NE-2097: Add logic to pick up next InstallPlan if desired version jumps upgrade graph

Author: alebedev87

manifests/00-cluster-role.yaml

@@ -180,6 +180,7 @@ rules:
   resources:
   - subscriptions
   - installplans
+  - clusterserviceversions
   verbs:
   - '*'
 

manifests/02-deployment.yaml

@@ -81,6 +81,12 @@ spec:
               value: openshift/origin-cluster-ingress-operator:latest
             - name: GATEWAY_API_OPERATOR_CHANNEL
               value: stable
+            # NOTE:
+            # Use an operator version that exists in the catalog.
+            # If the specified version is not found, the latest available
+            # OSSM operator version will be installed as of the upgrade date.
+            # Downgrades are not supported; the new version must be
+            # semantically greater than the previous one.
             - name: GATEWAY_API_OPERATOR_VERSION
               value: servicemeshoperator3.v3.0.1
           resources:

pkg/operator/controller/gatewayclass/subscription.go

@@ -176,7 +176,9 @@ func (r *reconciler) ensureServiceMeshOperatorInstallPlan(ctx context.Context) (
 }
 
 // currentInstallPlan returns the InstallPlan that describes installing the expected version of the GatewayAPI
-// implementation, if one exists.
+// implementation. If no InstallPlan exists for the expected version, return the InstallPlan which replaces the currently installed one.
+// This InstallPlan is expected to advance the OSSM operator toward the next CSV in the upgrade graph,
+// assuming that the configured version is available further up the graph.
 func (r *reconciler) currentInstallPlan(ctx context.Context) (bool, *operatorsv1alpha1.InstallPlan, error) {
 	_, subscription, err := r.currentSubscription(ctx, operatorcontroller.ServiceMeshOperatorSubscriptionName())
 	if err != nil {
@@ -189,7 +191,7 @@ func (r *reconciler) currentInstallPlan(ctx context.Context) (bool, *operatorsv1
 	if installPlans == nil || len(installPlans.Items) == 0 {
 		return false, nil, nil
 	}
-	var currentInstallPlan *operatorsv1alpha1.InstallPlan
+	var currentInstallPlan, nextInstallPlan *operatorsv1alpha1.InstallPlan
 	multipleInstallPlans := false
 	for _, installPlan := range installPlans.Items {
 		if len(installPlan.OwnerReferences) == 0 || len(installPlan.Spec.ClusterServiceVersionNames) == 0 {
@@ -209,6 +211,7 @@ func (r *reconciler) currentInstallPlan(ctx context.Context) (bool, *operatorsv1
 		if installPlan.Status.Phase != operatorsv1alpha1.InstallPlanPhaseRequiresApproval {
 			continue
 		}
+		// Check whether InstallPlan implements the expected operator version.
 		for _, csvName := range installPlan.Spec.ClusterServiceVersionNames {
 			if csvName == r.config.GatewayAPIOperatorVersion {
 				// Keep the newest InstallPlan to return at the end of the loop.
@@ -223,10 +226,56 @@ func (r *reconciler) currentInstallPlan(ctx context.Context) (bool, *operatorsv1
 				}
 			}
 		}
+		// Check whether InstallPlan implements the next operator version in the upgrade graph.
+		for _, csvName := range installPlan.Spec.ClusterServiceVersionNames {
+			// The definitions of InstalledCSV and CurrentCSV are non-trivial:
+			//
+			// - InstalledCSV represents the currently running CSV.
+			// - CurrentCSV represents the version that "subscription is progressing to"
+			//   which practically means "the next CSV in the upgrade graph."
+			//
+			// - If InstalledCSV < CurrentCSV:
+			//     No CSV replacement is ongoing. InstalledCSV is the current version,
+			//     and CurrentCSV is the next one in the upgrade graph.
+			// - If InstalledCSV == CurrentCSV:
+			//     One of the following scenarios is possible:
+			//       1. CSV replacement is in progress "InstalledCSV-1" is being replaced with CurrentCSV.
+			//       2. Installation of the first CSV is in progress.
+			//       3. There is no "next CSV" in the upgrade graph, so CurrentCSV
+			//          cannot point to a future version.
+			//   CurrentCSV only becomes the next version once the replacement finishes,
+			//   and the next InstallPlan appears around the same time.
+			//
+			// The first condition (below) prevents setting the next InstallPlan while a replacement
+			// or installation is ongoing, or when the end of the upgrade graph is reached.
+			if subscription.Status.InstalledCSV != subscription.Status.CurrentCSV && csvName == subscription.Status.CurrentCSV {
+				if nextInstallPlan == nil {
+					nextInstallPlan = &installPlan
+					break
+				}
+			}
+		}
 	}
 	if multipleInstallPlans {
 		log.Info(fmt.Sprintf("found multiple valid InstallPlans. using %s because it's the newest", currentInstallPlan.Name))
 	}
+	// No InstallPlan with the expected operator version was found,
+	// but the next one in the upgrade graph exists.
+	// Return the next InstallPlan to continue the upgrade.
+	if currentInstallPlan == nil && nextInstallPlan != nil {
+		log.Info("next install plan time")
+		// The condition below prevents approving an InstallPlan
+		// that targets a version beyond the expected operator version.
+		// This can happen when:
+		// - InstallPlan with the expected version is complete (no approval needed).
+		// - Newer versions exist in the upgrade graph.
+		// The check ensures that the currently running CSV is different
+		// from the expected version. Once they match, no further action is needed.
+		if subscription.Status.InstalledCSV != r.config.GatewayAPIOperatorVersion {
+			log.Info("installplan with expected operator version was not found; proceedng with an intermedite installplan", "name", nextInstallPlan.Name, "csv", subscription.Status.CurrentCSV)
+			currentInstallPlan = nextInstallPlan
+		}
+	}
 	return (currentInstallPlan != nil), currentInstallPlan, nil
 }