multi: add support for simple payments related metrics

monitoring/metrics.go

@@ -0,0 +1,76 @@
+//go:build monitoring
+// +build monitoring
+
+package monitoring
+
+import (
+	"sync"
+
+	"github.com/lightningnetwork/lnd/lncfg"
+)
+
+// MetricGroupCreator is a factory function that initializes a metric group.
+type MetricGroupCreator func(cfg *lncfg.Prometheus) (MetricGroup, error)
+
+// Global registry for all metric groups.
+var (
+	// metricGroups is a global variable of all registered metrics
+	// protected by the mutex below. All new MetricGroups should add
+	// themselves to this map within the init() method of their file.
+	metricGroups = make(map[string]MetricGroupCreator)
+
+	// activeGroups is a global map of all active metric groups. This can
+	// be used by some of the "static' package level methods to look up the
+	// target metric group to export observations.
+	activeMetrics = make(map[string]MetricGroup)
+
+	metricsMtx sync.Mutex
+)
+
+// MetricGroup is the primary interface for metric groups.
+type MetricGroup interface {
+	// Name returns the name of the metric group.
+	Name() string
+
+	// RegisterMetrics registers all metrics within the group.
+	RegisterMetrics() error
+
+	// ShouldRegister indicates whether this groups metrics should actually
+	// be registered.
+	ShouldRegister(cfg *lncfg.Prometheus) bool
+}
+
+// RegisterMetricGroup adds a new metric group to the registry.
+func RegisterMetricGroup(name string, creator MetricGroupCreator) {
+	metricsMtx.Lock()
+	defer metricsMtx.Unlock()
+	metricGroups[name] = creator
+}
+
+// InitializeMetrics initializes and registers all active metric groups.
+func InitializeMetrics(cfg *lncfg.Prometheus) error {
+	metricsMtx.Lock()
+	defer metricsMtx.Unlock()
+
+	for name, creator := range metricGroups {
+		// We'll pass the configuration struct to permit conditional
+		// metric registration in the future.
+		group, err := creator(cfg)
+		if err != nil {
+			return err
+		}
+
+		// Check whether this metric group should be registered.
+		if !group.ShouldRegister(cfg) {
+			continue
+		}
+
+		if err := group.RegisterMetrics(); err != nil {
+			return err
+		}
+
+		activeMetrics[name] = group
+	}
+
+	return nil
+}

monitoring/monitoring_off.go

@@ -23,3 +23,13 @@ func ExportPrometheusMetrics(_ *grpc.Server, _ lncfg.Prometheus) error {
 	return fmt.Errorf("lnd must be built with the monitoring tag to " +
 		"enable exporting Prometheus metrics")
 }
+
+// IncrementPaymentCount increments a counter tracking the number of payments
+// made by lnd when monitoring is enabled. This method no-ops as monitoring is
+// disabled.
+func IncrementPaymentCount() {}
+
+// IncrementHTLCAttemptCount increments a counter tracking the number of HTLC
+// attempts made by lnd when monitoring is enabled. This method no-ops as
+// monitoring is disabled.
+func IncrementHTLCAttemptCount() {}

monitoring/monitoring_on.go

@@ -27,12 +27,14 @@ func GetPromInterceptors() ([]grpc.UnaryServerInterceptor, []grpc.StreamServerIn
 	return unaryInterceptors, streamInterceptors
 }
 
-// ExportPrometheusMetrics sets server options, registers gRPC metrics and
-// launches the Prometheus exporter on the specified address.
+// ExportPrometheusMetrics sets server options, registers gRPC metrics,
+// and launches the Prometheus exporter on the specified address.
 func ExportPrometheusMetrics(grpcServer *grpc.Server, cfg lncfg.Prometheus) error {
+	var metricErr error
 	started.Do(func() {
 		log.Infof("Prometheus exporter started on %v/metrics", cfg.Listen)
 
+		// Register gRPC Prometheus interceptors.
 		grpc_prometheus.Register(grpcServer)
 
 		// Enable the histograms which can allow plotting latency
@@ -43,11 +45,21 @@ func ExportPrometheusMetrics(grpcServer *grpc.Server, cfg lncfg.Prometheus) erro
 			grpc_prometheus.EnableHandlingTimeHistogram()
 		}
 
+		// Initialize additional metric groups (e.g., payment tracking).
+		err := InitializeMetrics(&cfg)
+		if err != nil {
+			log.Warnf("Failed to initialize additional metrics: %v",
+				err)
+
+			metricErr = err
+		}
+
+		// Start the Prometheus HTTP handler.
 		http.Handle("/metrics", promhttp.Handler())
 		go func() {
 			http.ListenAndServe(cfg.Listen, nil)
 		}()
 	})
 
