Add Versioned Hash

pkg/apis/v1beta1/ec2nodeclass.go

@@ -329,6 +329,12 @@ type EC2NodeClass struct {
 	Status EC2NodeClassStatus `json:"status,omitempty"`
 }
 
+// We need to bump the EC2NodeClassHashVersion under these condition when we make an update the EC2NodeClass CRD
+// 1. A EC2NodeClass adds default values to existing fields that are included in the hash calculations
+// 2. A EC2NodeClass field is added to the hash calculation with an already-set value
+// 3. A EC2NodeClass field is removed from the hash calculations
+const EC2NodeClassHashVersion = "v1"
+
 func (in *EC2NodeClass) Hash() string {
 	return fmt.Sprint(lo.Must(hashstructure.Hash(in.Spec, hashstructure.FormatV2, &hashstructure.HashOptions{
 		SlicesAsSets:    true,

pkg/apis/v1beta1/labels.go

@@ -113,6 +113,7 @@ var (
 	LabelInstanceAcceleratorManufacturer      = Group + "/instance-accelerator-manufacturer"
 	LabelInstanceAcceleratorCount             = Group + "/instance-accelerator-count"
 	AnnotationEC2NodeClassHash                = Group + "/ec2nodeclass-hash"
+	AnnotationEC2NodeClassHashVersion         = Group + "/ec2nodeclass-hash-version"
 	AnnotationInstanceTagged                  = Group + "/tagged"
 
 	TagNodeClaim = v1beta1.Group + "/nodeclaim"

pkg/cloudprovider/cloudprovider.go

@@ -101,7 +101,10 @@ func (c *CloudProvider) Create(ctx context.Context, nodeClaim *corev1beta1.NodeC
 		return i.Name == instance.Type
 	})
 	nc := c.instanceToNodeClaim(instance, instanceType)
-	nc.Annotations = lo.Assign(nc.Annotations, map[string]string{v1beta1.AnnotationEC2NodeClassHash: nodeClass.Hash()})
+	nc.Annotations = lo.Assign(nodeClass.Annotations, map[string]string{
+		v1beta1.AnnotationEC2NodeClassHash:        nodeClass.Hash(),
+		v1beta1.AnnotationEC2NodeClassHashVersion: v1beta1.EC2NodeClassHashVersion,
+	})
 	return nc, nil
 }
 

pkg/cloudprovider/drift.go

@@ -136,8 +136,15 @@ func (c *CloudProvider) areSecurityGroupsDrifted(ctx context.Context, ec2Instanc
 
 func (c *CloudProvider) areStaticFieldsDrifted(nodeClaim *corev1beta1.NodeClaim, nodeClass *v1beta1.EC2NodeClass) cloudprovider.DriftReason {
 	nodeClassHash, foundHashNodeClass := nodeClass.Annotations[v1beta1.AnnotationEC2NodeClassHash]
+	nodeClassHashVersion, foundHashNodeClassVersion := nodeClass.Annotations[v1beta1.AnnotationEC2NodeClassHashVersion]
 	nodeClaimHash, foundHashNodeClaim := nodeClaim.Annotations[v1beta1.AnnotationEC2NodeClassHash]
-	if !foundHashNodeClass || !foundHashNodeClaim {
+	nodeClaimHashVersion, foundHashNodeClaimVersion := nodeClaim.Annotations[v1beta1.AnnotationEC2NodeClassHashVersion]
+
+	if !foundHashNodeClass || !foundHashNodeClaim || !foundHashNodeClassVersion || !foundHashNodeClaimVersion {
+		return ""
+	}
+	// validate that the version of the crd is the same
+	if nodeClassHashVersion != nodeClaimHashVersion {
 		return ""
 	}
 	return lo.Ternary(nodeClassHash != nodeClaimHash, NodeClassDrift, "")

