fix(ip): Recreate pending installplan if deleted before approval

At the moment, OLM will not recreate the pending-approval
installplan. The subscription will be stuck in UpgradePending forever
as a result. This PR will enable installplan recreation if it is deleted
before approval.

Signed-off-by: Vu Dinh <[email protected]>

Author: dinhxuanvu

pkg/controller/operators/catalog/operator.go

@@ -831,6 +831,12 @@ func (o *Operator) syncResolvingNamespace(obj interface{}) error {
 		return err
 	}
 
+	ips, err := o.listInstallPlansMap(namespace)
+	if err != nil {
+		logger.WithError(err).Debug("couldn't list installplan")
+		return err
+	}
+
 	// TODO: parallel
 	maxGeneration := 0
 	subscriptionUpdated := false
@@ -847,7 +853,7 @@ func (o *Operator) syncResolvingNamespace(obj interface{}) error {
 		}
 
 		// ensure the installplan reference is correct
-		sub, changedIP, err := o.ensureSubscriptionInstallPlanState(logger, sub)
+		sub, changedIP, err := o.ensureSubscriptionInstallPlanState(logger, sub, ips)
 		if err != nil {
 			logger.Debugf("error ensuring installplan state: %v", err)
 			return err
@@ -952,9 +958,29 @@ func (o *Operator) nothingToUpdate(logger *logrus.Entry, sub *v1alpha1.Subscript
 	return false
 }
 
-func (o *Operator) ensureSubscriptionInstallPlanState(logger *logrus.Entry, sub *v1alpha1.Subscription) (*v1alpha1.Subscription, bool, error) {
-	if sub.Status.InstallPlanRef != nil {
-		return sub, false, nil
+// checkMissingInstallPlan checks if the installplan is missing or not when
+// the subscription is in pending upgrade state
+func (o *Operator) checkMissingInstallPlan(sub *v1alpha1.Subscription, ips map[string]struct{}) (*v1alpha1.Subscription, bool, error) {
+	_, ok := ips[sub.Status.InstallPlanRef.Name]
+	if !ok && sub.Status.State == v1alpha1.SubscriptionStateUpgradePending {
+		out := sub.DeepCopy()
+		out.Status.InstallPlanRef = nil
+		out.Status.Install = nil
+		out.Status.CurrentCSV = ""
+		out.Status.State = v1alpha1.SubscriptionStateNone
+		out.Status.LastUpdated = o.now()
+		updated, err := o.client.OperatorsV1alpha1().Subscriptions(sub.GetNamespace()).UpdateStatus(context.TODO(), out, metav1.UpdateOptions{})
+		if err != nil {
+			return out, false, nil
+		}
+		return updated, true, nil
+	}
+	return sub, false, nil
+}
+
+func (o *Operator) ensureSubscriptionInstallPlanState(logger *logrus.Entry, sub *v1alpha1.Subscription, ips map[string]struct{}) (*v1alpha1.Subscription, bool, error) {
+	if sub.Status.InstallPlanRef != nil || sub.Status.Install != nil {
+		return o.checkMissingInstallPlan(sub, ips)
 	}
 
 	logger.Debug("checking for existing installplan")
@@ -2020,6 +2046,20 @@ func (o *Operator) listInstallPlans(namespace string) (ips []*v1alpha1.InstallPl
 	return
 }
 
+func (o *Operator) listInstallPlansMap(namespace string) (ips map[string]struct{}, err error) {
+	list, err := o.client.OperatorsV1alpha1().InstallPlans(namespace).List(context.TODO(), metav1.ListOptions{})
+	if err != nil {
+		return
+	}
+
+	ips = make(map[string]struct{})
+	for i := range list.Items {
+		ips[list.Items[i].GetName()] = struct{}{}
+	}
+
+	return
+}
+
 // competingCRDOwnersExist returns true if there exists a CSV that owns at least one of the given CSVs owned CRDs (that's not the given CSV)
 func competingCRDOwnersExist(namespace string, csv *v1alpha1.ClusterServiceVersion, existingOwners map[string][]string) (bool, error) {
 	// Attempt to find a pre-existing owner in the namespace for any owned crd

test/e2e/subscription_e2e_test.go

@@ -217,8 +217,34 @@ var _ = Describe("Subscription", func() {
 		require.Equal(GinkgoT(), v1alpha1.ApprovalManual, installPlan.Spec.Approval)
 		require.Equal(GinkgoT(), v1alpha1.InstallPlanPhaseRequiresApproval, installPlan.Status.Phase)
 
-		installPlan.Spec.Approved = true
-		_, err = crc.OperatorsV1alpha1().InstallPlans(testNamespace).Update(context.Background(), installPlan, metav1.UpdateOptions{})
+		// Delete the current installplan
+		err = crc.OperatorsV1alpha1().InstallPlans(testNamespace).Delete(context.Background(), installPlan.Name, metav1.DeleteOptions{})
+		require.NoError(GinkgoT(), err)
+
+		var ipName string
+		Eventually(func() bool {
+			fetched, err := crc.OperatorsV1alpha1().Subscriptions(testNamespace).Get(context.TODO(), "manual-subscription", metav1.GetOptions{})
+			if err != nil {
+				return false
+			}
+			if fetched.Status.Install != nil {
+				ipName = fetched.Status.Install.Name
+				return fetched.Status.Install.Name != installPlan.Name
+			}
+			return false
+		}, 5*time.Minute, 10*time.Second).Should(BeTrue())
+
+		// Fetch new installplan
+		newInstallPlan, err := fetchInstallPlan(GinkgoT(), crc, ipName, buildInstallPlanPhaseCheckFunc(v1alpha1.InstallPlanPhaseRequiresApproval))
+		require.NoError(GinkgoT(), err)
+		require.NotNil(GinkgoT(), newInstallPlan)
+
+		require.NotEqual(GinkgoT(), installPlan.Name, newInstallPlan.Name, "expected new installplan recreated")
+		require.Equal(GinkgoT(), v1alpha1.ApprovalManual, newInstallPlan.Spec.Approval)
+		require.Equal(GinkgoT(), v1alpha1.InstallPlanPhaseRequiresApproval, newInstallPlan.Status.Phase)
+
+		newInstallPlan.Spec.Approved = true
+		_, err = crc.OperatorsV1alpha1().InstallPlans(testNamespace).Update(context.Background(), newInstallPlan, metav1.UpdateOptions{})
 		require.NoError(GinkgoT(), err)
 
 		subscription, err = fetchSubscription(crc, testNamespace, "manual-subscription", subscriptionStateAtLatestChecker)