fix(op-node): Peer Scoring Metrics

op-node/flags/p2p_flags.go

@@ -34,6 +34,15 @@ var (
 		Value:    "none",
 		EnvVar:   p2pEnv("PEER_SCORING"),
 	}
+	PeerScoreBands = cli.StringFlag{
+		Name: "p2p.score.bands",
+		Usage: "Sets the peer score bands used primarily for peer score metrics. " +
+			"Should be provided in following format: <threshold>:<label>;<threshold>:<label>;..." +
+			"For example: -40:graylist;-20:restricted;0:nopx;20:friend;",
+		Required: false,
+		Value:    "-40:graylist;-20:restricted;0:nopx;20:friend;",
+		EnvVar:   p2pEnv("SCORE_BANDS"),
+	}
 
 	// Banning Flag - whether or not we want to act on the scoring
 	Banning = cli.BoolFlag{
@@ -276,6 +285,10 @@ var p2pFlags = []cli.Flag{
 	NoDiscovery,
 	P2PPrivPath,
 	P2PPrivRaw,
+	PeerScoring,
+	PeerScoreBands,
+	Banning,
+	TopicScoring,
 	ListenIP,
 	ListenTCPPort,
 	ListenUDPPort,

op-node/metrics/metrics.go

@@ -15,7 +15,6 @@ import (
 
 	pb "github.com/libp2p/go-libp2p-pubsub/pb"
 	libp2pmetrics "github.com/libp2p/go-libp2p/core/metrics"
-	"github.com/libp2p/go-libp2p/core/peer"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/collectors"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
@@ -66,7 +65,7 @@ type Metricer interface {
 	RecordSequencerSealingTime(duration time.Duration)
 	Document() []metrics.DocumentedMetric
 	// P2P Metrics
-	RecordPeerScoring(peerID peer.ID, score float64)
+	SetPeerScores(scores map[string]float64)
 }
 
 // Metrics tracks all the metrics for the op-node.
@@ -287,21 +286,24 @@ func NewMetrics(procName string) *Metrics {
 			Name:      "peer_count",
 			Help:      "Count of currently connected p2p peers",
 		}),
-		StreamCount: factory.NewGauge(prometheus.GaugeOpts{
-			Namespace: ns,
-			Subsystem: "p2p",
-			Name:      "stream_count",
-			Help:      "Count of currently connected p2p streams",
-		}),
+		// Notice: We cannot use peer ids as [Labels] in the GaugeVec
+		// since peer ids would open a service attack vector.
+		// Each peer id would be a separate metric, flooding prometheus.
+		//
+		// [Labels]: https://prometheus.io/docs/practices/naming/#labels
 		PeerScores: factory.NewGaugeVec(prometheus.GaugeOpts{
 			Namespace: ns,
 			Subsystem: "p2p",
 			Name:      "peer_scores",
-			Help:      "Peer scoring",
+			Help:      "Count of peer scores grouped by score",
 		}, []string{
-			// No label names here since peer ids would open a service attack vector.
-			// Each peer id would be a separate metric, flooding prometheus.
-			// See: https://prometheus.io/docs/practices/naming/#labels
+			"band",
+		}),
+		StreamCount: factory.NewGauge(prometheus.GaugeOpts{
+			Namespace: ns,
+			Subsystem: "p2p",
+			Name:      "stream_count",
+			Help:      "Count of currently connected p2p streams",
 		}),
 		GossipEventsTotal: factory.NewCounterVec(prometheus.CounterOpts{
 			Namespace: ns,
@@ -350,6 +352,14 @@ func NewMetrics(procName string) *Metrics {
 	}
 }
 
+// SetPeerScores updates the peer score [prometheus.GaugeVec].
+// This takes a map of labels to scores.
+func (m *Metrics) SetPeerScores(scores map[string]float64) {
+	for label, score := range scores {
+		m.PeerScores.WithLabelValues(label).Set(score)
+	}
+}
+
 // RecordInfo sets a pseudo-metric that contains versioning and
 // config info for the opnode.
 func (m *Metrics) RecordInfo(version string) {
@@ -491,10 +501,6 @@ func (m *Metrics) RecordGossipEvent(evType int32) {
 	m.GossipEventsTotal.WithLabelValues(pb.TraceEvent_Type_name[evType]).Inc()
 }
 