pkg/cloudprovider/suite_test.go

@@ -593,10 +593,14 @@ var _ = Describe("CloudProvider", func() {
 				Reservations: []*ec2.Reservation{{Instances: []*ec2.Instance{instance}}},
 			})
 			nodeClass.Annotations = lo.Assign(nodeClass.Annotations, map[string]string{
-				v1beta1.AnnotationEC2NodeClassHash: nodeClass.Hash(),
+				v1beta1.AnnotationEC2NodeClassHash:        nodeClass.Hash(),
+				v1beta1.AnnotationEC2NodeClassHashVersion: v1beta1.EC2NodeClassHashVersion,
 			})
 			nodeClaim.Status.ProviderID = fake.ProviderID(lo.FromPtr(instance.InstanceId))
-			nodeClaim.Annotations = lo.Assign(nodeClaim.Annotations, map[string]string{v1beta1.AnnotationEC2NodeClassHash: nodeClass.Hash()})
+			nodeClaim.Annotations = lo.Assign(nodeClaim.Annotations, map[string]string{
+				v1beta1.AnnotationEC2NodeClassHash:        nodeClass.Hash(),
+				v1beta1.AnnotationEC2NodeClassHashVersion: v1beta1.EC2NodeClassHashVersion,
+			})
 			nodeClaim.Labels = lo.Assign(nodeClaim.Labels, map[string]string{v1.LabelInstanceTypeStable: selectedInstanceType.Name})
 		})
 		It("should not fail if NodeClass does not exist", func() {
@@ -770,6 +774,20 @@ var _ = Describe("CloudProvider", func() {
 				Expect(err).NotTo(HaveOccurred())
 				Expect(isDrifted).To(BeEmpty())
 			})
+			It("should not return drifted if karpenter.sh/ec2nodeclass-hash-version annotation version does not match the EC2NodeClass on the NodeClaim", func() {
+				nodeClass.ObjectMeta.Annotations = map[string]string{
+					v1beta1.AnnotationEC2NodeClassHash:        "test-hash-1",
+					v1beta1.AnnotationEC2NodeClassHashVersion: "test-version-1",
+				}
+				nodeClaim.ObjectMeta.Annotations = map[string]string{
+					v1beta1.AnnotationEC2NodeClassHash:        "test-hash-1",
+					v1beta1.AnnotationEC2NodeClassHashVersion: "test-version-2",
+				}
+				ExpectApplied(ctx, env.Client, nodePool, nodeClaim)
+				isDrifted, err := cloudProvider.IsDrifted(ctx, nodeClaim)
+				Expect(err).NotTo(HaveOccurred())
+				Expect(isDrifted).To(BeEmpty())
+			})
 		})
 	})
 	Context("Subnet Compatibility", func() {

pkg/controllers/nodeclass/controller.go

@@ -80,7 +80,17 @@ func NewController(kubeClient client.Client, recorder events.Recorder, subnetPro
 func (c *Controller) Reconcile(ctx context.Context, nodeClass *v1beta1.EC2NodeClass) (reconcile.Result, error) {
 	stored := nodeClass.DeepCopy()
 	controllerutil.AddFinalizer(nodeClass, v1beta1.TerminationFinalizer)
-	nodeClass.Annotations = lo.Assign(nodeClass.Annotations, map[string]string{v1beta1.AnnotationEC2NodeClassHash: nodeClass.Hash()})
+
+	if nodeClass.Annotations[v1beta1.AnnotationEC2NodeClassHashVersion] != v1beta1.EC2NodeClassHashVersion {
+		if err := c.updateNodeClaimHash(ctx, nodeClass); err != nil {
+			return reconcile.Result{}, err
+		}
+	}
+	nodeClass.Annotations = lo.Assign(nodeClass.Annotations, map[string]string{
+		v1beta1.AnnotationEC2NodeClassHash:        nodeClass.Hash(),
+		v1beta1.AnnotationEC2NodeClassHashVersion: v1beta1.EC2NodeClassHashVersion,
+	})
+
 	err := multierr.Combine(
 		c.resolveSubnets(ctx, nodeClass),
 		c.resolveSecurityGroups(ctx, nodeClass),
@@ -222,6 +232,39 @@ func (c *Controller) resolveInstanceProfile(ctx context.Context, nodeClass *v1be
 	return nil
 }
 
+func (c *Controller) updateNodeClaimHash(ctx context.Context, nodeClass *v1beta1.EC2NodeClass) error {
+	ncList := &corev1beta1.NodeClaimList{}
+	if err := c.kubeClient.List(ctx, ncList, client.MatchingFields{"spec.nodeClassRef.name": nodeClass.Name}); err != nil {
+		return err
+	}
+
+	for i := range ncList.Items {
+		stored := ncList.Items[i].DeepCopy()
+
+		if ncList.Items[i].Annotations[v1beta1.EC2NodeClassHashVersion] != v1beta1.EC2NodeClassHashVersion {
+			ncList.Items[i].Annotations = lo.Assign(ncList.Items[i].Annotations, map[string]string{
+				v1beta1.AnnotationEC2NodeClassHashVersion: v1beta1.EC2NodeClassHashVersion,
+			})
+
+			// Any NodeClaim that is already drifted will remain drifted if the karpenter.sh/ec2nodeclass-hash-version doesn't match
+			// Since the hashing mechanism has changed we will not be able to determine if the drifted status of the node has changed
+			if ncList.Items[i].StatusConditions().GetCondition(corev1beta1.Drifted) == nil {
+				ncList.Items[i].Annotations = lo.Assign(ncList.Items[i].Annotations, map[string]string{
+					v1beta1.AnnotationEC2NodeClassHash: nodeClass.Hash(),
+				})
+			}
+
+			if !equality.Semantic.DeepEqual(stored, ncList.Items[i]) {
+				if err := c.kubeClient.Patch(ctx, &ncList.Items[i], client.MergeFrom(stored)); err != nil {
+					return client.IgnoreNotFound(err)
+				}
+			}
+		}
+	}
+
+	return nil
+}
+
 func (c *Controller) Name() string {
 	return "nodeclass"
 }

