Use ProviderID to map k8s nodeName to Azure VMs.

This patch allows Kubernetes nodeName to be different from
osProfile.computerName.

Author: tootoonchian

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

@@ -275,6 +275,8 @@ type Cloud struct {
 	// nodeZones is a mapping from Zone to a sets.String of Node's names in the Zone
 	// it is updated by the nodeInformer
 	nodeZones map[string]sets.String
+	// nodeProvidersIDs maps kube node name to its Azure ProviderID
+	nodeProviderIDs map[string]string
 	// nodeResourceGroups holds nodes external resource groups
 	nodeResourceGroups map[string]string
 	// unmanagedNodes holds a list of nodes not managed by Azure cloud provider.
@@ -344,6 +346,7 @@ func NewCloudWithoutFeatureGates(configReader io.Reader) (*Cloud, error) {
 	az := &Cloud{
 		nodeNames:                sets.NewString(),
 		nodeZones:                map[string]sets.String{},
+		nodeProviderIDs:          map[string]string{},
 		nodeResourceGroups:       map[string]string{},
 		unmanagedNodes:           sets.NewString(),
 		excludeLoadBalancerNodes: sets.NewString(),
@@ -803,6 +806,11 @@ func (az *Cloud) updateNodeCaches(prevNode, newNode *v1.Node) {
 			}
 		}
 
+		// Remove from nodeProviderIDs cache.
+		if len(prevNode.Spec.ProviderID) != 0 {
+			delete(az.nodeProviderIDs, prevNode.ObjectMeta.Name)
+		}
+
 		// Remove from nodeResourceGroups cache.
 		_, ok = prevNode.ObjectMeta.Labels[externalResourceGroupLabel]
 		if ok {
@@ -836,6 +844,11 @@ func (az *Cloud) updateNodeCaches(prevNode, newNode *v1.Node) {
 			az.nodeZones[newZone].Insert(newNode.ObjectMeta.Name)
 		}
 
+		// Add to nodeProviderIDs cache.
+		if len(newNode.Spec.ProviderID) != 0 {
+			az.nodeProviderIDs[newNode.ObjectMeta.Name] = newNode.Spec.ProviderID
+		}
+
 		// Add to nodeResourceGroups cache.
 		newRG, ok := newNode.ObjectMeta.Labels[externalResourceGroupLabel]
 		if ok && len(newRG) > 0 {

staging/src/k8s.io/legacy-cloud-providers/azure/azure_controller_common.go

@@ -107,7 +107,7 @@ func (c *controllerCommon) getNodeVMSet(nodeName types.NodeName, crt azcache.Azu
 	}
 
 	// 3. If the node is managed by availability set, then return ss.availabilitySet.
-	managedByAS, err := ss.isNodeManagedByAvailabilitySet(mapNodeNameToVMName(nodeName), crt)
+	managedByAS, err := ss.isNodeManagedByAvailabilitySet(c.cloud.mapNodeNameToVMName(nodeName), crt)
 	if err != nil {
 		return nil, err
 	}

staging/src/k8s.io/legacy-cloud-providers/azure/azure_controller_standard.go

@@ -39,7 +39,7 @@ func (as *availabilitySet) AttachDisk(isManagedDisk bool, diskName, diskURI stri
 		return err
 	}
 
-	vmName := mapNodeNameToVMName(nodeName)
+	vmName := as.Cloud.mapNodeNameToVMName(nodeName)
 	nodeResourceGroup, err := as.GetNodeResourceGroup(vmName)
 	if err != nil {
 		return err
@@ -126,7 +126,7 @@ func (as *availabilitySet) DetachDisk(diskName, diskURI string, nodeName types.N
 		return nil
 	}
 
-	vmName := mapNodeNameToVMName(nodeName)
+	vmName := as.Cloud.mapNodeNameToVMName(nodeName)
 	nodeResourceGroup, err := as.GetNodeResourceGroup(vmName)
 	if err != nil {
 		return err

staging/src/k8s.io/legacy-cloud-providers/azure/azure_controller_vmss.go

@@ -34,7 +34,7 @@ import (
 // AttachDisk attaches a vhd to vm
 // the vhd must exist, can be identified by diskName, diskURI, and lun.
 func (ss *scaleSet) AttachDisk(isManagedDisk bool, diskName, diskURI string, nodeName types.NodeName, lun int32, cachingMode compute.CachingTypes, diskEncryptionSetID string, writeAcceleratorEnabled bool) error {
-	vmName := mapNodeNameToVMName(nodeName)
+	vmName := ss.Cloud.mapNodeNameToVMName(nodeName)
 	ssName, instanceID, vm, err := ss.getVmssVM(vmName, azcache.CacheReadTypeDefault)
 	if err != nil {
 		return err
@@ -121,7 +121,7 @@ func (ss *scaleSet) AttachDisk(isManagedDisk bool, diskName, diskURI string, nod
 // DetachDisk detaches a disk from host
 // the vhd can be identified by diskName or diskURI
 func (ss *scaleSet) DetachDisk(diskName, diskURI string, nodeName types.NodeName) error {
-	vmName := mapNodeNameToVMName(nodeName)
+	vmName := ss.Cloud.mapNodeNameToVMName(nodeName)
 	ssName, instanceID, vm, err := ss.getVmssVM(vmName, azcache.CacheReadTypeDefault)
 	if err != nil {
 		return err

staging/src/k8s.io/legacy-cloud-providers/azure/azure_instances.go

@@ -257,7 +257,7 @@ func (az *Cloud) InstanceShutdownByProviderID(ctx context.Context, providerID st
 
 func (az *Cloud) isCurrentInstance(name types.NodeName, metadataVMName string) (bool, error) {
 	var err error
-	nodeName := mapNodeNameToVMName(name)
+	nodeName := az.mapNodeNameToVMName(name)
 
