op-node: Add a histogram to report current peer scores

op-node/flags/p2p_flags.go

@@ -35,13 +35,11 @@ var (
 		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;",
+		Name:     "p2p.score.bands",
+		Usage:    "Deprecated. This option is ignored and is only present for backwards compatibility.",
 		Required: false,
-		Value:    "-40:<=-40;-10:<=-10;-5:<=-05;-0.05:<=-00.05;0:<=0;0.05:<=00.05;5:<=05;10:<=10;20:<=20;100:>20;",
-		EnvVar:   p2pEnv("SCORE_BANDS"),
+		Value:    "",
+		Hidden:   true,
 	}
 
 	// Banning Flag - whether or not we want to act on the scoring

op-node/metrics/metrics.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	ophttp "github.com/ethereum-optimism/optimism/op-node/http"
+	"github.com/ethereum-optimism/optimism/op-node/p2p/store"
 	"github.com/ethereum-optimism/optimism/op-service/metrics"
 
 	pb "github.com/libp2p/go-libp2p-pubsub/pb"
@@ -66,7 +67,7 @@ type Metricer interface {
 	Document() []metrics.DocumentedMetric
 	RecordChannelInputBytes(num int)
 	// P2P Metrics
-	SetPeerScores(scores map[string]float64)
+	SetPeerScores(allScores []store.PeerScores)
 	ClientPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration)
 	ServerPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration)
 	PayloadsQuarantineSize(n int)
@@ -134,13 +135,13 @@ type Metrics struct {
 	// P2P Metrics
 	PeerCount         prometheus.Gauge
 	StreamCount       prometheus.Gauge
-	PeerScores        *prometheus.GaugeVec
 	GossipEventsTotal *prometheus.CounterVec
 	BandwidthTotal    *prometheus.GaugeVec
 	PeerUnbans        prometheus.Counter
 	IPUnbans          prometheus.Counter
 	Dials             *prometheus.CounterVec
 	Accepts           *prometheus.CounterVec
+	PeerScores        *prometheus.HistogramVec
 
 	ChannelInputBytes prometheus.Counter
 
@@ -161,6 +162,7 @@ func NewMetrics(procName string) *Metrics {
 	registry.MustRegister(collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}))
 	registry.MustRegister(collectors.NewGoCollector())
 	factory := metrics.With(registry)
+
 	return &Metrics{
 		Info: factory.NewGaugeVec(prometheus.GaugeOpts{
 			Namespace: ns,
@@ -308,19 +310,12 @@ func NewMetrics(procName string) *Metrics {
 			Name:      "peer_count",
 			Help:      "Count of currently connected p2p peers",
 		}),
-		// 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{
+		PeerScores: factory.NewHistogramVec(prometheus.HistogramOpts{
 			Namespace: ns,
-			Subsystem: "p2p",
 			Name:      "peer_scores",
-			Help:      "Count of peer scores grouped by score",
-		}, []string{
-			"band",
-		}),
+			Help:      "Histogram of currrently connected peer scores",
+			Buckets:   []float64{-100, -40, -20, -10, -5, -2, -1, -0.5, -0.05, 0, 0.05, 0.5, 1, 2, 5, 10, 20, 40},
+		}, []string{"type"}),
 		StreamCount: factory.NewGauge(prometheus.GaugeOpts{
 			Namespace: ns,
 			Subsystem: "p2p",
@@ -450,11 +445,17 @@ 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)
+// SetPeerScores updates the peer score metrics.
+// Accepts a slice of peer scores in any order.
+func (m *Metrics) SetPeerScores(allScores []store.PeerScores) {
+	for _, scores := range allScores {
+		m.PeerScores.WithLabelValues("total").Observe(scores.Gossip.Total)
+		m.PeerScores.WithLabelValues("ipColocation").Observe(scores.Gossip.IPColocationFactor)
+		m.PeerScores.WithLabelValues("behavioralPenalty").Observe(scores.Gossip.BehavioralPenalty)
+		m.PeerScores.WithLabelValues("blocksFirstMessage").Observe(scores.Gossip.Blocks.FirstMessageDeliveries)
+		m.PeerScores.WithLabelValues("blocksTimeInMesh").Observe(scores.Gossip.Blocks.TimeInMesh)
+		m.PeerScores.WithLabelValues("blocksMessageDeliveries").Observe(scores.Gossip.Blocks.MeshMessageDeliveries)
+		m.PeerScores.WithLabelValues("blocksInvalidMessageDeliveries").Observe(scores.Gossip.Blocks.InvalidMessageDeliveries)
 	}
 }
 
@@ -785,7 +786,7 @@ func (n *noopMetricer) RecordSequencerReset() {
 func (n *noopMetricer) RecordGossipEvent(evType int32) {
 }
 
-func (n *noopMetricer) SetPeerScores(scores map[string]float64) {
+func (n *noopMetricer) SetPeerScores(allScores []store.PeerScores) {
 }
 
 func (n *noopMetricer) IncPeerCount() {

op-node/p2p/cli/load_config.go

@@ -58,10 +58,6 @@ 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 := loadBanningOptions(conf, ctx); err != nil {
 		return nil, fmt.Errorf("failed to load banning option: %w", err)
 	}
@@ -124,17 +120,6 @@ 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
-}
-
 // loadBanningOptions loads whether or not to ban peers from the CLI context.
 func loadBanningOptions(conf *p2p.Config, ctx *cli.Context) error {
 	conf.BanningEnabled = ctx.GlobalBool(flags.Banning.Name)

op-node/p2p/config.go

@@ -47,7 +47,6 @@ type SetupP2P interface {
 	BanPeers() bool
 	BanThreshold() float64
 	BanDuration() time.Duration
-	PeerBandScorer() *BandScoreThresholds
 	GossipSetupConfigurables
 	ReqRespSyncEnabled() bool
 }
@@ -67,9 +66,6 @@ type Config struct {
 	PeerScoring  pubsub.PeerScoreParams
 	TopicScoring pubsub.TopicScoreParams
 
-	// Peer Score Band Thresholds
-	BandScoreThresholds BandScoreThresholds
-
 	// Whether to ban peers based on their [PeerScoring] score. Should be negative.
 	BanningEnabled bool
 	// Minimum score before peers are disconnected and banned
@@ -142,10 +138,6 @@ 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
 }