Merge pull request #6911 from ethereum-optimism/jg/challenger_metrics

op-challenger: Add basic metrics & pprof setup

op-batcher/batcher/batch_submitter.go

@@ -68,7 +68,7 @@ func Main(version string, cliCtx *cli.Context) error {
 		l.Info("starting metrics server", "addr", metricsCfg.ListenAddr, "port", metricsCfg.ListenPort)
 		go func() {
 			if err := m.Serve(ctx, metricsCfg.ListenAddr, metricsCfg.ListenPort); err != nil {
-				l.Error("error starting metrics server", err)
+				l.Error("error starting metrics server", "err", err)
 			}
 		}()
 		m.StartBalanceMetrics(ctx, l, batchSubmitter.L1Client, batchSubmitter.TxManager.From())

op-challenger/config/config.go

@@ -5,6 +5,8 @@ import (
 	"fmt"
 
 	"github.com/ethereum-optimism/optimism/op-node/chaincfg"
+	opmetrics "github.com/ethereum-optimism/optimism/op-service/metrics"
+	oppprof "github.com/ethereum-optimism/optimism/op-service/pprof"
 	"github.com/ethereum-optimism/optimism/op-service/txmgr"
 	"github.com/ethereum/go-ethereum/common"
 )
@@ -105,7 +107,9 @@ type Config struct {
 	CannonL2               string // L2 RPC Url
 	CannonSnapshotFreq     uint   // Frequency of snapshots to create when executing cannon (in VM instructions)
 
-	TxMgrConfig txmgr.CLIConfig
+	TxMgrConfig   txmgr.CLIConfig
+	MetricsConfig opmetrics.CLIConfig
+	PprofConfig   oppprof.CLIConfig
 }
 
 func NewConfig(
@@ -122,7 +126,9 @@ func NewConfig(
 
 		TraceType: traceType,
 
-		TxMgrConfig: txmgr.NewCLIConfig(l1EthRpc),
+		TxMgrConfig:   txmgr.NewCLIConfig(l1EthRpc),
+		MetricsConfig: opmetrics.DefaultCLIConfig(),
+		PprofConfig:   oppprof.DefaultCLIConfig(),
 
 		CannonSnapshotFreq: DefaultCannonSnapshotFreq,
 	}
@@ -182,5 +188,11 @@ func (c Config) Check() error {
 	if err := c.TxMgrConfig.Check(); err != nil {
 		return err
 	}
+	if err := c.MetricsConfig.Check(); err != nil {
+		return err
+	}
+	if err := c.PprofConfig.Check(); err != nil {
+		return err
+	}
 	return nil
 }

op-challenger/fault/service.go

@@ -8,9 +8,11 @@ import (
 	"github.com/ethereum-optimism/optimism/op-bindings/bindings"
 	"github.com/ethereum-optimism/optimism/op-challenger/config"
 	"github.com/ethereum-optimism/optimism/op-challenger/fault/types"
+	"github.com/ethereum-optimism/optimism/op-challenger/metrics"
+	"github.com/ethereum-optimism/optimism/op-challenger/version"
 	"github.com/ethereum-optimism/optimism/op-service/client"
+	oppprof "github.com/ethereum-optimism/optimism/op-service/pprof"
 	"github.com/ethereum-optimism/optimism/op-service/txmgr"
-	"github.com/ethereum-optimism/optimism/op-service/txmgr/metrics"
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/crypto"
 	"github.com/ethereum/go-ethereum/log"
@@ -25,12 +27,14 @@ type Service interface {
 
 type service struct {
 	logger  log.Logger
+	metrics metrics.Metricer
 	monitor *gameMonitor
 }
 
 // NewService creates a new Service.
 func NewService(ctx context.Context, logger log.Logger, cfg *config.Config) (*service, error) {
-	txMgr, err := txmgr.NewSimpleTxManager("challenger", logger, &metrics.NoopTxMetrics{}, cfg.TxMgrConfig)
+	m := metrics.NewMetrics()
+	txMgr, err := txmgr.NewSimpleTxManager("challenger", logger, &m.TxMetrics, cfg.TxMgrConfig)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create the transaction manager: %w", err)
 	}
@@ -40,6 +44,27 @@ func NewService(ctx context.Context, logger log.Logger, cfg *config.Config) (*se
 		return nil, fmt.Errorf("failed to dial L1: %w", err)
 	}
 
