UPSTREAM: <carry>: fix(test): drop blocking namespace-deletion wait between both-watch-modes scenarios

The both-watch-modes test loops over two scenarios (singlens, ownns) inside
a single It block and was blocking on full namespace deletion between them.
This caused flaky 300s timeouts on GCP techpreview clusters where master
nodes run at 94-99% CPU, which starves the namespace controller and makes
namespace termination arbitrarily slow.

The wait was not guarding anything real:
- EnsureCleanupClusterExtension already ensures the CE and CRD are gone;
  since CE deletion uses ForegroundPropagation, the ClusterObjectSet teardown
  must complete before the CE disappears, meaning all managed resources
  (Deployments, Services, etc.) are already deleted at that point.
- The singleown bundle installs no ValidatingWebhookConfiguration or
  MutatingWebhookConfiguration, so there is no webhook admission risk.
- Each scenario generates unique namespace names and CRD group suffixes via
  rand.String(4), so a terminating namespace from scenario 1 cannot collide
  with or interfere with scenario 2's resources.

Trigger both namespace deletions and proceed without waiting. The DeferCleanup
registrations that already exist will handle any residual cleanup after the
spec exits.

Fixes: OCPBUGS-84943

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Todd Short <tshort@redhat.com>

Author: 2 people

openshift/tests-extension/test/olmv1-singleownnamespace.go

@@ -13,7 +13,6 @@ import (
 	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
 	apiextensionsv1 "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
-	"k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/api/meta"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/rand"
@@ -417,7 +416,7 @@ var _ = Describe("[sig-olmv1][OCPFeatureGate:NewOLMOwnSingleNamespace] OLMv1 ope
 				DeferCleanup(func() {
 					By(fmt.Sprintf("cleanup: deleting ClusterExtension %s", ce.Name))
 					_ = k8sClient.Delete(context.Background(), ce, client.PropagationPolicy(metav1.DeletePropagationForeground))
-					helpers.EnsureCleanupClusterExtension(context.Background(), ceName, nsName)
+					helpers.EnsureCleanupClusterExtension(context.Background(), packageName, crdName)
 				})
 
 				By(fmt.Sprintf("waiting for the ClusterExtension %s to be installed for %s scenario", ceName, sc.label))
@@ -445,33 +444,27 @@ var _ = Describe("[sig-olmv1][OCPFeatureGate:NewOLMOwnSingleNamespace] OLMv1 ope
 					g.Expect(found).To(BeTrue(), "failed to find deployment with olm.targetNamespaces annotation")
 				}).WithTimeout(helpers.DefaultTimeout).WithPolling(helpers.DefaultPolling).Should(Succeed())
 
-				// Ginkgo never invokes those deferred cleanups until we exit the whole spec, so the first scenario’s
-				// cluster resources survive long enough to collide with the second scenario.
-				By(fmt.Sprintf("cleaning up resources created for %s scenario to allow next scenario", sc.label))
+				// DeferCleanup handlers don’t run between loop iterations; they fire after the whole spec exits.
+				// Explicitly clean up cluster-scoped resources (CE, CRD) so EnsureCleanupClusterExtension can
+				// block until COS teardown completes before the next scenario starts. Unique names per scenario
+				// prevent any resource collision regardless.
+				By(fmt.Sprintf("cleaning up resources created for %s scenario before next scenario", sc.label))
 				deletePolicy := metav1.DeletePropagationForeground
 				Expect(k8sClient.Delete(ctx, ce, client.PropagationPolicy(deletePolicy))).To(Succeed(), "failed to delete ClusterExtension %q", ceName)
 				helpers.EnsureCleanupClusterExtension(context.Background(), packageName, crdName)
 
 				Expect(k8sClient.Delete(ctx, crb, client.PropagationPolicy(deletePolicy))).To(Succeed(), "failed to delete ClusterRoleBinding %q", crbName)
 				Expect(k8sClient.Delete(ctx, sa, client.PropagationPolicy(deletePolicy))).To(Succeed(), "failed to delete ServiceAccount %q", saName)
 
+				// Trigger namespace deletion and proceed without blocking. By this point
+				// EnsureCleanupClusterExtension has completed, meaning the ClusterObjectSet
+				// teardown has deleted all managed resources (Deployment, Service, etc.) and the
+				// CRD is gone. Each scenario uses unique namespace names and CRD groups, so the
+				// terminating namespace from scenario 1 cannot interfere with scenario 2.
 				if watchNSObj != nil {
-					Expect(k8sClient.Delete(ctx, watchNSObj, client.PropagationPolicy(deletePolicy))).To(Succeed(), "failed to delete watch namespace %q", watchNamespace)
-				}
-				Expect(k8sClient.Delete(ctx, installNS, client.PropagationPolicy(deletePolicy))).To(Succeed(), "failed to delete install namespace %q", installNamespace)
-
-				By(fmt.Sprintf("waiting for namespace %s to be fully deleted before next scenario", installNamespace))
-				Eventually(func() bool {
-					err := k8sClient.Get(ctx, client.ObjectKey{Name: installNamespace}, &corev1.Namespace{})
-					return errors.IsNotFound(err)
-				}).WithTimeout(helpers.DefaultTimeout).WithPolling(helpers.DefaultPolling).Should(BeTrue(), "expected namespace %s to be deleted", installNamespace)
-				if watchNSObj != nil {
-					By(fmt.Sprintf("waiting for namespace %s to be fully deleted before next scenario", watchNamespace))
-					Eventually(func() bool {
-						err := k8sClient.Get(ctx, client.ObjectKey{Name: watchNamespace}, &corev1.Namespace{})
-						return errors.IsNotFound(err)
-					}).WithTimeout(helpers.DefaultTimeout).WithPolling(helpers.DefaultPolling).Should(BeTrue(), "expected namespace %s to be deleted", watchNamespace)
+					_ = k8sClient.Delete(ctx, watchNSObj, client.PropagationPolicy(deletePolicy))
 				}
+				_ = k8sClient.Delete(ctx, installNS, client.PropagationPolicy(deletePolicy))
 			}
 		})
 })