OCPBUGS-81544: requeue when AutoNodeEnabled is progressing

hypershift-operator/controllers/hostedcluster/hostedcluster_controller.go

@@ -847,8 +847,8 @@ func (r *HostedClusterReconciler) reconcile(ctx context.Context, req ctrl.Reques
 	propagateControlPlaneVersion(hcluster, hcp)
 
 	// Set the AutoNodeEnabled condition reflecting both spec intent and actual component rollout progress.
-	meta.SetStatusCondition(&hcluster.Status.Conditions,
-		r.reconcileAutoNodeEnabledCondition(ctx, hcluster, controlPlaneNamespace.Name))
+	autoNodeCondition, autoNodeProgressing := r.reconcileAutoNodeEnabledCondition(ctx, hcluster, controlPlaneNamespace.Name)
+	meta.SetStatusCondition(&hcluster.Status.Conditions, autoNodeCondition)
 
 	// Copy the AWSDefaultSecurityGroupCreated condition from the hostedcontrolplane
 	if hcluster.Spec.Platform.Type == hyperv1.AWSPlatform {
@@ -2103,6 +2103,12 @@ func (r *HostedClusterReconciler) reconcile(ctx context.Context, req ctrl.Reques
 	if requeueAfter != nil {
 		result.RequeueAfter = *requeueAfter
 	}
+	if autoNodeProgressing {
+		autoNodeRequeue := 15 * time.Second
+		if result.RequeueAfter == 0 || autoNodeRequeue < result.RequeueAfter {
+			result.RequeueAfter = autoNodeRequeue
+		}
+	}
 	return result, nil
 }
 

hypershift-operator/controllers/hostedcluster/karpenter.go

@@ -164,7 +164,13 @@ func isKASAvailable(ctx context.Context, cpNamespace string, c client.Client) (b
 //   - True  / AsExpected          — Karpenter enabled in spec AND both components fully rolled out.
 //   - False / AutoNodeProgressing — Enable or disable operation is in progress.
 //   - False / AutoNodeNotConfigured — Karpenter not in spec AND no components present.
-func (r *HostedClusterReconciler) reconcileAutoNodeEnabledCondition(ctx context.Context, hcluster *hyperv1.HostedCluster, hcpNamespace string) metav1.Condition {
+//
+// reconcileAutoNodeEnabledCondition returns the AutoNodeEnabled condition and whether
+// the caller should requeue to poll for progress. The second return value is true when
+// an enable or disable operation is still in flight; the HC reconciler does not watch
+// ControlPlaneComponent resources, so a periodic requeue is needed to pick up status
+// changes from the karpenter-operator's CPC updates.
+func (r *HostedClusterReconciler) reconcileAutoNodeEnabledCondition(ctx context.Context, hcluster *hyperv1.HostedCluster, hcpNamespace string) (metav1.Condition, bool) {
 	condition := metav1.Condition{
 		Type:               string(hyperv1.AutoNodeEnabled),
 		ObservedGeneration: hcluster.Generation,
@@ -178,7 +184,7 @@ func (r *HostedClusterReconciler) reconcileAutoNodeEnabledCondition(ctx context.
 		condition.Status = metav1.ConditionUnknown
 		condition.Reason = hyperv1.AutoNodeEvaluationFailedReason
 		condition.Message = fmt.Sprintf("failed to list ControlPlaneComponents: %v", err)
-		return condition
+		return condition, false
 	}
 
 	// Grab all of our karpenter components
@@ -195,7 +201,7 @@ func (r *HostedClusterReconciler) reconcileAutoNodeEnabledCondition(ctx context.
 			condition.Status = metav1.ConditionFalse
 			condition.Reason = hyperv1.AutoNodeProgressingReason
 			condition.Message = "AutoNode is being enabled: waiting for components to be created"
-			return condition
+			return condition, true
 		}
 		// Check if they're ready
 		var notReady []string
@@ -214,13 +220,13 @@ func (r *HostedClusterReconciler) reconcileAutoNodeEnabledCondition(ctx context.
 			condition.Status = metav1.ConditionFalse
 			condition.Reason = hyperv1.AutoNodeProgressingReason
 			condition.Message = fmt.Sprintf("AutoNode is being enabled: %s", strings.Join(notReady, "; "))
-			return condition
+			return condition, true
 		}
 		// Otherwise report ready
 		condition.Status = metav1.ConditionTrue
 		condition.Reason = hyperv1.AsExpectedReason
 		condition.Message = "AutoNode is ready"
-		return condition
+		return condition, false
 	}
 