-	return nil
+	return metricErr
 }

monitoring/payments_metrics.go

@@ -0,0 +1,79 @@
+//go:build monitoring
+// +build monitoring
+
+package monitoring
+
+import (
+	"github.com/lightningnetwork/lnd/lncfg"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+const paymentsMetricGroupName string = "payments"
+
+// paymentMetrics tracks payment-related Prometheus metrics.
+type paymentMetrics struct {
+	totalPayments     prometheus.Counter
+	totalHTLCAttempts prometheus.Counter
+}
+
+// NewPaymentMetrics creates a new instance of payment metrics.
+func NewPaymentMetrics(cfg *lncfg.Prometheus) (MetricGroup, error) {
+	return &paymentMetrics{
+		totalPayments: prometheus.NewCounter(prometheus.CounterOpts{
+			Name: "lnd_total_payments",
+			Help: "Total number of payments initiated",
+		}),
+		totalHTLCAttempts: prometheus.NewCounter(prometheus.CounterOpts{
+			Name: "lnd_total_htlc_attempts",
+			Help: "Total number of HTLC attempts",
+		}),
+	}, nil
+}
+
+// Name returns the metric group name.
+func (p *paymentMetrics) Name() string {
+	return paymentsMetricGroupName
+}
+
+// RegisterMetrics registers payment-related Prometheus metrics.
+func (p *paymentMetrics) RegisterMetrics() error {
+	err := prometheus.Register(p.totalPayments)
+	if err != nil {
+		return err
+	}
+	return prometheus.Register(p.totalHTLCAttempts)
+}
+
+// ShouldRegister indicates whether the payments related metrics should be
+// registered with prometheus.
+func (p *paymentMetrics) ShouldRegister(cfg *lncfg.Prometheus) bool {
+	// TODO: Can condition this on application config.
+	return true
+}
+
+// IncrementPaymentCount increments the counter tracking the number of payments
+// made by lnd when monitoring is enabled and the metric is configured
+func IncrementPaymentCount() {
+	group, ok := activeMetrics[paymentsMetricGroupName].(*paymentMetrics)
+	if !ok {
+		return
+	}
+
+	group.totalPayments.Inc()
+}
+
+// IncrementHTLCAttemptCount increments the counter tracking the number of HTLC
+// attempts made by lnd when monitoring is enabled and the metric is configured.
+func IncrementHTLCAttemptCount() {
+	group, ok := activeMetrics[paymentsMetricGroupName].(*paymentMetrics)
+	if !ok {
+		return
+	}
+
+	group.totalHTLCAttempts.Inc()
+}
+
+// Register the payments metric group.
+func init() {
+	RegisterMetricGroup(paymentsMetricGroupName, NewPaymentMetrics)
+}

routing/control_tower.go

@@ -5,6 +5,7 @@ import (
 
 	"github.com/lightningnetwork/lnd/channeldb"
 	"github.com/lightningnetwork/lnd/lntypes"
+	"github.com/lightningnetwork/lnd/monitoring"
 	"github.com/lightningnetwork/lnd/multimutex"
 	"github.com/lightningnetwork/lnd/queue"
 )
@@ -191,6 +192,9 @@ func (p *controlTower) InitPayment(paymentHash lntypes.Hash,
 		return err
 	}
 
+	// Observe the creation of a new payment.
+	monitoring.IncrementPaymentCount()
+
 	// Take lock before querying the db to prevent missing or duplicating
 	// an update.
 	p.paymentsMtx.Lock(paymentHash)

routing/payment_lifecycle.go

@@ -15,6 +15,7 @@ import (
 	"github.com/lightningnetwork/lnd/htlcswitch"
 	"github.com/lightningnetwork/lnd/lntypes"
 	"github.com/lightningnetwork/lnd/lnwire"
+	"github.com/lightningnetwork/lnd/monitoring"
 	"github.com/lightningnetwork/lnd/routing/route"
 	"github.com/lightningnetwork/lnd/routing/shards"
 	"github.com/lightningnetwork/lnd/tlv"
@@ -596,6 +597,9 @@ func (p *paymentLifecycle) registerAttempt(rt *route.Route,
 		p.identifier, &attempt.HTLCAttemptInfo,
 	)
 
+	// Observe a new HTLC attempt for the payment.
+	monitoring.IncrementHTLCAttemptCount()
+
 	return attempt, err
 }