RHIROS-1326 Kruize 0.0.20_rm Integration

RedHatInsights · Sep 28, 2023 · 01b75a8 · 01b75a8
1 parent 6f54036
commit 01b75a8
Show file tree

Hide file tree

Showing 7 changed files with 223 additions and 169 deletions.
diff --git a/Makefile b/Makefile
@@ -118,9 +118,9 @@ upload-msg-to-rosocp:
 get-recommendations:
 ifdef env
 	$(eval APIPOD=$(shell oc get pods -o custom-columns=POD:.metadata.name --no-headers -n ${env} | grep ros-ocp-backend-api))
-	oc exec ${APIPOD} -c ros-ocp-backend-api -n ${env} -- /bin/bash -c 'curl -s -H "X-Rh-Identity: ${b64_identity}" -H "x-rh-request_id: testtesttest" http://localhost:8000/api/cost-management/v1/recommendations/openshift' | python -m json.tool
+	oc exec ${APIPOD} -c ros-ocp-backend-api -n ${env} -- /bin/bash -c 'curl -s -H "X-Rh-Identity: ${b64_identity}" -H "x-rh-request_id: testtesttest" http://localhost:8000/api/cost-management/v1/recommendations/openshift?start_date=1992-12-26' | python -m json.tool
 else
 	curl -s -H "x-rh-identity: ${b64_identity}" \
 		 -H "x-rh-request_id: testtesttest" \
-		 http://localhost:8000/api/cost-management/v1/recommendations/openshift | python -m json.tool
+		 http://localhost:8000/api/cost-management/v1/recommendations/openshift?start_date=1992-12-26 | python -m json.tool
 endif
diff --git a/internal/api/utils.go b/internal/api/utils.go
@@ -223,11 +223,6 @@ func TransformComponentUnits(jsonData datatypes.JSON) map[string]interface{} {
 		return nil
 	}
 
-	durationBased, ok := data["duration_based"].(map[string]interface{})
-	if !ok {
-		fmt.Printf("duration_based not found in JSON")
-	}
-
 	convertMemory := func(memory map[string]interface{}) error {
 		amount, ok := memory["amount"].(float64)
 		if ok {
@@ -274,6 +269,11 @@ func TransformComponentUnits(jsonData datatypes.JSON) map[string]interface{} {
 		return nil
 	}
 
+	durationBased, ok := data["duration_based"].(map[string]interface{})
+	if !ok {
+		fmt.Printf("duration_based not found in JSON")
+	}
+
 	/*
 		Recommendation data is available for three periods
 		For each of these actual values will be present in

diff --git a/internal/services/report_processor.go b/internal/services/report_processor.go
@@ -199,7 +199,7 @@ func ProcessReport(msg *kafka.Message) {
 				continue
 			}
 
-			if kruize.Is_valid_recommendation(recommendation) {
+			if kruize.Is_valid_recommendation(recommendation, experiment_name, maxEndTime.String()) {
 				containers := recommendation[0].Kubernetes_objects[0].Containers
 				for _, container := range containers {
 					for _, v := range container.Recommendations.Data {
@@ -212,8 +212,8 @@ func ProcessReport(msg *kafka.Message) {
 						recommendationSet := model.RecommendationSet{
 							WorkloadID:          workload.ID,
 							ContainerName:       container.Container_name,
-							MonitoringStartTime: v.Duration_based.Short_term.Monitoring_start_time,
-							MonitoringEndTime:   v.Duration_based.Short_term.Monitoring_end_time,
+							MonitoringStartTime: v.RecommendationTerms.Short_term.MonitoringStartTime,
+							MonitoringEndTime:   v.MonitoringEndTime,
 							Recommendations:     marshalData,
 						}
 						if err := recommendationSet.CreateRecommendationSet(); err != nil {
@@ -227,8 +227,8 @@ func ProcessReport(msg *kafka.Message) {
 						historicalRecommendationSet := model.HistoricalRecommendationSet{
 							WorkloadID:          workload.ID,
 							ContainerName:       container.Container_name,
-							MonitoringStartTime: v.Duration_based.Short_term.Monitoring_start_time,
-							MonitoringEndTime:   v.Duration_based.Short_term.Monitoring_end_time,
+							MonitoringStartTime: v.RecommendationTerms.Short_term.MonitoringStartTime,
+							MonitoringEndTime:   v.MonitoringEndTime,
 							Recommendations:     marshalData,
 						}
 						if err := historicalRecommendationSet.CreateHistoricalRecommendationSet(); err != nil {

diff --git a/internal/types/kruizePayload/common.go b/internal/types/kruizePayload/common.go
@@ -38,36 +38,47 @@ type aggregation_info struct {
 }
 
 type recommendation struct {
-	Data          map[string]recommendationType `json:"data,omitempty"`
+	Version       string                		`json:"version,omitempty"`
+	Data          map[string]RecommendationData `json:"data,omitempty"`
 	Notifications map[string]notification       `json:"notifications,omitempty"`
 }
 
+
 type notification struct {
 	NotifyType string `json:"type,omitempty"`
 	Message    string `json:"message,omitempty"`
 	Code       int    `json:"code,omitempty"`
 }
 
