mitigate upgrade-e2e flakiness

Signed-off-by: Per Goncalves da Silva <[email protected]>

Author: Per Goncalves da Silva

test/upgrade-e2e/post_upgrade_test.go

@@ -32,29 +32,13 @@ func TestClusterExtensionAfterOLMUpgrade(t *testing.T) {
 	ctx := context.Background()
 	defer utils.CollectTestArtifacts(t, artifactName, c, cfg)
 
-	managerLabelSelector := labels.Set{"control-plane": "operator-controller-controller-manager"}
+	// wait for catalogd deployment to finish
+	t.Log("Wait for catalogd deployment to be ready")
+	catalogdManagerPod := waitForDeployment(t, ctx, "catalogd-controller-manager")
 
-	t.Log("Checking that the controller-manager deployment is updated")
-	require.EventuallyWithT(t, func(ct *assert.CollectT) {
-		var managerDeployments appsv1.DeploymentList
-		assert.NoError(ct, c.List(ctx, &managerDeployments, client.MatchingLabelsSelector{Selector: managerLabelSelector.AsSelector()}))
-		assert.Len(ct, managerDeployments.Items, 1)
-		managerDeployment := managerDeployments.Items[0]
-
-		assert.True(ct,
-			managerDeployment.Status.UpdatedReplicas == *managerDeployment.Spec.Replicas &&
-				managerDeployment.Status.Replicas == *managerDeployment.Spec.Replicas &&
-				managerDeployment.Status.AvailableReplicas == *managerDeployment.Spec.Replicas &&
-				managerDeployment.Status.ReadyReplicas == *managerDeployment.Spec.Replicas,
-		)
-	}, time.Minute, time.Second)
-
-	var managerPods corev1.PodList
-	t.Log("Waiting for only one controller-manager Pod to remain")
-	require.EventuallyWithT(t, func(ct *assert.CollectT) {
-		assert.NoError(ct, c.List(ctx, &managerPods, client.MatchingLabelsSelector{Selector: managerLabelSelector.AsSelector()}))
-		assert.Len(ct, managerPods.Items, 1)
-	}, time.Minute, time.Second)
+	// wait for operator-controller deployment to finish
+	t.Log("Wait for operator-controller deployment to be ready")
+	managerPod := waitForDeployment(t, ctx, "operator-controller-controller-manager")
 
 	t.Log("Reading logs to make sure that ClusterExtension was reconciled by operator-controller before we update it")
 	// Make sure that after we upgrade OLM itself we can still reconcile old objects without any changes
@@ -64,28 +48,42 @@ func TestClusterExtensionAfterOLMUpgrade(t *testing.T) {
 		"reconcile ending",
 		fmt.Sprintf(`ClusterExtension=%q`, testClusterExtensionName),
 	}
-	found, err := watchPodLogsForSubstring(logCtx, &managerPods.Items[0], "manager", substrings...)
+	found, err := watchPodLogsForSubstring(logCtx, managerPod, "manager", substrings...)
 	require.NoError(t, err)
 	require.True(t, found)
 
-	t.Log("Checking that the ClusterCatalog is serving")
+	t.Log("Checking that the ClusterCatalog is unpacked")
 	require.EventuallyWithT(t, func(ct *assert.CollectT) {
 		var clusterCatalog catalogd.ClusterCatalog
 		assert.NoError(ct, c.Get(ctx, types.NamespacedName{Name: testClusterCatalogName}, &clusterCatalog))
+
+		// check serving condition
 		cond := apimeta.FindStatusCondition(clusterCatalog.Status.Conditions, catalogd.TypeServing)
-		if !assert.NotNil(ct, cond) {
+		if assert.Nil(ct, cond) {
 			return
 		}
 		assert.Equal(ct, metav1.ConditionTrue, cond.Status)
 		assert.Equal(ct, catalogd.ReasonAvailable, cond.Reason)
+
+		// mitigation for upgrade-e2e flakiness caused by the following bug
+		// https://github.com/operator-framework/operator-controller/issues/1626
+		// wait until the unpack time > than the catalogd controller pod creation time
+		cond = apimeta.FindStatusCondition(clusterCatalog.Status.Conditions, catalogd.TypeProgressing)
+		if assert.Nil(ct, cond) {
+			return
+		}
+		assert.Equal(ct, metav1.ConditionTrue, cond.Status)
+		assert.Equal(ct, catalogd.ReasonSucceeded, cond.Reason)
+
+		assert.True(ct, clusterCatalog.Status.LastUnpacked.After(catalogdManagerPod.CreationTimestamp.Time))
 	}, time.Minute, time.Second)
 