 	// VMSS vmName is not same with hostname, use hostname instead.
 	if az.VMType == vmTypeVMSS {
@@ -280,7 +280,7 @@ func (az *Cloud) isCurrentInstance(name types.NodeName, metadataVMName string) (
 // InstanceID returns the cloud provider ID of the specified instance.
 // Note that if the instance does not exist or is no longer running, we must return ("", cloudprovider.InstanceNotFound)
 func (az *Cloud) InstanceID(ctx context.Context, name types.NodeName) (string, error) {
-	nodeName := mapNodeNameToVMName(name)
+	nodeName := az.mapNodeNameToVMName(name)
 	unmanaged, err := az.IsNodeUnmanaged(nodeName)
 	if err != nil {
 		return "", err
@@ -425,8 +425,15 @@ func (az *Cloud) CurrentNodeName(ctx context.Context, hostname string) (types.No
 	return types.NodeName(hostname), nil
 }
 
-// mapNodeNameToVMName maps a k8s NodeName to an Azure VM Name
-// This is a simple string cast.
-func mapNodeNameToVMName(nodeName types.NodeName) string {
-	return string(nodeName)
+// mapNodeNameToVMName maps a k8s NodeName to an Azure VM Name using  providerID.
+func (az *Cloud) mapNodeNameToVMName(nodeName types.NodeName) string {
+	vmName := string(nodeName)
+	providerID, ok := az.nodeProviderIDs[vmName]
+	if ok {
+		if name, err := az.VMSet.GetNodeNameByProviderID(providerID); err == nil {
+			vmName = string(name)
+		}
+	}
+
+	return vmName
 }

staging/src/k8s.io/legacy-cloud-providers/azure/azure_standard.go

@@ -818,7 +818,7 @@ func (as *availabilitySet) getPrimaryInterfaceWithVMSet(nodeName, vmSetName stri
 // EnsureHostInPool ensures the given VM's Primary NIC's Primary IP Configuration is
 // participating in the specified LoadBalancer Backend Pool.
 func (as *availabilitySet) EnsureHostInPool(service *v1.Service, nodeName types.NodeName, backendPoolID string, vmSetName string, isInternal bool) (string, string, string, *compute.VirtualMachineScaleSetVM, error) {
-	vmName := mapNodeNameToVMName(nodeName)
+	vmName := as.Cloud.mapNodeNameToVMName(nodeName)
 	serviceName := getServiceName(service)
 	nic, _, err := as.getPrimaryInterfaceWithVMSet(vmName, vmSetName)
 	if err != nil {
@@ -989,7 +989,7 @@ func (as *availabilitySet) EnsureBackendPoolDeleted(service *v1.Service, backend
 			continue
 		}
 
-		vmName := mapNodeNameToVMName(types.NodeName(nodeName))
+		vmName := as.Cloud.mapNodeNameToVMName(types.NodeName(nodeName))
 		nic, vmasID, err := as.getPrimaryInterfaceWithVMSet(vmName, vmSetName)
 		if err != nil {
 			if err == errNotInVMSet {

staging/src/k8s.io/legacy-cloud-providers/azure/azure_vmss.go

@@ -695,11 +695,13 @@ func (ss *scaleSet) getNodeIdentityByNodeName(nodeName string, crt azcache.Azure
 		return node, nil
 	}
 
-	if _, err := getScaleSetVMInstanceID(nodeName); err != nil {
+	vmName := ss.Cloud.mapNodeNameToVMName(types.NodeName(nodeName))
+
+	if _, err := getScaleSetVMInstanceID(vmName); err != nil {
 		return nil, err
 	}
 
-	node, err := getter(nodeName, crt)
+	node, err := getter(vmName, crt)
 	if err != nil {
 		return nil, err
 	}
@@ -708,7 +710,7 @@ func (ss *scaleSet) getNodeIdentityByNodeName(nodeName string, crt azcache.Azure
 	}
 
 	klog.V(2).Infof("Couldn't find VMSS for node %s, refreshing the cache", nodeName)
-	node, err = getter(nodeName, azcache.CacheReadTypeForceRefresh)
+	node, err = getter(vmName, azcache.CacheReadTypeForceRefresh)
 	if err != nil {
 		return nil, err
 	}
@@ -967,7 +969,7 @@ func (ss *scaleSet) getConfigForScaleSetByIPFamily(config *compute.VirtualMachin
 // participating in the specified LoadBalancer Backend Pool, which returns (resourceGroup, vmssName, instanceID, vmssVM, error).
 func (ss *scaleSet) EnsureHostInPool(service *v1.Service, nodeName types.NodeName, backendPoolID string, vmSetName string, isInternal bool) (string, string, string, *compute.VirtualMachineScaleSetVM, error) {
 	klog.V(3).Infof("ensuring node %q of scaleset %q in LB backendpool %q", nodeName, vmSetName, backendPoolID)
-	vmName := mapNodeNameToVMName(nodeName)
+	vmName := ss.Cloud.mapNodeNameToVMName(nodeName)
 	ssName, instanceID, vm, err := ss.getVmssVM(vmName, azcache.CacheReadTypeDefault)
 	if err != nil {
 		if errors.Is(err, cloudprovider.InstanceNotFound) {