pkg/controllers/nodeclass/suite_test.go

@@ -23,6 +23,7 @@ import (
 	"github.com/imdario/mergo"
 	"github.com/samber/lo"
 	v1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/client-go/tools/record"
 	_ "knative.dev/pkg/system/testing"
 	"sigs.k8s.io/controller-runtime/pkg/client"
@@ -833,7 +834,7 @@ var _ = Describe("NodeClassController", func() {
 		})
 	})
 	Context("Static Drift Hash", func() {
-		DescribeTable("should update the static drift hash when static field is updated", func(changes *v1beta1.EC2NodeClass) {
+		DescribeTable("should update the drift hash when static field is updated", func(changes *v1beta1.EC2NodeClass) {
 			ExpectApplied(ctx, env.Client, nodeClass)
 			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
 			nodeClass = ExpectExists(ctx, env.Client, nodeClass)
@@ -860,7 +861,7 @@ var _ = Describe("NodeClassController", func() {
 			Entry("MetadataOptions Drift", &v1beta1.EC2NodeClass{Spec: v1beta1.EC2NodeClassSpec{MetadataOptions: &v1beta1.MetadataOptions{HTTPEndpoint: aws.String("disabled")}}}),
 			Entry("Context Drift", &v1beta1.EC2NodeClass{Spec: v1beta1.EC2NodeClassSpec{Context: aws.String("context-2")}}),
 		)
-		It("should not update the static drift hash when dynamic field is updated", func() {
+		It("should not update the drift hash when dynamic field is updated", func() {
 			ExpectApplied(ctx, env.Client, nodeClass)
 			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
 			nodeClass = ExpectExists(ctx, env.Client, nodeClass)
@@ -889,6 +890,118 @@ var _ = Describe("NodeClassController", func() {
 			nodeClass = ExpectExists(ctx, env.Client, nodeClass)
 			Expect(nodeClass.Annotations[v1beta1.AnnotationEC2NodeClassHash]).To(Equal(expectedHash))
 		})
+		It("should update ec2nodeclass hash version when the ec2nodeclass hash version is out of sync with the controller", func() {
+			nodeClass.Annotations = map[string]string{v1beta1.AnnotationEC2NodeClassHash: "test"}
+			ExpectApplied(ctx, env.Client, nodeClass)
+
+			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
+			nodeClass = ExpectExists(ctx, env.Client, nodeClass)
+
+			expectedHash := nodeClass.Hash()
+			Expect(nodeClass.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+			Expect(nodeClass.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+		})
+		It("should update ec2nodeclass hash versions on all nodeclaims when the hash versions don't match", func() {
+			nodeClaimOne := coretest.NodeClaim(corev1beta1.NodeClaim{
+				ObjectMeta: metav1.ObjectMeta{
+					Annotations: map[string]string{
+						v1beta1.AnnotationEC2NodeClassHash:        "123456",
+						v1beta1.AnnotationEC2NodeClassHashVersion: "test",
+					},
+				},
+				Spec: corev1beta1.NodeClaimSpec{
+					NodeClassRef: &corev1beta1.NodeClassReference{
+						Name: nodeClass.Name,
+					},
+				},
+			})
+			nodeClaimTwo := coretest.NodeClaim(corev1beta1.NodeClaim{
+				ObjectMeta: metav1.ObjectMeta{
+					Annotations: map[string]string{
+						v1beta1.AnnotationEC2NodeClassHash:        "123456",
+						v1beta1.AnnotationEC2NodeClassHashVersion: "test",
+					},
+				},
+				Spec: corev1beta1.NodeClaimSpec{
+					NodeClassRef: &corev1beta1.NodeClassReference{
+						Name: nodeClass.Name,
+					},
+				},
+			})
+
+			ExpectApplied(ctx, env.Client, nodeClass, nodeClaimOne, nodeClaimTwo)
+
+			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
+			nodeClass = ExpectExists(ctx, env.Client, nodeClass)
+			nodeClaimOne = ExpectExists(ctx, env.Client, nodeClaimOne)
+			nodeClaimTwo = ExpectExists(ctx, env.Client, nodeClaimTwo)
+
+			expectedHash := nodeClass.Hash()
+			Expect(nodeClaimOne.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+			Expect(nodeClaimOne.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+			Expect(nodeClaimTwo.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+			Expect(nodeClaimTwo.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+		})
+		It("should not update ec2nodeclass hash on all nodeclaims when the hash versions match", func() {
+			nodeClaim := coretest.NodeClaim(corev1beta1.NodeClaim{
+				ObjectMeta: metav1.ObjectMeta{
+					Annotations: map[string]string{
+						v1beta1.AnnotationEC2NodeClassHash:        "1234564654",
+						v1beta1.AnnotationEC2NodeClassHashVersion: "test-version",
+					},
+				},
+				Spec: corev1beta1.NodeClaimSpec{
+					NodeClassRef: &corev1beta1.NodeClassReference{
+						Name: nodeClass.Name,
+					},
+				},
+			})
+			ExpectApplied(ctx, env.Client, nodeClass, nodeClaim)
+
+			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
+			nodeClass = ExpectExists(ctx, env.Client, nodeClass)
+			nodeClaim = ExpectExists(ctx, env.Client, nodeClaim)
+
+			expectedHash := nodeClass.Hash()
+
+			// Expect NodeClaims to have been updated to the original hash
+			Expect(nodeClass.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+			Expect(nodeClass.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+			Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+			Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+
+			nodeClaim.Annotations[v1beta1.AnnotationEC2NodeClassHash] = "test-hash"
+			ExpectApplied(ctx, env.Client, nodeClass, nodeClaim)
+
+			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
+
+			Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, "test-hash"))
+			Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+		})
+		It("should not update ec2nodeclass hash on the nodeclaim if it's drifted", func() {
+			nodeClaim := coretest.NodeClaim(corev1beta1.NodeClaim{
+				ObjectMeta: metav1.ObjectMeta{
+					Annotations: map[string]string{
+						v1beta1.AnnotationEC2NodeClassHash:        "123456",
+						v1beta1.AnnotationEC2NodeClassHashVersion: "test",
+					},
+				},
+				Spec: corev1beta1.NodeClaimSpec{
+					NodeClassRef: &corev1beta1.NodeClassReference{
+						Name: nodeClass.Name,
+					},
+				},
+			})
+			nodeClaim.StatusConditions().MarkTrue(corev1beta1.Drifted)
+			ExpectApplied(ctx, env.Client, nodeClaim)
+
+			ExpectReconcileSucceeded(ctx, nodeClassController, client.ObjectKeyFromObject(nodeClass))
+			nodeClaim = ExpectExists(ctx, env.Client, nodeClaim)
+
+			// Expect NodeClaims to have been updated to the original hash
+			Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, "123456"))
+			Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, "test"))
+		})
 	})
 	Context("NodeClass Termination", func() {
 		var profileName string