 	t.Log("Checking that the ClusterExtension is installed")
 	var clusterExtension ocv1.ClusterExtension
 	require.EventuallyWithT(t, func(ct *assert.CollectT) {
 		assert.NoError(ct, c.Get(ctx, types.NamespacedName{Name: testClusterExtensionName}, &clusterExtension))
 		cond := apimeta.FindStatusCondition(clusterExtension.Status.Conditions, ocv1.TypeInstalled)
-		if !assert.NotNil(ct, cond) {
+		if assert.Nil(ct, cond) {
 			return
 		}
 		assert.Equal(ct, metav1.ConditionTrue, cond.Status)
@@ -107,7 +105,7 @@ func TestClusterExtensionAfterOLMUpgrade(t *testing.T) {
 	require.EventuallyWithT(t, func(ct *assert.CollectT) {
 		assert.NoError(ct, c.Get(ctx, types.NamespacedName{Name: testClusterExtensionName}, &clusterExtension))
 		cond := apimeta.FindStatusCondition(clusterExtension.Status.Conditions, ocv1.TypeInstalled)
-		if !assert.NotNil(ct, cond) {
+		if assert.Nil(ct, cond) {
 			return
 		}
 		assert.Equal(ct, ocv1.ReasonSucceeded, cond.Reason)
@@ -117,6 +115,39 @@ func TestClusterExtensionAfterOLMUpgrade(t *testing.T) {
 	}, time.Minute, time.Second)
 }
 
+// waitForDeployment checks that the updated deployment with the given control-plane label
+// has reached the desired number of replicas and that the number pods matches that number
+// i.e. no old pods remain. It will return a pointer to the first pod. This is only necessary
+// to facilitate the mitigation put in place for https://github.com/operator-framework/operator-controller/issues/1626
+func waitForDeployment(t *testing.T, ctx context.Context, controlPlaneLabel string) *corev1.Pod {
+	deploymentLabelSelector := labels.Set{"control-plane": controlPlaneLabel}.AsSelector()
+
+	t.Log("Checking that the deployment is updated")
+	var desiredNumReplicas int32
+	require.EventuallyWithT(t, func(ct *assert.CollectT) {
+		var managerDeployments appsv1.DeploymentList
+		assert.NoError(ct, c.List(ctx, &managerDeployments, client.MatchingLabelsSelector{Selector: deploymentLabelSelector}))
+		assert.Len(ct, managerDeployments.Items, 1)
+		managerDeployment := managerDeployments.Items[0]
+
+		assert.True(ct,
+			managerDeployment.Status.UpdatedReplicas == *managerDeployment.Spec.Replicas &&
+				managerDeployment.Status.Replicas == *managerDeployment.Spec.Replicas &&
+				managerDeployment.Status.AvailableReplicas == *managerDeployment.Spec.Replicas &&
+				managerDeployment.Status.ReadyReplicas == *managerDeployment.Spec.Replicas,
+		)
+		desiredNumReplicas = *managerDeployment.Spec.Replicas
+	}, time.Minute, time.Second)
+
+	var managerPods corev1.PodList
+	t.Logf("Ensure the number of remaining pods equal the desired number of replicas (%d)", desiredNumReplicas)
+	require.EventuallyWithT(t, func(ct *assert.CollectT) {
+		assert.NoError(ct, c.List(ctx, &managerPods, client.MatchingLabelsSelector{Selector: deploymentLabelSelector}))
+		assert.Len(ct, managerPods.Items, 1)
+	}, time.Minute, time.Second)
+	return &managerPods.Items[0]
+}
+
 func watchPodLogsForSubstring(ctx context.Context, pod *corev1.Pod, container string, substrings ...string) (bool, error) {
 	podLogOpts := corev1.PodLogOptions{
 		Follow:    true,