus.MarkedCompleted() {
+			var cost int
+			if el9, _ := r.Plan.VSphereUsesEl9VirtV2v(); el9 {
+				/// virt-v2v transfers one disk at a time
+				cost = 1
+			} else {
+				// CDI transfers the disks in parallel by different pods
+				cost = len(vm.Disks)
+			}
+
 			pending := &pendingVM{
 				status: vmStatus,
-				cost:   len(vm.Disks),
+				cost:   cost,
 			}
 			r.pending[vm.Host] = append(r.pending[vm.Host], pending)
 		}