-func (m *Metrics) RecordPeerScoring(peerID peer.ID, score float64) {
-	m.PeerScores.WithLabelValues(peerID.String()).Set(score)
-}
-
 func (m *Metrics) IncPeerCount() {
 	m.PeerCount.Inc()
 }
@@ -627,7 +633,7 @@ func (n *noopMetricer) RecordSequencerReset() {
 func (n *noopMetricer) RecordGossipEvent(evType int32) {
 }
 
-func (n *noopMetricer) RecordPeerScoring(peerID peer.ID, score float64) {
+func (n *noopMetricer) SetPeerScores(scores map[string]float64) {
 }
 
 func (n *noopMetricer) IncPeerCount() {

op-node/p2p/band_scorer_test.go

@@ -0,0 +1,83 @@
+package p2p
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+// TestBandScorer_ParseDefault tests the [BandScorer.Parse] function
+// on the default band scores cli flag value.
+func TestBandScorer_ParseDefault(t *testing.T) {
+	// Create a new band scorer.
+	bandScorer, err := NewBandScorer("-40:graylist;-20:restricted;0:nopx;20:friend;")
+	require.NoError(t, err)
+
+	// Validate the [BandScorer] internals.
+	require.ElementsMatch(t, bandScorer.bands, []scorePair{
+		{band: "graylist", threshold: -40},
+		{band: "restricted", threshold: -20},
+		{band: "nopx", threshold: 0},
+		{band: "friend", threshold: 20},
+	})
+}
+
+// TestBandScorer_BucketCorrectly tests the [BandScorer.Bucket] function
+// on a variety of scores.
+func TestBandScorer_BucketCorrectly(t *testing.T) {
+	// Create a new band scorer.
+	bandScorer, err := NewBandScorer("-40:graylist;-20:restricted;0:nopx;20:friend;")
+	require.NoError(t, err)
+
+	// Validate the [BandScorer] internals.
+	require.Equal(t, bandScorer.Bucket(-100), "graylist")
+	require.Equal(t, bandScorer.Bucket(-40), "graylist")
+	require.Equal(t, bandScorer.Bucket(-39), "restricted")
+	require.Equal(t, bandScorer.Bucket(-20), "restricted")
+	require.Equal(t, bandScorer.Bucket(-19), "nopx")
+	require.Equal(t, bandScorer.Bucket(0), "nopx")
+	require.Equal(t, bandScorer.Bucket(1), "friend")
+	require.Equal(t, bandScorer.Bucket(20), "friend")
+	require.Equal(t, bandScorer.Bucket(21), "friend")
+}
+
+// TestBandScorer_BucketInverted tests the [BandScorer.Bucket] function
+// on a variety of scores, in descending order.
+func TestBandScorer_BucketInverted(t *testing.T) {
+	// Create a new band scorer.
+	bandScorer, err := NewBandScorer("20:friend;0:nopx;-20:restricted;-40:graylist;")
+	require.NoError(t, err)
+
+	// Validate the [BandScorer] internals.
+	require.Equal(t, bandScorer.Bucket(-100), "graylist")
+	require.Equal(t, bandScorer.Bucket(-40), "graylist")
+	require.Equal(t, bandScorer.Bucket(-39), "restricted")
+	require.Equal(t, bandScorer.Bucket(-20), "restricted")
+	require.Equal(t, bandScorer.Bucket(-19), "nopx")
+	require.Equal(t, bandScorer.Bucket(0), "nopx")
+	require.Equal(t, bandScorer.Bucket(1), "friend")
+	require.Equal(t, bandScorer.Bucket(20), "friend")
+	require.Equal(t, bandScorer.Bucket(21), "friend")
+}
+
+// TestBandScorer_ParseEmpty tests the [BandScorer.Parse] function
+// on an empty string.
+func TestBandScorer_ParseEmpty(t *testing.T) {
+	// Create a band scorer on an empty string.
+	bandScorer, err := NewBandScorer("")
+	require.NoError(t, err)
+
+	// Validate the [BandScorer] internals.
+	require.Len(t, bandScorer.bands, 0)
+}
+
+// TestBandScorer_ParseWhitespace tests the [BandScorer.Parse] function
+// on a variety of whitespaced strings.
+func TestBandScorer_ParseWhitespace(t *testing.T) {
+	// Create a band scorer on an empty string.
+	bandScorer, err := NewBandScorer("  ;  ;  ;  ")
+	require.NoError(t, err)
+
+	// Validate the [BandScorer] internals.
+	require.Len(t, bandScorer.bands, 0)
+}

