Merge pull request #5166 from serngawy/versionGate

🐛 ROSA: Fix version gate Ack for rosa hcp y-stream version upgrade

Author: k8s-ci-robot

config/crd/bases/controlplane.cluster.x-k8s.io_rosacontrolplanes.yaml

@@ -787,6 +787,19 @@ spec:
               version:
                 description: OpenShift semantic version, for example "4.14.5".
                 type: string
+              versionGate:
+                default: WaitForAcknowledge
+                description: |-
+                  VersionGate requires acknowledgment when upgrading ROSA-HCP y-stream versions (e.g., from 4.15 to 4.16).
+                  Default is WaitForAcknowledge.
+                  WaitForAcknowledge: If acknowledgment is required, the upgrade will not proceed until VersionGate is set to Acknowledge or AlwaysAcknowledge.
+                  Acknowledge: If acknowledgment is required, apply it for the upgrade. After upgrade is done set the version gate to WaitForAcknowledge.
+                  AlwaysAcknowledge: If acknowledgment is required, apply it and proceed with the upgrade.
+                enum:
+                - Acknowledge
+                - WaitForAcknowledge
+                - AlwaysAcknowledge
+                type: string
               workerRoleARN:
                 description: WorkerRoleARN is an AWS IAM role that will be attached
                   to worker instances.
@@ -801,11 +814,17 @@ spec:
             - subnets
             - supportRoleARN
             - version
+            - versionGate
             - workerRoleARN
             type: object
           status:
             description: RosaControlPlaneStatus defines the observed state of ROSAControlPlane.
             properties:
+              availableUpgrades:
+                description: Available upgrades for the ROSA hosted control plane.
+                items:
+                  type: string
+                type: array
               conditions:
                 description: Conditions specifies the conditions for the managed control
                   plane

controlplane/rosa/api/v1beta2/rosacontrolplane_types.go

@@ -38,6 +38,20 @@ const (
 	Private RosaEndpointAccessType = "Private"
 )
 
+// VersionGateAckType specifies the version gate acknowledgement.
+type VersionGateAckType string
+
+const (
+	// Acknowledge if acknowledgment is required and proceed with the upgrade.
+	Acknowledge VersionGateAckType = "Acknowledge"
+
+	// WaitForAcknowledge if acknowledgment is required, wait not to proceed with the upgrade.
+	WaitForAcknowledge VersionGateAckType = "WaitForAcknowledge"
+
+	// AlwaysAcknowledge always acknowledg if required and proceed with the upgrade.
+	AlwaysAcknowledge VersionGateAckType = "AlwaysAcknowledge"
+)
+
 // RosaControlPlaneSpec defines the desired state of ROSAControlPlane.
 type RosaControlPlaneSpec struct { //nolint: maligned
 	// Cluster name must be valid DNS-1035 label, so it must consist of lower case alphanumeric
@@ -77,6 +91,16 @@ type RosaControlPlaneSpec struct { //nolint: maligned
 	// OpenShift semantic version, for example "4.14.5".
 	Version string `json:"version"`
 
+	// VersionGate requires acknowledgment when upgrading ROSA-HCP y-stream versions (e.g., from 4.15 to 4.16).
+	// Default is WaitForAcknowledge.
+	// WaitForAcknowledge: If acknowledgment is required, the upgrade will not proceed until VersionGate is set to Acknowledge or AlwaysAcknowledge.
+	// Acknowledge: If acknowledgment is required, apply it for the upgrade. After upgrade is done set the version gate to WaitForAcknowledge.
+	// AlwaysAcknowledge: If acknowledgment is required, apply it and proceed with the upgrade.
+	//
+	// +kubebuilder:validation:Enum=Acknowledge;WaitForAcknowledge;AlwaysAcknowledge
+	// +kubebuilder:default=WaitForAcknowledge
+	VersionGate VersionGateAckType `json:"versionGate"`
+
 	// AWS IAM roles used to perform credential requests by the openshift operators.
 	RolesRef AWSRolesRef `json:"rolesRef"`
 
@@ -697,6 +721,9 @@ type RosaControlPlaneStatus struct {
 	ConsoleURL string `json:"consoleURL,omitempty"`
 	// OIDCEndpointURL is the endpoint url for the managed OIDC provider.
 	OIDCEndpointURL string `json:"oidcEndpointURL,omitempty"`
+
+	// Available upgrades for the ROSA hosted control plane.
+	AvailableUpgrades []string `json:"availableUpgrades,omitempty"`
 }
 