+	pprofConfig := cfg.PprofConfig
+	if pprofConfig.Enabled {
+		logger.Info("starting pprof", "addr", pprofConfig.ListenAddr, "port", pprofConfig.ListenPort)
+		go func() {
+			if err := oppprof.ListenAndServe(ctx, pprofConfig.ListenAddr, pprofConfig.ListenPort); err != nil {
+				logger.Error("error starting pprof", "err", err)
+			}
+		}()
+	}
+
+	metricsCfg := cfg.MetricsConfig
+	if metricsCfg.Enabled {
+		logger.Info("starting metrics server", "addr", metricsCfg.ListenAddr, "port", metricsCfg.ListenPort)
+		go func() {
+			if err := m.Serve(ctx, metricsCfg.ListenAddr, metricsCfg.ListenPort); err != nil {
+				logger.Error("error starting metrics server", "err", err)
+			}
+		}()
+		m.StartBalanceMetrics(ctx, logger, client, txMgr.From())
+	}
+
 	factory, err := bindings.NewDisputeGameFactory(cfg.GameFactoryAddress, client)
 	if err != nil {
 		return nil, fmt.Errorf("failed to bind the fault dispute game factory contract: %w", err)
@@ -49,9 +74,14 @@ func NewService(ctx context.Context, logger log.Logger, cfg *config.Config) (*se
 	monitor := newGameMonitor(logger, client.BlockNumber, cfg.GameAddress, loader, func(addr common.Address) (gamePlayer, error) {
 		return NewGamePlayer(ctx, logger, cfg, addr, txMgr, client)
 	})
+
+	m.RecordInfo(version.SimpleWithMeta)
+	m.RecordUp()
+
 	return &service{
-		monitor: monitor,
 		logger:  logger,
+		metrics: m,
+		monitor: monitor,
 	}, nil
 }
 

op-challenger/flags/flags.go

@@ -9,6 +9,8 @@ import (
 	opservice "github.com/ethereum-optimism/optimism/op-service"
 	openum "github.com/ethereum-optimism/optimism/op-service/enum"
 	oplog "github.com/ethereum-optimism/optimism/op-service/log"
+	opmetrics "github.com/ethereum-optimism/optimism/op-service/metrics"
+	oppprof "github.com/ethereum-optimism/optimism/op-service/pprof"
 	"github.com/ethereum-optimism/optimism/op-service/txmgr"
 	"github.com/ethereum/go-ethereum/common"
 
@@ -134,6 +136,8 @@ var optionalFlags = []cli.Flag{
 func init() {
 	optionalFlags = append(optionalFlags, oplog.CLIFlags(envVarPrefix)...)
 	optionalFlags = append(optionalFlags, txmgr.CLIFlags(envVarPrefix)...)
+	optionalFlags = append(optionalFlags, opmetrics.CLIFlags(envVarPrefix)...)
+	optionalFlags = append(optionalFlags, oppprof.CLIFlags(envVarPrefix)...)
 
 	Flags = append(requiredFlags, optionalFlags...)
 }
@@ -201,6 +205,8 @@ func NewConfigFromCLI(ctx *cli.Context) (*config.Config, error) {
 	}
 
 	txMgrConfig := txmgr.ReadCLIConfig(ctx)
+	metricsConfig := opmetrics.ReadCLIConfig(ctx)
+	pprofConfig := oppprof.ReadCLIConfig(ctx)
 
 	traceTypeFlag := config.TraceType(strings.ToLower(ctx.String(TraceTypeFlag.Name)))
 
