OCPBUGS-27390: IPI install fails on Nutanix with DHCP network

go.mod

@@ -127,5 +127,3 @@ require (
 	sigs.k8s.io/structured-merge-diff/v4 v4.4.1 // indirect
 	sigs.k8s.io/yaml v1.4.0 // indirect
 )
-
-replace k8s.io/kubectl => k8s.io/kubectl v0.28.2

go.sum

@@ -869,8 +869,8 @@ k8s.io/klog/v2 v2.110.1 h1:U/Af64HJf7FcwMcXyKm2RPM22WZzyR7OSpYj5tg3cL0=
 k8s.io/klog/v2 v2.110.1/go.mod h1:YGtd1984u+GgbuZ7e08/yBuAfKLSO0+uR1Fhi6ExXjo=
 k8s.io/kube-openapi v0.0.0-20231010175941-2dd684a91f00 h1:aVUu9fTY98ivBPKR9Y5w/AuzbMm96cd3YHRTU83I780=
 k8s.io/kube-openapi v0.0.0-20231010175941-2dd684a91f00/go.mod h1:AsvuZPBlUDVuCdzJ87iajxtXuR9oktsTctW/R9wwouA=
-k8s.io/kubectl v0.28.2 h1:fOWOtU6S0smdNjG1PB9WFbqEIMlkzU5ahyHkc7ESHgM=
-k8s.io/kubectl v0.28.2/go.mod h1:6EQWTPySF1fn7yKoQZHYf9TPwIl2AygHEcJoxFekr64=
+k8s.io/kubectl v0.29.0 h1:Oqi48gXjikDhrBF67AYuZRTcJV4lg2l42GmvsP7FmYI=
+k8s.io/kubectl v0.29.0/go.mod h1:0jMjGWIcMIQzmUaMgAzhSELv5WtHo2a8pq67DtviAJs=
 k8s.io/utils v0.0.0-20240102154912-e7106e64919e h1:eQ/4ljkx21sObifjzXwlPKpdGLrCfRziVtos3ofG/sQ=
 k8s.io/utils v0.0.0-20240102154912-e7106e64919e/go.mod h1:OLgZIPagt7ERELqWJFomSt595RzquPNLL48iOWgYOg0=
 rsc.io/binaryregexp v0.2.0/go.mod h1:qTv7/COck+e2FymRvadv62gMdZztPaShugOCi3I+8D8=

pkg/actuators/machine/machine_scope.go

@@ -60,6 +60,9 @@ type machineScope struct {
 	// For Machine vm create/update use, take a copy of the providerSpec that we can mutate.
 	// This must never be written back to the Machine itself.
 	providerSpecValidated *machinev1.NutanixMachineProviderConfig
+
+	// The machine referenced Node
+	node *corev1.Node
 }
 
 func newMachineScope(params machineScopeParams) (*machineScope, error) {
@@ -288,26 +291,65 @@ func (s *machineScope) setProviderStatus(vm *nutanixClientV3.VMIntentResponse, c
 	s.providerStatus.VmUUID = vm.Metadata.UUID
 
 	// update machine.status.addresses
-	addresses := s.machine.Status.Addresses
-	addr := getExistingAddress(addresses, corev1.NodeInternalIP)
-	if addr != nil {
-		addr.Address = *vm.Status.Resources.NicList[0].IPEndpointList[0].IP
-	} else {
-		addresses = append(addresses, corev1.NodeAddress{
-			Type:    corev1.NodeInternalIP,
-			Address: *vm.Status.Resources.NicList[0].IPEndpointList[0].IP,
-		})
+	machineAddresses := []corev1.NodeAddress{}
+	vmIPEndpoints := make(map[string]bool)
+
+	for _, nic := range vm.Status.Resources.NicList {
+		for _, ipEndpoint := range nic.IPEndpointList {
+			if ipEndpoint.IP != nil && *ipEndpoint.IP != "" {
+				vmIPEndpoints[*ipEndpoint.IP] = true
+			}
+		}
 	}
-	addr = getExistingAddress(addresses, corev1.NodeInternalDNS)
-	if addr != nil {
-		addr.Address = *vm.Spec.Name
+
+	// If the node is linked to the machine, we use the node object's status.addresses to populate that of the machine object.
+	// Also validate the NodeInternalIP addresses are included in the VM's IPEndpointList
+	if s.isNodeLinked() {
+		node, err := s.getNode()
+		if err != nil {
+			return fmt.Errorf("%s: failed to get node: %w", s.machine.Name, err)
+		}
+
+		for _, naddr := range node.Status.Addresses {
+			// verify that the node's InternalIP address is included in the vm's IPEndpointList
+			if naddr.Type == corev1.NodeInternalIP {
+				if _, ok := vmIPEndpoints[naddr.Address]; !ok {
+					return fmt.Errorf("%s: the node's InternalIP address %s is not found in the Machine VM's ip_endpoint_list.", s.machine.Name, naddr.Address)
+				}
+			}
+
+			// add the node's address to the Machine object.
+			machineAddresses = append(machineAddresses, corev1.NodeAddress{
+				Type:    naddr.Type,
+				Address: naddr.Address,
+			})
+
+		}
+		klog.V(3).Infof("%s: use the node object's addresses to populate that of the machine object.", s.machine.Name)
+
 	} else {
-		addresses = append(addresses, corev1.NodeAddress{
-			Type:    corev1.NodeInternalDNS,
+		// Node is not ready
+		klog.V(3).Infof("%s: the node is not ready yet.", s.machine.Name)
+	}
+
+	// if the node is not ready or not having addresses data yet,
+	// use the VM's data to popute the machine addresses
+	if len(machineAddresses) == 0 {
+		for ipAddr, _ := range vmIPEndpoints {
+			machineAddresses = append(machineAddresses, corev1.NodeAddress{
+				Type:    corev1.NodeInternalIP,
+				Address: ipAddr,
+			})
+		}
+
+		machineAddresses = append(machineAddresses, corev1.NodeAddress{
+			Type:    corev1.NodeHostName,
 			Address: *vm.Spec.Name,
 		})
 	}
-	s.machine.Status.Addresses = addresses
+
+	s.machine.Status.Addresses = machineAddresses
+	klog.V(3).Infof("%s: the machine status.addresses=%+v.", s.machine.Name, s.machine.Status.Addresses)
 
