Merge pull request #1148 from cdlliuy/add_canary_analysis_result_as_metric

Add canary analysis result as Prometheus metrics

docs/gitbook/usage/monitoring.md

@@ -117,4 +117,8 @@ flagger_canary_duration_seconds_bucket{name="podinfo",namespace="test",le="10"}
 flagger_canary_duration_seconds_bucket{name="podinfo",namespace="test",le="+Inf"} 6
 flagger_canary_duration_seconds_sum{name="podinfo",namespace="test"} 17.3561329
 flagger_canary_duration_seconds_count{name="podinfo",namespace="test"} 6
+
+# Last canary metric analysis result per different metrics
+flagger_canary_metric_analysis{metric="podinfo-http-successful-rate",name="podinfo",namespace="test"} 1
+flagger_canary_metric_analysis{metric="podinfo-custom-metric",name="podinfo",namespace="test"} 0.918223108974359
 ```

pkg/controller/scheduler_metrics.go

@@ -146,7 +146,7 @@ func (c *Controller) runBuiltinMetricChecks(canary *flaggerv1.Canary) bool {
 				}
 				return false
 			}
-
+			c.recorder.SetAnalysis(canary, metric.Name, val)
 			if metric.ThresholdRange != nil {
 				tr := *metric.ThresholdRange
 				if tr.Min != nil && val < *tr.Min {
@@ -177,6 +177,7 @@ func (c *Controller) runBuiltinMetricChecks(canary *flaggerv1.Canary) bool {
 				}
 				return false
 			}
+			c.recorder.SetAnalysis(canary, metric.Name, val.Seconds())
 			if metric.ThresholdRange != nil {
 				tr := *metric.ThresholdRange
 				if tr.Min != nil && val < time.Duration(*tr.Min)*time.Millisecond {
@@ -209,6 +210,7 @@ func (c *Controller) runBuiltinMetricChecks(canary *flaggerv1.Canary) bool {
 				}
 				return false
 			}
+			c.recorder.SetAnalysis(canary, metric.Name, val)
 			if metric.ThresholdRange != nil {
 				tr := *metric.ThresholdRange
 				if tr.Min != nil && val < *tr.Min {
@@ -283,6 +285,8 @@ func (c *Controller) runMetricChecks(canary *flaggerv1.Canary) bool {
 				return false
 			}
 
+			c.recorder.SetAnalysis(canary, metric.Name, val)
+
 			if metric.ThresholdRange != nil {
 				tr := *metric.ThresholdRange
 				if tr.Min != nil && val < *tr.Min {

pkg/metrics/recorder.go

@@ -31,6 +31,7 @@ type Recorder struct {
 	total    *prometheus.GaugeVec
 	status   *prometheus.GaugeVec
 	weight   *prometheus.GaugeVec
+	analysis *prometheus.GaugeVec
 }
 
 // NewRecorder creates a new recorder and registers the Prometheus metrics
@@ -67,12 +68,19 @@ func NewRecorder(controller string, register bool) Recorder {
 		Help:      "The virtual service destination weight current value",
 	}, []string{"workload", "namespace"})
 
+	analysis := prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Subsystem: controller,
+		Name:      "canary_metric_analysis",
+		Help:      "Last canary analysis result per metric",
+	}, []string{"name", "namespace", "metric"})
+
 	if register {
 		prometheus.MustRegister(info)
 		prometheus.MustRegister(duration)
 		prometheus.MustRegister(total)
 		prometheus.MustRegister(status)
 		prometheus.MustRegister(weight)
+		prometheus.MustRegister(analysis)
 	}
 
 	return Recorder{
@@ -81,6 +89,7 @@ func NewRecorder(controller string, register bool) Recorder {
 		total:    total,
 		status:   status,
 		weight:   weight,
+		analysis: analysis,
 	}
 }
 
@@ -99,6 +108,10 @@ func (cr *Recorder) SetTotal(namespace string, total int) {
 	cr.total.WithLabelValues(namespace).Set(float64(total))
 }
 
+func (cr *Recorder) SetAnalysis(cd *flaggerv1.Canary, metricTemplateName string, val float64) {
+	cr.analysis.WithLabelValues(cd.Spec.TargetRef.Name, cd.Namespace, metricTemplateName).Set(val)
+}
+
 // SetStatus sets the last known canary analysis status
 func (cr *Recorder) SetStatus(cd *flaggerv1.Canary, phase flaggerv1.CanaryPhase) {
 	var status int