add initialize conditions to MakePA to avoid potential race conditions

config/core/300-resources/podautoscaler.yaml

@@ -125,9 +125,28 @@ spec:
             status:
               description: Status communicates the observed state of the PodAutoscaler (from the controller).
               type: object
-              required:
-                - metricsServiceName
-                - serviceName
+              default:
+                conditions:
+                  - lastTransitionTime: "1970-01-01T00:00:00Z"
+                    message: Waiting for controller
+                    reason: Pending
+                    status: Unknown
+                    type: Active
+                  - lastTransitionTime: "1970-01-01T00:00:00Z"
+                    message: Waiting for controller
+                    reason: Pending
+                    status: Unknown
+                    type: Ready
+                  - lastTransitionTime: "1970-01-01T00:00:00Z"
+                    message: Waiting for controller
+                    reason: Pending
+                    status: Unknown
+                    type: SKSReady
+                  - lastTransitionTime: "1970-01-01T00:00:00Z"
+                    message: Waiting for controller
+                    reason: Pending
+                    status: Unknown
+                    type: ScaleTargetInitialized
               properties:
                 actualScale:
                   description: ActualScale shows the actual number of replicas for the revision.

docs/serving-api.md

@@ -459,6 +459,7 @@ string
 </em>
 </td>
 <td>
+<em>(Optional)</em>
 <p>ServiceName is the K8s Service name that serves the revision, scaled by this PA.
 The service is created and owned by the ServerlessService object owned by this PA.</p>
 </td>
@@ -471,6 +472,7 @@ string
 </em>
 </td>
 <td>
+<em>(Optional)</em>
 <p>MetricsServiceName is the K8s Service name that provides revision metrics.
 The service is managed by the PA object.</p>
 </td>

pkg/apis/autoscaling/v1alpha1/pa_types.go

@@ -112,14 +112,17 @@ const (
 
 // PodAutoscalerStatus communicates the observed state of the PodAutoscaler (from the controller).
 type PodAutoscalerStatus struct {
+	// +kubebuilder:default={"conditions": {{"type":"Active", "status": "Unknown", "reason": "Pending",  "message": "Waiting for controller", "lastTransitionTime": "1970-01-01T00:00:00Z"}, {"type":"Ready", "status": "Unknown", "reason": "Pending", "message": "Waiting for controller", "lastTransitionTime": "1970-01-01T00:00:00Z"}, {"type":"SKSReady", "status": "Unknown", "reason": "Pending", "message": "Waiting for controller", "lastTransitionTime": "1970-01-01T00:00:00Z"}, {"type":"ScaleTargetInitialized", "status": "Unknown", "reason": "Pending", "message": "Waiting for controller", "lastTransitionTime": "1970-01-01T00:00:00Z"}}}
 	duckv1.Status `json:",inline"`
 
 	// ServiceName is the K8s Service name that serves the revision, scaled by this PA.
 	// The service is created and owned by the ServerlessService object owned by this PA.
+	// +optional
 	ServiceName string `json:"serviceName"`
 
 	// MetricsServiceName is the K8s Service name that provides revision metrics.
 	// The service is managed by the PA object.
+	// +optional
 	MetricsServiceName string `json:"metricsServiceName"`
 
 	// DesiredScale shows the current desired number of replicas for the revision.

pkg/apis/serving/v1/revision_lifecycle_test.go

@@ -726,6 +726,7 @@ func TestPropagateAutoscalerStatusReplicas(t *testing.T) {
 
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			tc.ps.InitializeConditions()
 			r.PropagateAutoscalerStatus(&tc.ps)
 
 			if !cmp.Equal(tc.wantActualReplicas, r.ActualReplicas) {

pkg/reconciler/revision/table_test.go

@@ -340,7 +340,7 @@ func TestReconcile(t *testing.T) {
 		},
 		WantStatusUpdates: []clientgotesting.UpdateActionImpl{{
 			Object: Revision("foo", "pa-inactive",
-				WithLogURL, withDefaultContainerStatuses(), MarkDeploying(""),
+				WithLogURL, withDefaultContainerStatuses(), MarkContainerHealthyUnknown(""),
 				// When we reconcile an "all ready" revision when the PA
 				// is inactive, we should see the following change.
 				MarkInactive("NoTraffic", "This thing is inactive."), WithRevisionObservedGeneration(1),

pkg/reconciler/testing/v1/factory.go

@@ -25,6 +25,7 @@ import (
 	fakenetworkingclient "knative.dev/networking/pkg/client/injection/client/fake"
 	fakekubeclient "knative.dev/pkg/client/injection/kube/client/fake"
 	fakedynamicclient "knative.dev/pkg/injection/clients/dynamicclient/fake"
+	autoscalingv1alpha1 "knative.dev/serving/pkg/apis/autoscaling/v1alpha1"
 	v1 "knative.dev/serving/pkg/apis/serving/v1"
 	fakeservingclient "knative.dev/serving/pkg/client/injection/client/fake"
 
@@ -95,6 +96,28 @@ func MakeFactory(ctor Ctor) rtesting.Factory {
 				return false, nil, nil
 			},
 		)
+
+		client.PrependReactor("create", "podautoscalers", func(action ktesting.Action) (handled bool, ret runtime.Object, err error) {
+			if createAction, ok := action.(ktesting.CreateAction); ok {
+				if pa, ok := createAction.GetObject().(*autoscalingv1alpha1.PodAutoscaler); ok {
+					// Initialize conditions first
+					pa.Status.InitializeConditions()
+
+					// Set all conditions to match kubebuilder defaults: status="Unknown", reason="Deploying"
+					// This matches the behavior defined in the +kubebuilder:default annotation
+					condSet := pa.GetConditionSet()
+					manager := condSet.Manage(&pa.Status)
+
+					// Set each condition to Unknown with "Deploying" reason to match kubebuilder defaults
+					manager.MarkUnknown(autoscalingv1alpha1.PodAutoscalerConditionActive, "Deploying", "")
+					manager.MarkUnknown(autoscalingv1alpha1.PodAutoscalerConditionReady, "Deploying", "")
+					manager.MarkUnknown(autoscalingv1alpha1.PodAutoscalerConditionSKSReady, "Deploying", "")
+					manager.MarkUnknown(autoscalingv1alpha1.PodAutoscalerConditionScaleTargetInitialized, "Deploying", "")
+				}
+			}
+			return false, nil, nil
+		})
+
 		// This is needed by the Configuration controller tests, which
 		// use GenerateName to produce Revisions.
 		rtesting.PrependGenerateNameReactor(&client.Fake)