feat: Add Support for Node Monitoring Agent (#7545)

aws · Jan 10, 2025 · 10202fe · 10202fe
1 parent b631d9e
commit 10202fe
Show file tree

Hide file tree

Showing 6 changed files with 66 additions and 4 deletions.
diff --git a/charts/karpenter-crd/templates/karpenter.sh_nodeclaims.yaml b/charts/karpenter-crd/templates/karpenter.sh_nodeclaims.yaml
@@ -38,6 +38,10 @@ spec:
         - jsonPath: .metadata.creationTimestamp
           name: Age
           type: date
+        - jsonPath: .status.imageID
+          name: ImageID
+          priority: 1
+          type: string
         - jsonPath: .status.providerID
           name: ID
           priority: 1

diff --git a/go.mod b/go.mod
@@ -43,7 +43,7 @@ require (
 	k8s.io/klog/v2 v2.130.1
 	k8s.io/utils v0.0.0-20241104100929-3ea5e8cea738
 	sigs.k8s.io/controller-runtime v0.19.3
-	sigs.k8s.io/karpenter v1.1.1
+	sigs.k8s.io/karpenter v1.1.2-0.20241220005608-b3fa6ebffc19
 	sigs.k8s.io/yaml v1.4.0
 )
 

diff --git a/go.sum b/go.sum
@@ -339,8 +339,8 @@ sigs.k8s.io/controller-runtime v0.19.3 h1:XO2GvC9OPftRst6xWCpTgBZO04S2cbp0Qqkj8b
 sigs.k8s.io/controller-runtime v0.19.3/go.mod h1:j4j87DqtsThvwTv5/Tc5NFRyyF/RF0ip4+62tbTSIUM=
 sigs.k8s.io/json v0.0.0-20241010143419-9aa6b5e7a4b3 h1:/Rv+M11QRah1itp8VhT6HoVx1Ray9eB4DBr+K+/sCJ8=
 sigs.k8s.io/json v0.0.0-20241010143419-9aa6b5e7a4b3/go.mod h1:18nIHnGi6636UCz6m8i4DhaJ65T6EruyzmoQqI2BVDo=
-sigs.k8s.io/karpenter v1.1.1 h1:QPpVC8DsaLgJ/YWcFpZKE4m3jD+Qp88/GtSPvMfffck=
-sigs.k8s.io/karpenter v1.1.1/go.mod h1:NQouOJNK6s1d4EIKa5cY7nAV3IG74qZ6gPzHBeCZNPw=
+sigs.k8s.io/karpenter v1.1.2-0.20241220005608-b3fa6ebffc19 h1:nCaZE6O7772FEEPGgTef05IanE8AWMKf7DBh1LiU1ik=
+sigs.k8s.io/karpenter v1.1.2-0.20241220005608-b3fa6ebffc19/go.mod h1:E1mtCutIoQJA05ClYYQo9y+5ujk6U2FxByauGSUXXZs=
 sigs.k8s.io/structured-merge-diff/v4 v4.4.2 h1:MdmvkGuXi/8io6ixD5wud3vOLwc1rj0aNqRlpuvjmwA=
 sigs.k8s.io/structured-merge-diff/v4 v4.4.2/go.mod h1:N8f93tFZh9U6vpxwRArLiikrE5/2tiu1w1AGfACIGE4=
 sigs.k8s.io/yaml v1.4.0 h1:Mk1wCc2gy/F0THH0TAp1QYyJNzRm2KCLy3o5ASXVI5E=

diff --git a/pkg/apis/crds/karpenter.sh_nodeclaims.yaml b/pkg/apis/crds/karpenter.sh_nodeclaims.yaml
@@ -35,6 +35,10 @@ spec:
         - jsonPath: .metadata.creationTimestamp
           name: Age
           type: date
+        - jsonPath: .status.imageID
+          name: ImageID
+          priority: 1
+          type: string
         - jsonPath: .status.providerID
           name: ID
           priority: 1

diff --git a/pkg/cloudprovider/cloudprovider.go b/pkg/cloudprovider/cloudprovider.go
@@ -259,7 +259,7 @@ func getTags(ctx context.Context, nodeClass *v1.EC2NodeClass, nodeClaim *karpv1.
 
