Add the beacon ID to some metrics (#969)

chain/beacon/store.go

@@ -9,6 +9,7 @@ import (
 
 	clock "github.com/jonboulle/clockwork"
 
+	"github.com/drand/drand/common"
 	"github.com/drand/drand/common/scheme"
 
 	"github.com/drand/drand/chain"
@@ -113,13 +114,20 @@ func (d *discrepancyStore) Put(b *chain.Beacon) error {
 	if err := d.Store.Put(b); err != nil {
 		return err
 	}
+
+	beaconID := d.group.ID
+	if beaconID == "" {
+		beaconID = common.DefaultBeaconID
+	}
+
 	actual := d.clock.Now().UnixNano()
 	expected := chain.TimeOfRound(d.group.Period, d.group.GenesisTime, b.Round) * 1e9
 	discrepancy := float64(actual-expected) / float64(time.Millisecond)
-	metrics.BeaconDiscrepancyLatency.Set(float64(actual-expected) / float64(time.Millisecond))
-	metrics.LastBeaconRound.Set(float64(b.GetRound()))
-	metrics.GroupSize.Set(float64(d.group.Len()))
-	metrics.GroupThreshold.Set(float64(d.group.Threshold))
+
+	metrics.BeaconDiscrepancyLatency.WithLabelValues(beaconID).Set(float64(actual-expected) / float64(time.Millisecond))
+	metrics.LastBeaconRound.WithLabelValues(beaconID).Set(float64(b.GetRound()))
+	metrics.GroupSize.WithLabelValues(beaconID).Set(float64(d.group.Len()))
+	metrics.GroupThreshold.WithLabelValues(beaconID).Set(float64(d.group.Threshold))
 	d.l.Infow("", "NEW_BEACON_STORED", b.String(), "time_discrepancy_ms", discrepancy)
 	return nil
 }

metrics/metrics.go

@@ -53,35 +53,41 @@ var (
 		Name: "api_call_counter",
 		Help: "Number of API calls that we have received",
 	}, []string{"api_method"})
+
 	// GroupDialFailures (Group) how manuy failures connecting outbound
 	GroupDialFailures = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Name: "dial_failures",
 		Help: "Number of times there have been network connection issues",
 	}, []string{"peer_address"})
+
 	// OutgoingConnections (Group) how many GrpcClient connections are present
 	OutgoingConnections = prometheus.NewGauge(prometheus.GaugeOpts{
 		Name: "outgoing_group_connections",
 		Help: "Number of peers with current outgoing GrpcClient connections",
 	})
-	GroupSize = prometheus.NewGauge(prometheus.GaugeOpts{
+
+	GroupSize = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: "group_size",
 		Help: "Number of peers in the current group",
-	})
-	GroupThreshold = prometheus.NewGauge(prometheus.GaugeOpts{
+	}, []string{"beacon_id"})
+
+	GroupThreshold = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: "group_threshold",
 		Help: "Number of shares needed for beacon reconstruction",
-	})
+	}, []string{"beacon_id"})
+
 	// BeaconDiscrepancyLatency (Group) millisecond duration between time beacon created and
 	// calculated time of round.
-	BeaconDiscrepancyLatency = prometheus.NewGauge(prometheus.GaugeOpts{
+	BeaconDiscrepancyLatency = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: "beacon_discrepancy_latency",
 		Help: "Discrepancy between beacon creation time and calculated round time",
-	})
+	}, []string{"beacon_id"})
+
 	// LastBeaconRound is the most recent round (as also seen at /health) stored.
-	LastBeaconRound = prometheus.NewGauge(prometheus.GaugeOpts{
+	LastBeaconRound = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: "last_beacon_round",
 		Help: "Last locally stored beacon",
-	})
+	}, []string{"beacon_id"})
 
 	// HTTPCallCounter (HTTP) how many http requests
 	HTTPCallCounter = prometheus.NewCounterVec(prometheus.CounterOpts{