-type recommendationType struct {
-	Duration_based termbased `json:"duration_based,omitempty"`
+type RecommendationEngineObject struct {
+	PodsCount        int             		 `json:"pods_count,omitempty"`
+	ConfidenceLevel  float64         		 `json:"confidence_level,omitempty"`
+	Config           ConfigObject			 `json:"config,omitempty"`
+	Variation        ConfigObject			 `json:"variation,omitempty"`
+	Notifications    map[string]notification `json:"notifications,omitempty"`
+}
+
+type RecommendationData struct {
+	Notifications       map[string]notification `json:"notifications"`
+	MonitoringEndTime   time.Time 				`json:"monitoring_end_time"`
+	Current             ConfigObject 			`json:"current"`
+	RecommendationTerms TermBased 				`json:"recommendation_terms"`
 }
 
-type termbased struct {
-	Short_term  recommendationObject `json:"short_term,omitempty"`
-	Medium_term recommendationObject `json:"medium_term,omitempty"`
-	Long_term   recommendationObject `json:"long_term,omitempty"`
+type RecommendationTerm struct {
+	DurationInHours 		float64 					`json:"duration_in_hours"`
+	Notifications   		map[string]notification 	`json:"notifications"`
+	MonitoringStartTime     time.Time
+	RecommendationEngines   struct {
+		Cost        RecommendationEngineObject `json:"cost"`
+		Performance RecommendationEngineObject `json:"performance"`
+	} `json:"recommendation_engines"`
 }
 
-type recommendationObject struct {
-	Monitoring_start_time time.Time               `json:"monitoring_start_time,omitempty"`
-	Monitoring_end_time   time.Time               `json:"monitoring_end_time,omitempty"`
-	Duration_in_hours     float64                 `json:"duration_in_hours,omitempty"`
-	Pods_count            int                     `json:"pods_count,omitempty"`
-	Confidence_level      float64                 `json:"confidence_level,omitempty"`
-	Current               ConfigObject            `json:"current,omitempty"`
-	Config                ConfigObject            `json:"config,omitempty"`
-	Variation             ConfigObject            `json:"variation,omitempty"`
-	Notifications         map[string]notification `json:"notifications,omitempty"`
+type TermBased struct {
+	Short_term  RecommendationTerm `json:"short_term,omitempty"`
+	Medium_term RecommendationTerm `json:"medium_term,omitempty"`
+	Long_term   RecommendationTerm `json:"long_term,omitempty"`
 }
 
 type ConfigObject struct {

diff --git a/internal/utils/kruize/kruize_api.go b/internal/utils/kruize/kruize_api.go
@@ -176,14 +176,50 @@ func Update_recommendations(experiment_name string, interval_end_time time.Time)
 
 }
 
-func Is_valid_recommendation(d []kruizePayload.ListRecommendations) bool {
+func Is_valid_recommendation(d []kruizePayload.ListRecommendations, experiment_name string, maxEndTime string) bool {
 	if len(d) > 0 {
+
+		// To maintain a local reference the following map has been created from 
+		// https://github.com/kruize/autotune/blob/master/design/NotificationCodes.md#detailed-codes
+		notificationCodeValidities := map[string]string{
+			"111000": "INFO",
+			"120001": "INFO",
+			"221001": "ERROR",
+			"221002": "ERROR",
+			"221003": "ERROR",
+			"221004": "ERROR",
+			"223001": "ERROR",
+			"223002": "ERROR",
+			"223003": "ERROR",
+			"223004": "ERROR",
+			"224001": "ERROR",
+			"224002": "ERROR",
+			"224003": "ERROR",
+			"224004": "ERROR",
+		}
+
 		notifications := d[0].Kubernetes_objects[0].Containers[0].Recommendations.Notifications
-		// 112101 is notification code for "Duration Based Recommendations Available".
-		if _, ok := notifications["112101"]; ok {
-			return true
-		} else {
-			return false
+
+		for key := range notifications{
+			notificationType, keyExists := notificationCodeValidities[key]
+			if !keyExists {
+				return false
+			}
+
+			if key == "111000" && notificationType == "INFO"{
+				notificationsLevelTwo := d[0].Kubernetes_objects[0].Containers[0].Recommendations.Data[maxEndTime].Notifications
+				for key := range notificationsLevelTwo{
+					if notificationCodeValidities[key] == "ERROR"{
+						kruizeInvalidRecommendationDetail.WithLabelValues(key, experiment_name).Set(1)
+					}
+				}
+				return true
+			} else {
+				// Setting the metric counter to 1 as we expect a single metric
+				// for a combination of notification_code and experiment_name
+				kruizeInvalidRecommendationDetail.WithLabelValues(key, experiment_name).Set(1)
+				return false
+			}
 		}
 	}
 	return false

diff --git a/internal/utils/kruize/metrics.go b/internal/utils/kruize/metrics.go
@@ -13,3 +13,14 @@ var (
 		[]string{"path"},
 	)
 )
+
+
+var (
+    kruizeInvalidRecommendationDetail = promauto.NewGaugeVec(
+        prometheus.GaugeOpts{
+            Name: "rosocp_kruize_invalid_recommendation_detail",
+            Help: "List of INFO/ERROR type recommendations from Kruize",
+        },
+        []string{"notification_code", "experiment_name"},
+    )
+)