@@ -222,5 +228,7 @@ func NewConfigFromCLI(ctx *cli.Context) (*config.Config, error) {
 		CannonSnapshotFreq:      ctx.Uint(CannonSnapshotFreqFlag.Name),
 		AgreeWithProposedOutput: ctx.Bool(AgreeWithProposedOutputFlag.Name),
 		TxMgrConfig:             txMgrConfig,
+		MetricsConfig:           metricsConfig,
+		PprofConfig:             pprofConfig,
 	}, nil
 }

op-challenger/metrics/metrics.go

+package metrics
+
+import (
+	"context"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/ethclient"
+	"github.com/ethereum/go-ethereum/log"
+	"github.com/prometheus/client_golang/prometheus"
+
+	opmetrics "github.com/ethereum-optimism/optimism/op-service/metrics"
+	txmetrics "github.com/ethereum-optimism/optimism/op-service/txmgr/metrics"
+)
+
+const Namespace = "op_challenger"
+
+type Metricer interface {
+	RecordInfo(version string)
+	RecordUp()
+
+	// Record Tx metrics
+	txmetrics.TxMetricer
+}
+
+type Metrics struct {
+	ns       string
+	registry *prometheus.Registry
+	factory  opmetrics.Factory
+
+	txmetrics.TxMetrics
+
+	info prometheus.GaugeVec
+	up   prometheus.Gauge
+}
+
+var _ Metricer = (*Metrics)(nil)
+
+func NewMetrics() *Metrics {
+	registry := opmetrics.NewRegistry()
+	factory := opmetrics.With(registry)
+
+	return &Metrics{
+		ns:       Namespace,
+		registry: registry,
+		factory:  factory,
+
+		TxMetrics: txmetrics.MakeTxMetrics(Namespace, factory),
+
+		info: *factory.NewGaugeVec(prometheus.GaugeOpts{
+			Namespace: Namespace,
+			Name:      "info",
+			Help:      "Pseudo-metric tracking version and config info",
+		}, []string{
+			"version",
+		}),
+		up: factory.NewGauge(prometheus.GaugeOpts{
+			Namespace: Namespace,
+			Name:      "up",
+			Help:      "1 if the op-challenger has finished starting up",
+		}),
+	}
+}
+
+func (m *Metrics) Serve(ctx context.Context, host string, port int) error {
+	return opmetrics.ListenAndServe(ctx, m.registry, host, port)
+}
+
+func (m *Metrics) StartBalanceMetrics(ctx context.Context, l log.Logger, client *ethclient.Client, account common.Address) {
+	opmetrics.LaunchBalanceMetrics(ctx, l, m.registry, m.ns, client, account)
+}
+
+// RecordInfo sets a pseudo-metric that contains versioning and
+// config info for the op-proposer.
+func (m *Metrics) RecordInfo(version string) {
+	m.info.WithLabelValues(version).Set(1)
+}
+
+// RecordUp sets the up metric to 1.
+func (m *Metrics) RecordUp() {
+	prometheus.MustRegister()
+	m.up.Set(1)
+}
+
+func (m *Metrics) Document() []opmetrics.DocumentedMetric {
+	return m.factory.Document()
+}

op-challenger/metrics/noop.go

+package metrics
+
+import (
+	txmetrics "github.com/ethereum-optimism/optimism/op-service/txmgr/metrics"
+)
+
+type noopMetrics struct {
+	txmetrics.NoopTxMetrics
+}
+
+var NoopMetrics Metricer = new(noopMetrics)
+
+func (*noopMetrics) RecordInfo(version string) {}
+func (*noopMetrics) RecordUp()                 {}

op-challenger/scripts/alphabet/charlie.sh

@@ -27,4 +27,6 @@ $CHALLENGER_DIR/bin/op-challenger \
   --game-address $FAULT_GAME_ADDRESS \
   --private-key $CHARLIE_KEY \
   --num-confirmations 1 \
+  --metrics.enabled --metrics.port=7304 \
+  --pprof.enabled --pprof.port=6064 \
   --agree-with-proposed-output=true

op-challenger/scripts/alphabet/mallory.sh

@@ -27,4 +27,6 @@ $CHALLENGER_DIR/bin/op-challenger \
   --game-address $FAULT_GAME_ADDRESS \
   --private-key $MALLORY_KEY \
   --num-confirmations 1 \
+  --metrics.enabled --metrics.port=7305 \
+  --pprof.enabled --pprof.port=6065 \
   --agree-with-proposed-output=false