 func (c *CloudProvider) RepairPolicies() []cloudprovider.RepairPolicy {
 	return []cloudprovider.RepairPolicy{
-		// Supported Kubelet fields
+		// Supported Kubelet Node Conditions
 		{
 			ConditionType:      corev1.NodeReady,
 			ConditionStatus:    corev1.ConditionFalse,
@@ -270,6 +270,33 @@ func (c *CloudProvider) RepairPolicies() []cloudprovider.RepairPolicy {
 			ConditionStatus:    corev1.ConditionUnknown,
 			TolerationDuration: 30 * time.Minute,
 		},
+		// Support Node Monitoring Agent Conditions
+		//
+		{
+			ConditionType:      "AcceleratedHardwareReady",
+			ConditionStatus:    corev1.ConditionFalse,
+			TolerationDuration: 10 * time.Minute,
+		},
+		{
+			ConditionType:      "StorageReady",
+			ConditionStatus:    corev1.ConditionFalse,
+			TolerationDuration: 30 * time.Minute,
+		},
+		{
+			ConditionType:      "NetworkingReady",
+			ConditionStatus:    corev1.ConditionFalse,
+			TolerationDuration: 30 * time.Minute,
+		},
+		{
+			ConditionType:      "KernelReady",
+			ConditionStatus:    corev1.ConditionFalse,
+			TolerationDuration: 30 * time.Minute,
+		},
+		{
+			ConditionType:      "ContainerRuntimeReady",
+			ConditionStatus:    corev1.ConditionFalse,
+			TolerationDuration: 30 * time.Minute,
+		},
 	}
 }
 

diff --git a/test/suites/integration/repair_policy_test.go b/test/suites/integration/repair_policy_test.go
@@ -73,6 +73,7 @@ var _ = Describe("Repair Policy", func() {
 		env.EventuallyExpectNotFound(pod, node)
 		env.EventuallyExpectHealthyPodCount(selector, numPods)
 	},
+		// Kubelet Supported Conditions
 		Entry("Node Ready False", corev1.NodeCondition{
 			Type:               corev1.NodeReady,
 			Status:             corev1.ConditionFalse,
@@ -83,6 +84,32 @@ var _ = Describe("Repair Policy", func() {
 			Status:             corev1.ConditionUnknown,
 			LastTransitionTime: metav1.Time{Time: time.Now().Add(-31 * time.Minute)},
 		}),
+		// Node Monitoring Agent Supported Conditions
+		Entry("Node AcceleratedHardwareReady False", corev1.NodeCondition{
+			Type:               "AcceleratedHardwareReady",
+			Status:             corev1.ConditionFalse,
+			LastTransitionTime: metav1.Time{Time: time.Now().Add(-11 * time.Minute)},
+		}),
+		Entry("Node StorageReady False", corev1.NodeCondition{
+			Type:               "StorageReady",
+			Status:             corev1.ConditionFalse,
+			LastTransitionTime: metav1.Time{Time: time.Now().Add(-31 * time.Minute)},
+		}),
+		Entry("Node NetworkingReady False", corev1.NodeCondition{
+			Type:               "NetworkingReady",
+			Status:             corev1.ConditionFalse,
+			LastTransitionTime: metav1.Time{Time: time.Now().Add(-31 * time.Minute)},
+		}),
+		Entry("Node KernelReady False", corev1.NodeCondition{
+			Type:               "KernelReady",
+			Status:             corev1.ConditionFalse,
+			LastTransitionTime: metav1.Time{Time: time.Now().Add(-31 * time.Minute)},
+		}),
+		Entry("Node ContainerRuntimeReady False", corev1.NodeCondition{
+			Type:               "ContainerRuntimeReady",
+			Status:             corev1.ConditionFalse,
+			LastTransitionTime: metav1.Time{Time: time.Now().Add(-31 * time.Minute)},
+		}),
 	)
 	It("should ignore disruption budgets", func() {
 		nodePool.Spec.Disruption.Budgets = []karpenterv1.Budget{