OCPBUGS-83281: Wait for Gateway deletion before removing GatewayClass in test cleanup

The Gateway API test cleanup was deleting Gateways and immediately
proceeding to delete the GatewayClass and istiod without waiting for
the Gateway resources to be fully removed. Since the gateway
deployment and pods have an owner reference to the Gateway, they are
cascade-deleted by Kubernetes GC only after the Gateway is gone. If
the GatewayClass and istiod are removed first, the gateway pods lose
their control plane and crash-loop, generating pathological "Back-off
restarting failed container" events that fail CI invariant monitors.

Add a shared waitForGatewayDeletion helper that both the upgrade test
Teardown and the controller test AfterEach use to ensure Gateways are
fully deleted before proceeding with GatewayClass cleanup.

https://redhat.atlassian.net/browse/OCPBUGS-83281

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Author: gcs278

test/extended/router/gatewayapi_upgrade.go

@@ -13,7 +13,6 @@ import (
 
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
-	"k8s.io/apimachinery/pkg/util/wait"
 	e2e "k8s.io/kubernetes/test/e2e/framework"
 	"k8s.io/kubernetes/test/e2e/upgrades"
 	gatewayv1 "sigs.k8s.io/gateway-api/apis/v1"
@@ -240,21 +239,8 @@ func (t *GatewayAPIUpgradeTest) Teardown(ctx context.Context, f *e2e.Framework)
 		e2e.Logf("Failed to delete Gateway %q: %v", t.gatewayName, err)
 	}
 
-	// Wait for Gateway to be fully deleted before removing GatewayClass
-	// This prevents orphaned resources if the controller (defined by GatewayClass) is removed
-	// before Istiod completes cleanup
 	g.By("Waiting for Gateway to be fully deleted")
-	err = wait.PollUntilContextTimeout(ctx, 2*time.Second, 2*time.Minute, false, func(ctx context.Context) (bool, error) {
-		_, err := t.oc.AdminGatewayApiClient().GatewayV1().Gateways(ingressNamespace).Get(ctx, t.gatewayName, metav1.GetOptions{})
-		if apierrors.IsNotFound(err) {
-			e2e.Logf("Gateway %q successfully deleted", t.gatewayName)
-			return true, nil
-		}
-		return false, nil
-	})
-	if err != nil {
-		e2e.Logf("Gateway %q still exists after 2 minutes, continuing cleanup anyway", t.gatewayName)
-	}
+	waitForGatewayDeletion(t.oc, t.gatewayName)
 
 	g.By("Deleting the GatewayClass")
 	err = t.oc.AdminGatewayApiClient().GatewayV1().GatewayClasses().Delete(ctx, gatewayClassName, metav1.DeleteOptions{})

test/extended/router/gatewayapicontroller.go

@@ -152,6 +152,11 @@ var _ = g.Describe("[sig-network-edge][OCPFeatureGate:GatewayAPIController][Feat
 				o.Expect(err).NotTo(o.HaveOccurred(), "Gateway %s could not be deleted", name)
 			}
 
+			g.By("Waiting for gateways to be fully deleted")
+			for _, name := range gateways {
+				waitForGatewayDeletion(oc, name)
+			}
+
 			g.By("Deleting the GatewayClass")
 
 			if err := oc.AdminGatewayApiClient().GatewayV1().GatewayClasses().Delete(context.Background(), gatewayClassName, metav1.DeleteOptions{}); err != nil && !apierrors.IsNotFound(err) {
@@ -1324,6 +1329,19 @@ func waitForIstiodPodDeletion(oc *exutil.CLI) {
 	}).WithTimeout(10 * time.Minute).WithPolling(10 * time.Second).Should(o.Succeed())
 }
 
+// waitForGatewayDeletion waits for a Gateway resource to be fully deleted.
+// The gateway deployment and pods have an owner reference to the Gateway, so
+// Kubernetes garbage collection will cascade-delete them once the Gateway is gone.
+// This must complete before removing the GatewayClass or istiod, otherwise
+// orphaned gateway pods will crash-loop without their control plane.
+func waitForGatewayDeletion(oc *exutil.CLI, gatewayName string) {
+	e2e.Logf("Waiting for Gateway %q to be fully deleted", gatewayName)
+	o.Eventually(func(g o.Gomega) {
+		_, err := oc.AdminGatewayApiClient().GatewayV1().Gateways(ingressNamespace).Get(context.Background(), gatewayName, metav1.GetOptions{})
+		g.Expect(apierrors.IsNotFound(err)).To(o.BeTrue(), "Gateway %q still exists", gatewayName)
+	}).WithTimeout(5 * time.Minute).WithPolling(5 * time.Second).Should(o.Succeed())
+}
+
 // validateOLMBasedOSSM validates that Gateway API is using OLM-based provisioning.
 func validateOLMBasedOSSM(oc *exutil.CLI, timeout time.Duration) {
 	pollInterval := 5 * time.Second