Add support for removing unsupported huge page sizes

When kubelet is restarted, it will now remove the resources for huge
page sizes no longer supported. This is required when:
- node disables huge pages
- changing the default huge page size in older versions of linux
(because it will then only support the newly set default).
- Software updates that change what sizes are supported (eg. by changing
boot parameters).

pkg/kubelet/kubelet_node_status.go

@@ -31,6 +31,7 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
+	"k8s.io/apimachinery/pkg/util/sets"
 	cloudprovider "k8s.io/cloud-provider"
 	cloudproviderapi "k8s.io/cloud-provider/api"
 	"k8s.io/klog"
@@ -117,6 +118,7 @@ func (kl *Kubelet) tryRegisterWithAPIServer(node *v1.Node) bool {
 	requiresUpdate := kl.reconcileCMADAnnotationWithExistingNode(node, existingNode)
 	requiresUpdate = kl.updateDefaultLabels(node, existingNode) || requiresUpdate
 	requiresUpdate = kl.reconcileExtendedResource(node, existingNode) || requiresUpdate
+	requiresUpdate = kl.reconcileHugePageResource(node, existingNode) || requiresUpdate
 	if requiresUpdate {
 		if _, _, err := nodeutil.PatchNodeStatus(kl.kubeClient.CoreV1(), types.NodeName(kl.nodeName), originalNode, existingNode); err != nil {
 			klog.Errorf("Unable to reconcile node %q with API server: error updating node: %v", kl.nodeName, err)
@@ -127,6 +129,53 @@ func (kl *Kubelet) tryRegisterWithAPIServer(node *v1.Node) bool {
 	return true
 }
 
+// reconcileHugePageResource will update huge page capacity for each page size and remove huge page sizes no longer supported
+func (kl *Kubelet) reconcileHugePageResource(initialNode, existingNode *v1.Node) bool {
+	requiresUpdate := false
+	supportedHugePageResources := sets.String{}
+
+	for resourceName := range initialNode.Status.Capacity {
+		if !v1helper.IsHugePageResourceName(resourceName) {
+			continue
+		}
+		supportedHugePageResources.Insert(string(resourceName))
+
+		initialCapacity := initialNode.Status.Capacity[resourceName]
+		initialAllocatable := initialNode.Status.Allocatable[resourceName]
+
+		capacity, resourceIsSupported := existingNode.Status.Capacity[resourceName]
+		allocatable := existingNode.Status.Allocatable[resourceName]
+
+		// Add or update capacity if it the size was previously unsupported or has changed
+		if !resourceIsSupported || capacity.Cmp(initialCapacity) != 0 {
+			existingNode.Status.Capacity[resourceName] = initialCapacity.DeepCopy()
+			requiresUpdate = true
+		}
+
+		// Add or update allocatable if it the size was previously unsupported or has changed
+		if !resourceIsSupported || allocatable.Cmp(initialAllocatable) != 0 {
+			existingNode.Status.Allocatable[resourceName] = initialAllocatable.DeepCopy()
+			requiresUpdate = true
+		}
+
+	}
+
+	for resourceName := range existingNode.Status.Capacity {
+		if !v1helper.IsHugePageResourceName(resourceName) {
+			continue
+		}
+
+		// If huge page size no longer is supported, we remove it from the node
+		if !supportedHugePageResources.Has(string(resourceName)) {
+			delete(existingNode.Status.Capacity, resourceName)
+			delete(existingNode.Status.Allocatable, resourceName)
+			klog.Infof("Removing now unsupported huge page resource named: %s", resourceName)
+			requiresUpdate = true
+		}
+	}
+	return requiresUpdate
+}
+
 // Zeros out extended resource capacity during reconciliation.
 func (kl *Kubelet) reconcileExtendedResource(initialNode, node *v1.Node) bool {
 	requiresUpdate := false