 	// Karpenter not enabled — check if Deployments are still terminating.
@@ -236,19 +242,19 @@ func (r *HostedClusterReconciler) reconcileAutoNodeEnabledCondition(ctx context.
 			condition.Status = metav1.ConditionUnknown
 			condition.Reason = hyperv1.AutoNodeEvaluationFailedReason
 			condition.Message = fmt.Sprintf("failed to check karpenter deployments: %v", err)
-			return condition
+			return condition, false
 		}
 	}
 
 	if len(runningDeployments) > 0 {
 		condition.Status = metav1.ConditionFalse
 		condition.Reason = hyperv1.AutoNodeProgressingReason
 		condition.Message = fmt.Sprintf("AutoNode is being disabled: waiting for deployments to be removed: %s", strings.Join(runningDeployments, ", "))
-		return condition
+		return condition, true
 	}
 
 	condition.Status = metav1.ConditionFalse
 	condition.Reason = hyperv1.AutoNodeNotConfiguredReason
 	condition.Message = "AutoNode provisioner is not configured"
-	return condition
+	return condition, false
 }

hypershift-operator/controllers/hostedcluster/karpenter_test.go

@@ -297,14 +297,16 @@ func TestReconcileAutoNodeEnabledCondition(t *testing.T) {
 	}
 
 	tests := map[string]struct {
-		autoNode    hyperv1.AutoNode
-		components  []hyperv1.ControlPlaneComponent
-		deployments []appsv1.Deployment
-		want        metav1.Condition
+		autoNode       hyperv1.AutoNode
+		components     []hyperv1.ControlPlaneComponent
+		deployments    []appsv1.Deployment
+		want           metav1.Condition
+		wantProgessing bool
 	}{
 		"When karpenter is enabled and components not yet created it should report progressing": {
-			autoNode:   karpenterEnabledAutoNode,
-			components: nil,
+			autoNode:       karpenterEnabledAutoNode,
+			components:     nil,
+			wantProgessing: true,
 			want: metav1.Condition{
 				Type:   string(hyperv1.AutoNodeEnabled),
 				Status: metav1.ConditionFalse,
@@ -319,6 +321,7 @@ func TestReconcileAutoNodeEnabledCondition(t *testing.T) {
 					Status:     hyperv1.ControlPlaneComponentStatus{Conditions: []metav1.Condition{rolloutCompleteTrue}},
 				},
 			},
+			wantProgessing: true,
 			want: metav1.Condition{
 				Type:   string(hyperv1.AutoNodeEnabled),
 				Status: metav1.ConditionFalse,
@@ -337,6 +340,7 @@ func TestReconcileAutoNodeEnabledCondition(t *testing.T) {
 					Status:     hyperv1.ControlPlaneComponentStatus{Conditions: []metav1.Condition{rolloutCompleteFalse}},
 				},
 			},
+			wantProgessing: true,
 			want: metav1.Condition{
 				Type:   string(hyperv1.AutoNodeEnabled),
 				Status: metav1.ConditionFalse,
@@ -367,6 +371,7 @@ func TestReconcileAutoNodeEnabledCondition(t *testing.T) {
 				{ObjectMeta: metav1.ObjectMeta{Name: karpenterv2.ComponentName, Namespace: hcpNamespace}},
 				{ObjectMeta: metav1.ObjectMeta{Name: karpenteroperatorv2.ComponentName, Namespace: hcpNamespace}},
 			},
+			wantProgessing: true,
 			want: metav1.Condition{
 				Type:   string(hyperv1.AutoNodeEnabled),
 				Status: metav1.ConditionFalse,
@@ -378,6 +383,7 @@ func TestReconcileAutoNodeEnabledCondition(t *testing.T) {
 			deployments: []appsv1.Deployment{
 				{ObjectMeta: metav1.ObjectMeta{Name: karpenterv2.ComponentName, Namespace: hcpNamespace}},
 			},
+			wantProgessing: true,
 			want: metav1.Condition{
 				Type:   string(hyperv1.AutoNodeEnabled),
 				Status: metav1.ConditionFalse,
@@ -439,13 +445,16 @@ func TestReconcileAutoNodeEnabledCondition(t *testing.T) {
 				},
 			}
 
-			got := r.reconcileAutoNodeEnabledCondition(context.Background(), hcluster, hcpNamespace)
+			got, progressing := r.reconcileAutoNodeEnabledCondition(context.Background(), hcluster, hcpNamespace)
 			got.ObservedGeneration = 0
 			got.Message = ""
 			got.LastTransitionTime = metav1.Time{}
 
 			if !equality.Semantic.DeepEqual(tc.want, got) {
-				t.Errorf("expected %+v, got %+v", tc.want, got)
+				t.Errorf("condition: expected %+v, got %+v", tc.want, got)
+			}
+			if progressing != tc.wantProgessing {
+				t.Errorf("progressing: expected %v, got %v", tc.wantProgessing, progressing)
 			}
 		})
 	}