op-node/p2p/cli/load_config.go

@@ -58,6 +58,10 @@ func NewConfig(ctx *cli.Context, blockTime uint64) (*p2p.Config, error) {
 		return nil, fmt.Errorf("failed to load p2p peer scoring options: %w", err)
 	}
 
+	if err := loadPeerScoreBands(conf, ctx); err != nil {
+		return nil, fmt.Errorf("failed to load p2p peer score bands: %w", err)
+	}
+
 	if err := loadBanningOption(conf, ctx); err != nil {
 		return nil, fmt.Errorf("failed to load banning option: %w", err)
 	}
@@ -121,6 +125,17 @@ func loadPeerScoringParams(conf *p2p.Config, ctx *cli.Context, blockTime uint64)
 	return nil
 }
 
+// loadPeerScoreBands loads [p2p.BandScorer] from the CLI context.
+func loadPeerScoreBands(conf *p2p.Config, ctx *cli.Context) error {
+	scoreBands := ctx.GlobalString(flags.PeerScoreBands.Name)
+	bandScorer, err := p2p.NewBandScorer(scoreBands)
+	if err != nil {
+		return err
+	}
+	conf.BandScoreThresholds = *bandScorer
+	return nil
+}
+
 // loadBanningOption loads whether or not to ban peers from the CLI context.
 func loadBanningOption(conf *p2p.Config, ctx *cli.Context) error {
 	ban := ctx.GlobalBool(flags.Banning.Name)

op-node/p2p/config.go

@@ -54,6 +54,9 @@ type Config struct {
 	PeerScoring  pubsub.PeerScoreParams
 	TopicScoring pubsub.TopicScoreParams
 
+	// Peer Score Band Thresholds
+	BandScoreThresholds BandScoreThresholds
+
 	// Whether to ban peers based on their [PeerScoring] score.
 	BanningEnabled bool
 
@@ -151,6 +154,10 @@ func (conf *Config) PeerScoringParams() *pubsub.PeerScoreParams {
 	return &conf.PeerScoring
 }
 
+func (conf *Config) PeerBandScorer() *BandScoreThresholds {
+	return &conf.BandScoreThresholds
+}
+
 func (conf *Config) BanPeers() bool {
 	return conf.BanningEnabled
 }

op-node/p2p/gossip.go

@@ -55,6 +55,7 @@ type GossipSetupConfigurables interface {
 	TopicScoringParams() *pubsub.TopicScoreParams
 	BanPeers() bool
 	ConfigureGossip(params *pubsub.GossipSubParams) []pubsub.Option
+	PeerBandScorer() *BandScoreThresholds
 }
 
 type GossipRuntimeConfig interface {
@@ -64,7 +65,8 @@ type GossipRuntimeConfig interface {
 //go:generate mockery --name GossipMetricer
 type GossipMetricer interface {
 	RecordGossipEvent(evType int32)
-	RecordPeerScoring(peerID peer.ID, score float64)
+	// Peer Scoring Metric Funcs
+	SetPeerScores(map[string]float64)
 }
 
 func blocksTopicV1(cfg *rollup.Config) string {