test/suites/drift/suite_test.go

@@ -719,6 +719,23 @@ var _ = Describe("Drift", func() {
 		env.EventuallyExpectNotFound(pod, node)
 		env.EventuallyExpectHealthyPodCount(selector, numPods)
 	})
+	It("should update NodeClass hash on the NodeClass and NodeClaim", func() {
+		nodeClass.Annotations = map[string]string{v1beta1.AnnotationEC2NodeClassHashVersion: "test-version"}
+		nodeClass.Annotations = map[string]string{v1beta1.AnnotationEC2NodeClassHash: "test-adfdhhcnkdskj"}
+
+		env.ExpectCreated(dep, nodeClass, nodePool)
+		env.EventuallyExpectHealthyPodCount(selector, numPods)
+		nodeClaim := env.EventuallyExpectCreatedNodeClaimCount("==", 1)[0]
+		err := env.Client.Get(env.Context, client.ObjectKeyFromObject(nodeClass), nodeClass)
+		Expect(err).ToNot(HaveOccurred())
+
+		expectedHash := nodeClass.Hash()
+
+		Expect(nodeClass.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+		Expect(nodeClass.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+		Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHash, expectedHash))
+		Expect(nodeClaim.Annotations).To(HaveKeyWithValue(v1beta1.AnnotationEC2NodeClassHashVersion, v1beta1.EC2NodeClassHashVersion))
+	})
 	Context("Failure", func() {
 		It("should not continue to drift if a node never registers", func() {
 			// launch a new nodeClaim