 	s.providerStatus.Conditions = setNutanixProviderConditions([]metav1.Condition{condition}, s.providerStatus.Conditions)
 
@@ -319,6 +361,10 @@ func (s *machineScope) isNodeLinked() bool {
 }
 
 func (s *machineScope) getNode() (*corev1.Node, error) {
+	if s.node != nil {
+		return s.node, nil
+	}
+
 	var node corev1.Node
 	if !s.isNodeLinked() {
 		return nil, fmt.Errorf("[Machine: %s] NodeRef empty, unable to get related Node", s.machine.Name)
@@ -339,7 +385,8 @@ func (s *machineScope) getNode() (*corev1.Node, error) {
 		return nil, err
 	}
 
-	return &node, nil
+	s.node = &node
+	return s.node, nil
 }
 
 // nodeHasVolumesAttached returns true if node status still have volumes attached

pkg/actuators/machine/reconciler.go

@@ -6,8 +6,6 @@ import (
 	"time"
 
 	nutanixClientV3 "github.com/nutanix-cloud-native/prism-go-client/pkg/nutanix/v3"
-	corev1 "k8s.io/api/core/v1"
-	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/klog/v2"
 
 	machinecontroller "github.com/openshift/machine-api-operator/pkg/controller/machine"
@@ -244,28 +242,6 @@ func (r *Reconciler) exists() (bool, error) {
 	return true, nil
 }
 
-// isMaster returns true if the machine is part of a cluster's control plane
-func (r *Reconciler) isMaster() (bool, error) {
-	if r.machine.Status.NodeRef == nil {
-		klog.Errorf("NodeRef not found in machine %q", r.machine.Name)
-		return false, nil
-	}
-	node := &corev1.Node{}
-	nodeKey := types.NamespacedName{
-		Name: r.machine.Status.NodeRef.Name,
-	}
-
-	err := r.client.Get(r.Context, nodeKey, node)
-	if err != nil {
-		return false, fmt.Errorf("failed to get node from machine %s", r.machine.Name)
-	}
-
-	if _, exists := node.Labels[masterLabel]; exists {
-		return true, nil
-	}
-	return false, nil
-}
-
 // setProviderID adds providerID in the machine spec
 func (r *Reconciler) setProviderID(vmUUID *string) error {
 	if vmUUID == nil {
@@ -282,34 +258,29 @@ func (r *Reconciler) setProviderID(vmUUID *string) error {
 		klog.Infof("%s: ProviderID set at machine.spec: %s", r.machine.Name, providerID)
 	}
 
-	if r.machine.Status.NodeRef == nil {
-		klog.Infof("%s: the Machine node is not ready yet.", r.machine.Name)
+	if !r.machineScope.isNodeLinked() {
+		// Node is not ready
+		klog.V(3).Infof("%s: the node is not ready yet.", r.machine.Name)
 		return nil
 	}
 
 	// update the corresponding node.Spec.ProviderID
-	nodeName := r.machine.Status.NodeRef.Name
-	if len(nodeName) == 0 {
-		nodeName = r.machine.Name
-	}
-	nodeKey := types.NamespacedName{Name: nodeName}
-	node := &corev1.Node{}
-	err := r.client.Get(r.Context, nodeKey, node)
+	node, err := r.machineScope.getNode()
 	if err != nil {
-		return fmt.Errorf("%s: failed to get node %s: %w", r.machine.Name, nodeName, err)
+		return fmt.Errorf("%s: failed to get node: %w", r.machine.Name, err)
 	}
 
 	existingNodeProviderID := node.Spec.ProviderID
 	if existingNodeProviderID == providerID {
-		klog.Infof("%s: The node %q spec.providerID is already set with value: %s", r.machine.Name, nodeName, existingNodeProviderID)
+		klog.Infof("%s: The node %q spec.providerID is already set with value: %s", r.machine.Name, node.Name, existingNodeProviderID)
 	} else {
 		node.Spec.ProviderID = providerID
 		err := r.client.Update(r.Context, node)
 		if err != nil {
-			klog.Errorf("%s: failed to update the node %q spec.providerID. error: %v", r.machine.Name, nodeName, err)
+			klog.Errorf("%s: failed to update the node %q spec.providerID. error: %v", r.machine.Name, node.Name, err)
 			return err
 		}
-		klog.Infof("%s: The node %q spec.providerID is set to: %s", r.machine.Name, nodeName, providerID)
+		klog.Infof("%s: The node %q spec.providerID is set to: %s", r.machine.Name, node.Name, providerID)
 	}
 
 	return nil