Add get_registration_delay_milliseconds metric (#2275)

signal/README.md

@@ -90,6 +90,9 @@ The Signal Server exposes the following metrics in Prometheus format:
 - **registration_delay_milliseconds**: A Histogram metric that measures the time
   it took to register a peer in
   milliseconds.
+- **get_registration_delay_milliseconds**: A Histogram metric that measures the time
+  it took to get a peer registration in
+  milliseconds.
 - **messages_forwarded_total**: A Counter metric that counts the total number of
   messages forwarded between peers.
 - **message_forward_failures_total**: A Counter metric that counts the total

signal/metrics/app.go

@@ -15,6 +15,7 @@ type AppMetrics struct {
 	Deregistrations      metric.Int64Counter
 	RegistrationFailures metric.Int64Counter
 	RegistrationDelay    metric.Float64Histogram
+	GetRegistrationDelay metric.Float64Histogram
 
 	MessagesForwarded      metric.Int64Counter
 	MessageForwardFailures metric.Int64Counter
@@ -54,6 +55,12 @@ func NewAppMetrics(meter metric.Meter) (*AppMetrics, error) {
 		return nil, err
 	}
 
+	getRegistrationDelay, err := meter.Float64Histogram("get_registration_delay_milliseconds",
+		metric.WithExplicitBucketBoundaries(getStandardBucketBoundaries()...))
+	if err != nil {
+		return nil, err
+	}
+
 	messagesForwarded, err := meter.Int64Counter("messages_forwarded_total")
 	if err != nil {
 		return nil, err
@@ -80,6 +87,7 @@ func NewAppMetrics(meter metric.Meter) (*AppMetrics, error) {
 		Deregistrations:      deregistrations,
 		RegistrationFailures: registrationFailures,
 		RegistrationDelay:    registrationDelay,
+		GetRegistrationDelay: getRegistrationDelay,
 
 		MessagesForwarded:      messagesForwarded,
 		MessageForwardFailures: messageForwardFailures,

signal/server/signal.go

@@ -30,6 +30,10 @@ const (
 	labelErrorMissingId    = "missing_id"
 	labelErrorMissingMeta  = "missing_meta"
 	labelErrorFailedHeader = "failed_header"
+
+	labelRegistrionStatus     = "status"
+	labelRegistrationFound    = "found"
+	labelRegistrationNotFound = "not_found"
 )
 
 // Server an instance of a Signal server
@@ -63,7 +67,10 @@ func (s *Server) Send(ctx context.Context, msg *proto.EncryptedMessage) (*proto.
 		return nil, fmt.Errorf("peer %s is not registered", msg.Key)
 	}
 
+	getRegistrationStart := time.Now()
+
 	if dstPeer, found := s.registry.Get(msg.RemoteKey); found {
+		s.metrics.GetRegistrationDelay.Record(ctx, float64(time.Since(getRegistrationStart).Nanoseconds())/1e6, metric.WithAttributes(attribute.String(labelType, labelTypeMessage), attribute.String(labelRegistrionStatus, labelRegistrationFound)))
 		start := time.Now()
 		//forward the message to the target peer
 		if err := dstPeer.Stream.Send(msg); err != nil {
@@ -76,6 +83,7 @@ func (s *Server) Send(ctx context.Context, msg *proto.EncryptedMessage) (*proto.
 			s.metrics.MessagesForwarded.Add(context.Background(), 1)
 		}
 	} else {
+		s.metrics.GetRegistrationDelay.Record(ctx, float64(time.Since(getRegistrationStart).Nanoseconds())/1e6, metric.WithAttributes(attribute.String(labelType, labelTypeMessage), attribute.String(labelRegistrionStatus, labelRegistrationNotFound)))
 		log.Debugf("message from peer [%s] can't be forwarded to peer [%s] because destination peer is not connected", msg.Key, msg.RemoteKey)
 		//todo respond to the sender?
 
@@ -125,8 +133,11 @@ func (s *Server) ConnectStream(stream proto.SignalExchange_ConnectStreamServer)
 
 		log.Debugf("received a new message from peer [%s] to peer [%s]", p.Id, msg.RemoteKey)
 
+		getRegistrationStart := time.Now()
+
 		// lookup the target peer where the message is going to
 		if dstPeer, found := s.registry.Get(msg.RemoteKey); found {
+			s.metrics.GetRegistrationDelay.Record(stream.Context(), float64(time.Since(getRegistrationStart).Nanoseconds())/1e6, metric.WithAttributes(attribute.String(labelType, labelTypeStream), attribute.String(labelRegistrionStatus, labelRegistrationFound)))
 			start := time.Now()
 			//forward the message to the target peer
 			if err := dstPeer.Stream.Send(msg); err != nil {
@@ -139,10 +150,10 @@ func (s *Server) ConnectStream(stream proto.SignalExchange_ConnectStreamServer)
 				s.metrics.MessagesForwarded.Add(stream.Context(), 1)
 			}
 		} else {
+			s.metrics.GetRegistrationDelay.Record(stream.Context(), float64(time.Since(getRegistrationStart).Nanoseconds())/1e6, metric.WithAttributes(attribute.String(labelType, labelTypeStream), attribute.String(labelRegistrionStatus, labelRegistrationNotFound)))
+			s.metrics.MessageForwardFailures.Add(stream.Context(), 1, metric.WithAttributes(attribute.String(labelType, labelTypeNotConnected)))
 			log.Debugf("message from peer [%s] can't be forwarded to peer [%s] because destination peer is not connected", p.Id, msg.RemoteKey)
 			//todo respond to the sender?
-
-			s.metrics.MessageForwardFailures.Add(stream.Context(), 1, metric.WithAttributes(attribute.String(labelType, labelTypeNotConnected)))
 		}
 	}
 	<-stream.Context().Done()