 // +kubebuilder:object:root=true

controlplane/rosa/api/v1beta2/zz_generated.deepcopy.go

@@ -410,6 +410,17 @@ func (r *ROSAControlPlaneReconciler) reconcileClusterVersion(rosaScope *scope.RO
 	version := rosaScope.ControlPlane.Spec.Version
 	if version == rosa.RawVersionID(cluster.Version()) {
 		conditions.MarkFalse(rosaScope.ControlPlane, rosacontrolplanev1.ROSAControlPlaneUpgradingCondition, "upgraded", clusterv1.ConditionSeverityInfo, "")
+
+		if cluster.Version() != nil {
+			rosaScope.ControlPlane.Status.AvailableUpgrades = cluster.Version().AvailableUpgrades()
+		}
+
+		// Set the version gate to WaitForAcknowledge as the previous upgrade is applied.
+		if rosaScope.ControlPlane.Spec.VersionGate == rosacontrolplanev1.Acknowledge {
+			rosaScope.ControlPlane.Spec.VersionGate = rosacontrolplanev1.WaitForAcknowledge
+		}
+
+		// return as there is no upgrade to schedule.
 		return nil
 	}
 
@@ -419,9 +430,18 @@ func (r *ROSAControlPlaneReconciler) reconcileClusterVersion(rosaScope *scope.RO
 	}
 
 	if scheduledUpgrade == nil {
-		scheduledUpgrade, err = rosa.ScheduleControlPlaneUpgrade(ocmClient, cluster, version, time.Now())
+		ack := (rosaScope.ControlPlane.Spec.VersionGate == rosacontrolplanev1.Acknowledge || rosaScope.ControlPlane.Spec.VersionGate == rosacontrolplanev1.AlwaysAcknowledge)
+		scheduledUpgrade, err = rosa.ScheduleControlPlaneUpgrade(ocmClient, cluster, version, time.Now(), ack)
 		if err != nil {
-			return fmt.Errorf("failed to schedule control plane upgrade to version %s: %w", version, err)
+			condition := &clusterv1.Condition{
+				Type:    rosacontrolplanev1.ROSAControlPlaneUpgradingCondition,
+				Status:  corev1.ConditionFalse,
+				Reason:  "failed",
+				Message: fmt.Sprintf("failed to schedule upgrade to version %s: %v", version, err),
+			}
+			conditions.Set(rosaScope.ControlPlane, condition)
+
+			return err
 		}
 	}
 

controlplane/rosa/controllers/rosacontrolplane_controller.go

@@ -27,7 +27,7 @@ func CheckExistingScheduledUpgrade(client *ocm.Client, cluster *cmv1.Cluster) (*
 }
 
 // ScheduleControlPlaneUpgrade schedules a new control plane upgrade to the specified version at the specified time.
-func ScheduleControlPlaneUpgrade(client *ocm.Client, cluster *cmv1.Cluster, version string, nextRun time.Time) (*cmv1.ControlPlaneUpgradePolicy, error) {
+func ScheduleControlPlaneUpgrade(client *ocm.Client, cluster *cmv1.Cluster, version string, nextRun time.Time, ack bool) (*cmv1.ControlPlaneUpgradePolicy, error) {
 	// earliestNextRun is set to at least 5 min from now by the OCM API.
 	// Set our next run request to something slightly longer than 5min to make sure we account for the latency between when we send this
 	// request and when the server processes it.
@@ -41,10 +41,32 @@ func ScheduleControlPlaneUpgrade(client *ocm.Client, cluster *cmv1.Cluster, vers
 		ScheduleType(cmv1.ScheduleTypeManual).
 		Version(version).
 		NextRun(nextRun).
+		EnableMinorVersionUpgrades(true).
 		Build()
 	if err != nil {
 		return nil, err
 	}
+
+	versionGates, err := client.GetMissingGateAgreementsHypershift(cluster.ID(), upgradePolicy)
+	if err != nil {
+		return nil, err
+	}
+
+	if !ack && len(versionGates) > 0 {
+		errMess := "version gate acknowledgement required"
+		for id := range versionGates {
+			errMess = fmt.Sprintf(errMess+"\nid:%s\n %s\n %s\n %s\n", versionGates[id].ID(), versionGates[id].Description(), versionGates[id].DocumentationURL(), versionGates[id].WarningMessage())
+		}
+
+		return nil, fmt.Errorf("%s", errMess)
+	}
+
+	for id := range versionGates {
+		if err = client.AckVersionGate(cluster.ID(), versionGates[id].ID()); err != nil {
+			return nil, err
+		}
+	}
+
 	return client.ScheduleHypershiftControlPlaneUpgrade(cluster.ID(), upgradePolicy)
 }
 
@@ -64,6 +86,7 @@ func ScheduleNodePoolUpgrade(client *ocm.Client, clusterID string, nodePool *cmv
 		ScheduleType(cmv1.ScheduleTypeManual).
 		Version(version).
 		NextRun(nextRun).
+		EnableMinorVersionUpgrades(true).
 		Build()
 	if err != nil {
 		return nil, err