set metric on errors

Signed-off-by: Will Kutler <wkutler@redhat.com>

controllers/configurationpolicy_controller.go

@@ -94,16 +94,29 @@ var policyEvalCounter = prometheus.NewCounterVec(
 	[]string{"name"},
 )
 
-// var policyUserErrorsGauge = prometheus.NewGaugeVec(
-// 	prometheus.GaugeOpts{
-// 		Name: "common_related_objects",
-// 		Help: "A gauge vector of related objects managed by multiple policies.",
-// 	},
-// 	[]string{
-// 		"relatedObject",
-// 		"policy",
-// 	},
-// )
+var policyUserErrorsGauge = prometheus.NewGaugeVec(
+	prometheus.GaugeOpts{
+		Name: "policy_user_errors",
+		Help: "The number of user errors encountered while processing policies",
+	},
+	[]string{
+		"policy",
+		"template",
+		"type",
+	},
+)
+
+var policySystemErrorsGauge = prometheus.NewGaugeVec(
+	prometheus.GaugeOpts{
+		Name: "policy_system_errors",
+		Help: "The number of system errors encountered while processing policies",
+	},
+	[]string{
+		"policy",
+		"template",
+		"type",
+	},
+)
 
 var compareObjSecondsCounter = prometheus.NewCounterVec(
 	prometheus.CounterOpts{
@@ -130,6 +143,8 @@ func init() {
 	metrics.Registry.MustRegister(policyEvalCounter)
 	metrics.Registry.MustRegister(compareObjSecondsCounter)
 	metrics.Registry.MustRegister(compareObjEvalCounter)
+	metrics.Registry.MustRegister(policySystemErrorsGauge)
+	metrics.Registry.MustRegister(policyUserErrorsGauge)
 }
 
 // SetupWithManager sets up the controller with the Manager.
@@ -192,6 +207,8 @@ func (r *ConfigurationPolicyReconciler) Reconcile(ctx context.Context, request c
 		_ = policyEvalCounter.DeleteLabelValues(request.Name)
 		_ = compareObjEvalCounter.DeletePartialMatch(prometheus.Labels{"config_policy_name": request.Name})
 		_ = compareObjSecondsCounter.DeletePartialMatch(prometheus.Labels{"config_policy_name": request.Name})
+		_ = policyUserErrorsGauge.DeletePartialMatch(prometheus.Labels{"template": request.Name})
+		_ = policySystemErrorsGauge.DeletePartialMatch(prometheus.Labels{"template": request.Name})
 	}
 
 	return reconcile.Result{}, nil
@@ -616,7 +633,6 @@ func (r *ConfigurationPolicyReconciler) handleObjectTemplates(plc policyv1.Confi
 	relatedObjects := []policyv1.RelatedObject{}
 	parentStatusUpdateNeeded := false
 
-	//TODO-METRIC - user error
 	validationErr := ""
 	if plc.Spec == nil {
 		validationErr = "Policy does not have a Spec specified"
@@ -637,6 +653,9 @@ func (r *ConfigurationPolicyReconciler) handleObjectTemplates(plc policyv1.Confi
 
 		r.checkRelatedAndUpdate(plc, relatedObjects, oldRelated, statusChanged)
 
+		parent := plc.OwnerReferences[0]
+		policyUserErrorsGauge.WithLabelValues(parent.Name, plc.GetName(), "invalid-template").Add(1)
+
 		return
 	}
 
@@ -1635,8 +1654,11 @@ func (r *ConfigurationPolicyReconciler) getMapping(
 			kind := afterPrefix[0:(strings.Index(afterPrefix, "\" "))]
 			mappingErrMsg = "couldn't find mapping resource with kind " + kind +
 				", please check if you have CRD deployed"
-			// TODO-METRIC - user error - missing CRD
+
 			log.Error(err, "Could not map resource, do you have the CRD deployed?", "kind", kind)
+
+			parent := policy.OwnerReferences[0]
+			policyUserErrorsGauge.WithLabelValues(parent.Name, policy.GetName(), "no-object-CRD").Add(1)
 		}
 
 		errMsg := err.Error()
@@ -2457,8 +2479,10 @@ func (r *ConfigurationPolicyReconciler) addForUpdate(policy *policyv1.Configurat
 	if k8serrors.IsConflict(err) {
 		policyLog.Error(err, "Tried to re-update status before previous update could be applied, retrying next loop")
 	} else if err != nil {
-		// TODO-METRIC - system error - can't send a status update
 		policyLog.Error(err, "Could not update status, retrying next loop")
+
+		parent := policy.OwnerReferences[0]
+		policySystemErrorsGauge.WithLabelValues(parent.Name, policy.GetName(), "status-update-failed").Add(1)
 	}
 }