operator: add skew doc links

Author: djoshy

install/0000_90_machine-config_01_prometheus-rules.yaml

@@ -49,7 +49,7 @@ spec:
             severity: info
           annotations:
             summary: "Boot image skew enforcement is disabled. Scaling operations may not be successful."
-            description: "Boot image skew enforcement mode is set to None. When scaling up, new nodes may be provisioned with older boot images that could introduce compatibility issues. Consider manually updating boot images to match the cluster version. Please refer to docs at [TODO-INSERTLINK] for additional details."            
+            description: "Boot image skew enforcement mode is set to None. When scaling up, new nodes may be provisioned with older boot images that could introduce compatibility issues. Consider manually updating boot images to match the cluster version. Please refer to docs at https://docs.redhat.com/en/documentation/openshift_container_platform/latest/html/machine_configuration/mco-update-boot-skew-mgmt for additional details."            
 ---
 apiVersion: monitoring.coreos.com/v1
 kind: PrometheusRule

pkg/operator/status.go

@@ -692,8 +692,14 @@ func (optr *Operator) checkBootImageSkewUpgradeableGuard() (bool, string, error)
 	}
 
 	if skewLimitExceeded {
-		// TODO: Update error message; tracked in https://issues.redhat.com/browse/MCO-2034
-		return true, fmt.Sprintf("Upgrades have been disabled because %s. To enable upgrades, please update your boot images following the documentation at [TODO: insert link], or disable boot image skew enforcement at [TODO: insert link]", skewLimitExceededMessage), nil
+		docsVersion := "latest"
+		if ocpVersion := optr.getCurrentOCPVersionFromClusterVersion(); ocpVersion != "" {
+			if parts := strings.SplitN(ocpVersion, ".", 3); len(parts) >= 2 {
+				docsVersion = parts[0] + "." + parts[1]
+			}
+		}
+		docsURL := fmt.Sprintf("https://docs.redhat.com/en/documentation/openshift_container_platform/%s/html/machine_configuration/mco-update-boot-skew-mgmt", docsVersion)
+		return true, fmt.Sprintf("Upgrades have been disabled because %s. To enable upgrades, please update your boot images or disable boot image skew enforcement by following the documentation at %s", skewLimitExceededMessage, docsURL), nil
 	}
 
 	return false, "", nil

pkg/operator/sync.go

@@ -2550,7 +2550,7 @@ func (optr *Operator) syncBootImageSkewEnforcementStatus(mcop *opv1.MachineConfi
 	}
 
 	// Grab install time OCP version
-	ocpVersionAtInstall := optr.getOCPVersionFromClusterVersion()
+	ocpVersionAtInstall := optr.getOCPInstallVersionFromClusterVersion()
 
 	// Spec override takes priority over all platform defaults.
 	if mcop.Spec.BootImageSkewEnforcement != (opv1.BootImageSkewEnforcementConfig{}) {
@@ -2591,10 +2591,10 @@ func (optr *Operator) syncBootImageSkewEnforcementStatus(mcop *opv1.MachineConfi
 	newMachineConfigurationStatus.BootImageSkewEnforcementStatus = apihelpers.GetSkewEnforcementStatusManualWithOCPVersion(ocpVersionAtInstall)
 }
 
-// getOCPVersionFromClusterVersion extracts the OCP version from ClusterVersion history.
-// It finds the last completed update in history (install version) and parses it to a clean version string.
+// getOCPInstallVersionFromClusterVersion extracts the original install version from ClusterVersion history.
+// It finds the oldest completed update in history and parses it to a clean version string.
 // Returns an empty string if ClusterVersion cannot be retrieved or parsed.
-func (optr *Operator) getOCPVersionFromClusterVersion() string {
+func (optr *Operator) getOCPInstallVersionFromClusterVersion() string {
 	clusterVersion, err := optr.clusterVersionLister.Get("version")
 	if err != nil {
 		klog.Warningf("Failed to get ClusterVersion: %v, skipping boot image skew enforcement configuration", err)
@@ -2627,6 +2627,30 @@ func (optr *Operator) getOCPVersionFromClusterVersion() string {
 	return fmt.Sprintf("%d.%d.%d", parsedVersion.Major(), parsedVersion.Minor(), parsedVersion.Patch())
 }
 
+// getCurrentOCPVersionFromClusterVersion extracts the current running OCP version from ClusterVersion history.
+// It finds the most recent completed update in history and parses it to a clean version string.
+// Returns an empty string if ClusterVersion cannot be retrieved or parsed.
+func (optr *Operator) getCurrentOCPVersionFromClusterVersion() string {
+	clusterVersion, err := optr.clusterVersionLister.Get("version")
+	if err != nil {
+		klog.Warningf("Failed to get ClusterVersion: %v", err)
+		return ""
+	}
+	// History is ordered by recency (newest first); find the first completed entry.
+	for _, h := range clusterVersion.Status.History {
+		if h.State == configv1.CompletedUpdate {
+			parsedVersion, err := k8sversion.ParseGeneric(h.Version)
+			if err != nil {
+				klog.Warningf("Failed to parse current OCP version %q: %v", h.Version, err)
+				return ""
+			}
+			return fmt.Sprintf("%d.%d.%d", parsedVersion.Major(), parsedVersion.Minor(), parsedVersion.Patch())
+		}
+	}
+	klog.Warningf("ClusterVersion has no completed updates in history")
+	return ""
+}
+
 // isBareMetalOnLegacyProvisioningPath checks whether the BareMetal cluster is still using the
 // legacy qcow2-based provisioning path. It does so by inspecting the provisioningOSDownloadURL
 // field of the singleton Provisioning CR (metal3.io/v1alpha1).