Merge pull request #92608 from lubronzhan/remove_node_when_instance_not_found_vsphere

Remove node when the vm is deleted in vsphere

staging/src/k8s.io/legacy-cloud-providers/vsphere/vclib/virtualmachine.go

@@ -213,6 +213,29 @@ func (vm *VirtualMachine) IsActive(ctx context.Context) (bool, error) {
 	return false, nil
 }
 
+// Exists checks if VM exists and is not terminated
+func (vm *VirtualMachine) Exists(ctx context.Context) (bool, error) {
+	vmMoList, err := vm.Datacenter.GetVMMoList(ctx, []*VirtualMachine{vm}, []string{"summary.runtime.powerState"})
+	if err != nil {
+		klog.Errorf("Failed to get VM Managed object with property summary. err: +%v", err)
+		return false, err
+	}
+	// We check for VMs which are still available in vcenter and has not been terminated/removed from
+	// disk and hence we consider PoweredOn,PoweredOff and Suspended as alive states.
+	aliveStates := []types.VirtualMachinePowerState{
+		types.VirtualMachinePowerStatePoweredOff,
+		types.VirtualMachinePowerStatePoweredOn,
+		types.VirtualMachinePowerStateSuspended,
+	}
+	currentState := vmMoList[0].Summary.Runtime.PowerState
+	for _, state := range aliveStates {
+		if state == currentState {
+			return true, nil
+		}
+	}
+	return false, nil
+}
+
 // GetAllAccessibleDatastores gets the list of accessible Datastores for the given Virtual Machine
 func (vm *VirtualMachine) GetAllAccessibleDatastores(ctx context.Context) ([]*DatastoreInfo, error) {
 	host, err := vm.HostSystem(ctx)

staging/src/k8s.io/legacy-cloud-providers/vsphere/vclib/virtualmachine_test.go

@@ -114,6 +114,26 @@ func TestVirtualMachine(t *testing.T) {
 			}
 		}
 
+		for _, turnOff := range []bool{true, false} {
+			// Turn off for checking if exist return true
+			if turnOff {
+				_, _ = vm.PowerOff(ctx)
+			}
+
+			exist, err := vm.Exists(ctx)
+			if err != nil {
+				t.Error(err)
+			}
+			if !exist {
+				t.Errorf("exist=%t, expected=%t", exist, true)
+			}
+
+			// Turn back on
+			if turnOff {
+				_, _ = vm.PowerOn(ctx)
+			}
+		}
+
 		for _, expect := range []bool{true, false} {
 			active, err := vm.IsActive(ctx)
 			if err != nil {
@@ -140,5 +160,11 @@ func TestVirtualMachine(t *testing.T) {
 				t.Error(err)
 			}
 		}
+
+		// Expecting Exists func to throw error if VM deleted
+		_, err = vm.Exists(ctx)
+		if err == nil {
+			t.Error("expected error")
+		}
 	}
 }

staging/src/k8s.io/legacy-cloud-providers/vsphere/vsphere.go

@@ -766,11 +766,14 @@ func (vs *VSphere) InstanceExistsByProviderID(ctx context.Context, providerID st
 		return false, err
 	}
 	_, err = vs.InstanceID(ctx, convertToK8sType(nodeName))
-	if err == nil {
-		return true, nil
+	if err != nil {
+		if err == cloudprovider.InstanceNotFound {
+			return false, nil
+		}
+		return false, err
 	}
 
-	return false, err
+	return true, nil
 }
 
 // InstanceShutdownByProviderID returns true if the instance is in safe state to detach volumes
@@ -832,15 +835,17 @@ func (vs *VSphere) InstanceID(ctx context.Context, nodeName k8stypes.NodeName) (
 			klog.Errorf("Failed to get VM object for node: %q. err: +%v", convertToString(nodeName), err)
 			return "", err
 		}
-		isActive, err := vm.IsActive(ctx)
+
+		exists, err := vm.Exists(ctx)
 		if err != nil {
-			klog.Errorf("Failed to check whether node %q is active. err: %+v.", convertToString(nodeName), err)
+			klog.Errorf("Failed to check whether node %q still exists. err: %+v.", convertToString(nodeName), err)
 			return "", err
 		}
-		if isActive {
+		if exists {
 			return vs.vmUUID, nil
 		}
-		klog.Warningf("The VM: %s is not in %s state", convertToString(nodeName), vclib.ActivePowerState)
+
+		klog.Warningf("The VM: %s doesn't exist", convertToString(nodeName))
 		return "", cloudprovider.InstanceNotFound
 	}