adding metrics

docker/telemetry/otel-collector/config.yml

@@ -15,18 +15,10 @@ exporters:
     endpoint: "otel-collector:8889"
     send_timestamps: true
     metric_expiration: 1800m
-  jaeger:
-    endpoint: "jaeger:14250"
-    tls:
-      insecure: true
 
 service:
   extensions: [health_check]
   pipelines:
     metrics:
       receivers: [otlp]
       exporters: [prometheus]
-    traces:
-      receivers: [otlp]
-      processors: []
-      exporters: [jaeger]

nodebuilder/pruner/config.go

@@ -15,3 +15,10 @@ func DefaultConfig() Config {
 		RecencyWindow:  time.Hour * 24 * 30,
 	}
 }
+
+// WithStoragePrunerMetrics is a utility function to turn on storage pruner metrics and that is
+// expected to be "invoked" by the fx lifecycle.
+func WithStoragePrunerMetrics(sp *StoragePruner) error {
+	return sp.WithMetrics()
+}
+

nodebuilder/pruner/metrics.go

@@ -0,0 +1,75 @@
+package pruner
+
+import (
+	"context"
+
+	"go.opentelemetry.io/otel"
+	"go.opentelemetry.io/otel/metric"
+)
+
+var (
+	meter = otel.Meter("storage_pruner")
+)
+
+type metrics struct {
+	registeredCounter metric.Int64Counter
+	prunedCounter     metric.Int64Counter
+
+	epochAmount metric.Int64ObservableGauge
+}
+
+func (sp *StoragePruner) WithMetrics() error {
+	registeredCounter, err := meter.Int64Counter("pruner_registered_counter",
+		metric.WithDescription("pruner registered datahash counter"))
+	if err != nil {
+		return err
+	}
+
+	prunedCounter, err := meter.Int64Counter("pruner_pruned_counter",
+		metric.WithDescription("pruner pruned datahash counter"))
+	if err != nil {
+		return err
+	}
+
+	epochAmount, err := meter.Int64ObservableGauge("pruner_epoch_amount",
+		metric.WithDescription("pruner epoch amount"))
+	if err != nil {
+		return err
+	}
+
+	callback := func(ctx context.Context, observer metric.Observer) error {
+		observer.ObserveInt64(epochAmount, int64(len(sp.activeEpochs)))
+		return nil
+	}
+
+	if _, err := meter.RegisterCallback(callback, epochAmount); err != nil {
+		return err
+	}
+
+	sp.metrics = &metrics{
+		registeredCounter: registeredCounter,
+		prunedCounter:     prunedCounter,
+		epochAmount:       epochAmount,
+	}
+	return nil
+}
+
+func (m *metrics) observeRegister(ctx context.Context) {
+	if m == nil {
+		return
+	}
+	if ctx.Err() != nil {
+		ctx = context.Background()
+	}
+	m.registeredCounter.Add(context.Background(), 1)
+}
+
+func (m *metrics) observePrune(ctx context.Context) {
+	if m == nil {
+		return
+	}
+	if ctx.Err() != nil {
+		ctx = context.Background()
+	}
+	m.prunedCounter.Add(context.Background(), 1)
+}

nodebuilder/pruner/pruner.go

@@ -37,6 +37,8 @@ type StoragePruner struct {
 	ds    datastore.Batching
 	store *eds.Store
 
+	metrics *metrics
+
 	done chan struct{}
 }
 
@@ -123,6 +125,7 @@ func (sp *StoragePruner) Register(ctx context.Context, h *header.ExtendedHeader)
 	}
 
 	datahashes = append(datahashes, h.DAH.Hash())
+	sp.metrics.observeRegister(ctx)
 	return sp.saveDatahashesToEpoch(ctx, epoch, datahashes)
 }
 
@@ -183,6 +186,7 @@ func (sp *StoragePruner) pruneEpoch(ctx context.Context, epoch uint64) error {
 		if err != nil {
 			return err
 		}
+		sp.metrics.observePrune(ctx)
 	}
 
 	delete(sp.activeEpochs, epoch)

nodebuilder/settings.go

@@ -27,6 +27,7 @@ import (
 	modheader "github.com/celestiaorg/celestia-node/nodebuilder/header"
 	"github.com/celestiaorg/celestia-node/nodebuilder/node"
 	"github.com/celestiaorg/celestia-node/nodebuilder/p2p"
+	"github.com/celestiaorg/celestia-node/nodebuilder/pruner"
 	"github.com/celestiaorg/celestia-node/nodebuilder/share"
 	"github.com/celestiaorg/celestia-node/state"
 )
@@ -101,6 +102,7 @@ func WithMetrics(metricOpts []otlpmetrichttp.Option, nodeType node.Type) fx.Opti
 			baseComponents,
 			fx.Invoke(share.WithStoreMetrics),
 			fx.Invoke(share.WithShrexServerMetrics),
+			fx.Invoke(pruner.WithStoragePrunerMetrics),
 			samplingMetrics,
 		)
 	case node.Light:
@@ -113,6 +115,7 @@ func WithMetrics(metricOpts []otlpmetrichttp.Option, nodeType node.Type) fx.Opti
 			baseComponents,
 			fx.Invoke(share.WithStoreMetrics),
 			fx.Invoke(share.WithShrexServerMetrics),
+			fx.Invoke(pruner.WithStoragePrunerMetrics),
 		)
 	default:
 		panic("invalid node type")