Add a metric for number of reconnections

README.md

@@ -65,6 +65,7 @@ Go-related metrics are available. We also have custom metrics:
 - `received_events_counter`
 - `relay_connection_subscriptions_gauge`
 - `relay_connection_received_messages_counter`
+- `relay_connection_reconnections_counter`
 
 See `service/adapters/prometheus`.
 

service/adapters/prometheus/prometheus.go

@@ -40,6 +40,7 @@ type Prometheus struct {
 	receivedEventsCounter                   *prometheus.CounterVec
 	relayConnectionSubscriptionsGauge       *prometheus.GaugeVec
 	relayConnectionReceivedMessagesCounter  *prometheus.CounterVec
+	relayConnectionReconnectionsCounter     *prometheus.CounterVec
 
 	registry *prometheus.Registry
 
@@ -109,6 +110,13 @@ func NewPrometheus(logger logging.Logger) (*Prometheus, error) {
 		},
 		[]string{labelAddress, labelMessageType, labelResult},
 	)
+	relayConnectionReconnectionsCounter := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "relay_connection_reconnections_counter",
+			Help: "Number of reconnections to a relay.",
+		},
+		[]string{labelAddress},
+	)
 
 	reg := prometheus.NewRegistry()
 	for _, v := range []prometheus.Collector{
@@ -121,6 +129,7 @@ func NewPrometheus(logger logging.Logger) (*Prometheus, error) {
 		receivedEventsCounter,
 		relayConnectionSubscriptionsGauge,
 		relayConnectionReceivedMessagesCounter,
+		relayConnectionReconnectionsCounter,
 		collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}),
 		collectors.NewGoCollector(),
 	} {
@@ -152,6 +161,7 @@ func NewPrometheus(logger logging.Logger) (*Prometheus, error) {
 		receivedEventsCounter:                   receivedEventsCounter,
 		relayConnectionSubscriptionsGauge:       relayConnectionSubscriptionsGauge,
 		relayConnectionReceivedMessagesCounter:  relayConnectionReceivedMessagesCounter,
+		relayConnectionReconnectionsCounter:     relayConnectionReconnectionsCounter,
 
 		registry: reg,
 
@@ -202,6 +212,10 @@ func (p *Prometheus) ReportMessageReceived(address domain.RelayAddress, messageT
 	p.relayConnectionReceivedMessagesCounter.With(labels).Inc()
 }
 
+func (p *Prometheus) ReportRelayReconnection(address domain.RelayAddress) {
+	p.relayConnectionReconnectionsCounter.With(prometheus.Labels{labelAddress: address.String()}).Inc()
+}
+
 func (p *Prometheus) Registry() *prometheus.Registry {
 	return p.registry
 }

service/domain/relays/relay_connection.go

@@ -154,6 +154,7 @@ func (r *RelayConnection) run(ctx context.Context) error {
 	defer r.setState(RelayConnectionStateDisconnected)
 
 	r.logger.Trace().Message("connecting")
+	r.metrics.ReportRelayReconnection(r.address)
 
 	conn, _, err := websocket.DefaultDialer.DialContext(ctx, r.address.String(), nil)
 	if err != nil {

service/domain/relays/relay_connections.go

@@ -13,6 +13,7 @@ type Metrics interface {
 	ReportRelayConnectionsState(m map[domain.RelayAddress]RelayConnectionState)
 	ReportNumberOfSubscriptions(address domain.RelayAddress, n int)
 	ReportMessageReceived(address domain.RelayAddress, messageType MessageType, err *error)
+	ReportRelayReconnection(address domain.RelayAddress)
 }
 
 type MessageType struct {