Merge pull request #449 from prometheus/beorn7/fix-partitioning-by-outcome

Adjust the partitioning by outcome.

notification/notification.go

@@ -39,11 +39,6 @@ const (
 const (
 	namespace = "prometheus"
 	subsystem = "notifications"
-
-	result  = "result"
-	success = "success"
-	failure = "failure"
-	dropped = "dropped"
 )
 
 var (
@@ -88,7 +83,9 @@ type NotificationHandler struct {
 	// HTTP client with custom timeout settings.
 	httpClient httpPoster
 
-	notificationLatency        *prometheus.SummaryVec
+	notificationLatency        prometheus.Summary
+	notificationErrors         prometheus.Counter
+	notificationDropped        prometheus.Counter
 	notificationsQueueLength   prometheus.Gauge
 	notificationsQueueCapacity prometheus.Metric
 
@@ -103,15 +100,24 @@ func NewNotificationHandler(alertmanagerURL string, notificationQueueCapacity in
 
 		httpClient: utility.NewDeadlineClient(*deadline),
 
-		notificationLatency: prometheus.NewSummaryVec(
+		notificationLatency: prometheus.NewSummary(prometheus.SummaryOpts{
-			prometheus.SummaryOpts{
+			Namespace: namespace,
-				Namespace: namespace,
+			Subsystem: subsystem,
-				Subsystem: subsystem,
+			Name:      "latency_milliseconds",
-				Name:      "latency_milliseconds",
+			Help:      "Latency quantiles for sending alert notifications (not including dropped notifications).",
-				Help:      "Latency quantiles for sending alert notifications.",
+		}),
-			},
+		notificationErrors: prometheus.NewCounter(prometheus.CounterOpts{
-			[]string{result},
+			Namespace: namespace,
-		),
+			Subsystem: subsystem,
+			Name:      "errors_total",
+			Help:      "Total number of errors sending alert notifications.",
+		}),
+		notificationDropped: prometheus.NewCounter(prometheus.CounterOpts{
+			Namespace: namespace,
+			Subsystem: subsystem,
+			Name:      "dropped_total",
+			Help:      "Total number of alert notifications dropped due to alert manager missing in configuration.",
+		}),
 		notificationsQueueLength: prometheus.NewGauge(prometheus.GaugeOpts{
 			Namespace: namespace,
 			Subsystem: subsystem,
@@ -175,22 +181,19 @@ func (n *NotificationHandler) Run() {
 	for reqs := range n.pendingNotifications {
 		if n.alertmanagerURL == "" {
 			glog.Warning("No alert manager configured, not dispatching notification")
-			n.notificationLatency.WithLabelValues(dropped).Observe(0)
+			n.notificationDropped.Inc()
 			continue
 		}
 
 		begin := time.Now()
 		err := n.sendNotifications(reqs)
-		labelValue := success
 
 		if err != nil {
 			glog.Error("Error sending notification: ", err)
-			labelValue = failure
+			n.notificationErrors.Inc()
 		}
 
-		n.notificationLatency.WithLabelValues(labelValue).Observe(
+		n.notificationLatency.Observe(float64(time.Since(begin) / time.Millisecond))
-			float64(time.Since(begin) / time.Millisecond),
-		)
 	}
 	close(n.stopped)
 }

storage/remote/queue_manager.go

@@ -59,7 +59,8 @@ type TSDBQueueManager struct {
 	drained        chan bool
 
 	samplesCount  *prometheus.CounterVec
-	sendLatency   *prometheus.SummaryVec
+	sendLatency   prometheus.Summary
+	sendErrors    prometheus.Counter
 	queueLength   prometheus.Gauge
 	queueCapacity prometheus.Metric
 }
@@ -81,15 +82,18 @@ func NewTSDBQueueManager(tsdb TSDBClient, queueCapacity int) *TSDBQueueManager {
 			},
 			[]string{result},
 		),
-		sendLatency: prometheus.NewSummaryVec(
+		sendLatency: prometheus.NewSummary(prometheus.SummaryOpts{
-			prometheus.SummaryOpts{
+			Namespace: namespace,
-				Namespace: namespace,
+			Subsystem: subsystem,
-				Subsystem: subsystem,
+			Name:      "sent_latency_milliseconds",
-				Name:      "sent_latency_milliseconds",
+			Help:      "Latency quantiles for sending sample batches to the remote TSDB.",
-				Help:      "Latency quantiles for sending samples to the remote TSDB.",
+		}),
-			},
+		sendErrors: prometheus.NewCounter(prometheus.CounterOpts{
-			[]string{result},
+			Namespace: namespace,
-		),
+			Subsystem: subsystem,
+			Name:      "sent_errors_total",
+			Help:      "Total number of errors sending sample batches to the remote TSDB.",
+		}),
 		queueLength: prometheus.NewGauge(prometheus.GaugeOpts{
 			Namespace: namespace,
 			Subsystem: subsystem,
@@ -164,9 +168,10 @@ func (t *TSDBQueueManager) sendSamples(s clientmodel.Samples) {
 	if err != nil {
 		glog.Warningf("error sending %d samples to TSDB: %s", len(s), err)
 		labelValue = failure
+		t.sendErrors.Inc()
 	}
 	t.samplesCount.WithLabelValues(labelValue).Add(float64(len(s)))
-	t.sendLatency.WithLabelValues(labelValue).Observe(float64(duration))
+	t.sendLatency.Observe(float64(duration))
 }
 
 // Run continuously sends samples to the TSDB.