op-challenger/version/version.go

@@ -4,3 +4,11 @@ var (
 	Version = "v0.1.0"
 	Meta    = "dev"
 )
+
+var SimpleWithMeta = func() string {
+	v := Version
+	if Meta != "" {
+		v += "-" + Meta
+	}
+	return v
+}()

op-proposer/proposer/l2_output_submitter.go

@@ -85,7 +85,7 @@ func Main(version string, cliCtx *cli.Context) error {
 		l.Info("starting metrics server", "addr", metricsCfg.ListenAddr, "port", metricsCfg.ListenPort)
 		go func() {
 			if err := m.Serve(ctx, metricsCfg.ListenAddr, metricsCfg.ListenPort); err != nil {
-				l.Error("error starting metrics server", err)
+				l.Error("error starting metrics server", "err", err)
 			}
 		}()
 		m.StartBalanceMetrics(ctx, l, proposerConfig.L1Client, proposerConfig.TxManager.From())

op-service/metrics/cli.go

@@ -13,8 +13,18 @@ const (
 	EnabledFlagName    = "metrics.enabled"
 	ListenAddrFlagName = "metrics.addr"
 	PortFlagName       = "metrics.port"
+	defaultListenAddr  = "0.0.0.0"
+	defaultListenPort  = 7300
 )
 
+func DefaultCLIConfig() CLIConfig {
+	return CLIConfig{
+		Enabled:    false,
+		ListenAddr: defaultListenAddr,
+		ListenPort: defaultListenPort,
+	}
+}
+
 func CLIFlags(envPrefix string) []cli.Flag {
 	return []cli.Flag{
 		&cli.BoolFlag{
@@ -25,13 +35,13 @@ func CLIFlags(envPrefix string) []cli.Flag {
 		&cli.StringFlag{
 			Name:    ListenAddrFlagName,
 			Usage:   "Metrics listening address",
-			Value:   "0.0.0.0", // TODO(CLI-4159): Switch to 127.0.0.1
+			Value:   defaultListenAddr, // TODO(CLI-4159): Switch to 127.0.0.1
 			EnvVars: opservice.PrefixEnvVar(envPrefix, "METRICS_ADDR"),
 		},
 		&cli.IntFlag{
 			Name:    PortFlagName,
 			Usage:   "Metrics listening port",
-			Value:   7300,
+			Value:   defaultListenPort,
 			EnvVars: opservice.PrefixEnvVar(envPrefix, "METRICS_PORT"),
 		},
 	}

op-service/pprof/cli.go

@@ -12,8 +12,18 @@ const (
 	EnabledFlagName    = "pprof.enabled"
 	ListenAddrFlagName = "pprof.addr"
 	PortFlagName       = "pprof.port"
+	defaultListenAddr  = "0.0.0.0"
+	defaultListenPort  = 6060
 )
 
+func DefaultCLIConfig() CLIConfig {
+	return CLIConfig{
+		Enabled:    false,
+		ListenAddr: defaultListenAddr,
+		ListenPort: defaultListenPort,
+	}
+}
+
 func CLIFlags(envPrefix string) []cli.Flag {
 	return []cli.Flag{
 		&cli.BoolFlag{
@@ -24,13 +34,13 @@ func CLIFlags(envPrefix string) []cli.Flag {
 		&cli.StringFlag{
 			Name:    ListenAddrFlagName,
 			Usage:   "pprof listening address",
-			Value:   "0.0.0.0", // TODO(CLI-4159): Switch to 127.0.0.1
+			Value:   defaultListenAddr, // TODO(CLI-4159): Switch to 127.0.0.1
 			EnvVars: opservice.PrefixEnvVar(envPrefix, "PPROF_ADDR"),
 		},
 		&cli.IntFlag{
 			Name:    PortFlagName,
 			Usage:   "pprof listening port",
-			Value:   6060,
+			Value:   defaultListenPort,
 			EnvVars: opservice.PrefixEnvVar(envPrefix, "PPROF_PORT"),
 		},
 	}