Add Support for Node Monitoring Agent

aws · Dec 20, 2024 · cb4b833 · cb4b833
1 parent c023ab7
commit cb4b833
Show file tree

Hide file tree

Showing 2 changed files with 55 additions and 1 deletion.
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{