feat(op-dispute-mon): Unexpected Claim Resolution (ethereum-optimism#…

…10031)

* feat(op-dispute-mon): track claims resolved against honest actors

* fix(op-dispute-mon): break after honest actor error is logged

* fix(op-dispute-mon): lints

op-dispute-mon/config/config.go

@@ -32,10 +32,11 @@ const (
 type Config struct {
 	L1EthRpc           string         // L1 RPC Url
 	GameFactoryAddress common.Address // Address of the dispute game factory
-	RollupRpc          string         // The rollup node RPC URL.
 
-	MonitorInterval time.Duration // Frequency to check for new games to monitor.
-	GameWindow      time.Duration // Maximum window to look for games to monitor.
+	HonestActors    []common.Address // List of honest actors to monitor claims for.
+	RollupRpc       string           // The rollup node RPC URL.
+	MonitorInterval time.Duration    // Frequency to check for new games to monitor.
+	GameWindow      time.Duration    // Maximum window to look for games to monitor.
 
 	MetricsConfig opmetrics.CLIConfig
 	PprofConfig   oppprof.CLIConfig
@@ -46,6 +47,7 @@ func NewConfig(gameFactoryAddress common.Address, l1EthRpc string) Config {
 		L1EthRpc:           l1EthRpc,
 		GameFactoryAddress: gameFactoryAddress,
 
+		HonestActors:    []common.Address{},
 		MonitorInterval: DefaultMonitorInterval,
 		GameWindow:      DefaultGameWindow,
 

op-dispute-mon/flags/flags.go

@@ -10,6 +10,7 @@ import (
 	oplog "github.com/ethereum-optimism/optimism/op-service/log"
 	opmetrics "github.com/ethereum-optimism/optimism/op-service/metrics"
 	"github.com/ethereum-optimism/optimism/op-service/oppprof"
+	"github.com/ethereum/go-ethereum/common"
 )
 
 const (
@@ -33,6 +34,11 @@ var (
 		EnvVars: prefixEnvVars("GAME_FACTORY_ADDRESS"),
 	}
 	// Optional Flags
+	HonestActorsFlag = &cli.StringSliceFlag{
+		Name:    "honest-actors",
+		Usage:   "List of honest actors that are monitored for any claims that are resolved against them.",
+		EnvVars: prefixEnvVars("HONEST_ACTORS"),
+	}
 	RollupRpcFlag = &cli.StringFlag{
 		Name:    "rollup-rpc",
 		Usage:   "HTTP provider URL for the rollup node",
@@ -62,6 +68,7 @@ var requiredFlags = []cli.Flag{
 // optionalFlags is a list of unchecked cli flags
 var optionalFlags = []cli.Flag{
 	RollupRpcFlag,
+	HonestActorsFlag,
 	MonitorIntervalFlag,
 	GameWindowFlag,
 }
@@ -96,13 +103,25 @@ func NewConfigFromCLI(ctx *cli.Context) (*config.Config, error) {
 		return nil, err
 	}
 
+	var actors []common.Address
+	if ctx.IsSet(HonestActorsFlag.Name) {
+		for _, addrStr := range ctx.StringSlice(HonestActorsFlag.Name) {
+			actor, err := opservice.ParseAddress(addrStr)
+			if err != nil {
+				return nil, fmt.Errorf("invalid honest actor address: %w", err)
+			}
+			actors = append(actors, actor)
+		}
+	}
+
 	metricsConfig := opmetrics.ReadCLIConfig(ctx)
 	pprofConfig := oppprof.ReadCLIConfig(ctx)
 
 	return &config.Config{
 		L1EthRpc:           ctx.String(L1EthRpcFlag.Name),
 		GameFactoryAddress: gameFactoryAddress,
 
+		HonestActors:    actors,
 		RollupRpc:       ctx.String(RollupRpcFlag.Name),
 		MonitorInterval: ctx.Duration(MonitorIntervalFlag.Name),
 		GameWindow:      ctx.Duration(GameWindowFlag.Name),

op-dispute-mon/metrics/metrics.go

@@ -69,6 +69,8 @@ type Metricer interface {
 	RecordInfo(version string)
 	RecordUp()
 
+	RecordUnexpectedClaimResolution(address common.Address, count int)
+
 	RecordGameResolutionStatus(complete bool, maxDurationReached bool, count int)
 
 	RecordCredit(expectation CreditExpectation, count int)
@@ -104,6 +106,8 @@ type Metrics struct {
 
 	claims prometheus.GaugeVec
 
+	unexpectedClaimResolutions prometheus.GaugeVec
+
 	withdrawalRequests prometheus.GaugeVec
 
 	info prometheus.GaugeVec
@@ -161,6 +165,13 @@ func NewMetrics() *Metrics {
 			Name:      "claim_resolution_delay_max",
 			Help:      "Maximum claim resolution delay in seconds",
 		}),
+		unexpectedClaimResolutions: *factory.NewGaugeVec(prometheus.GaugeOpts{
+			Namespace: Namespace,
+			Name:      "unexpected_claim_resolutions",
+			Help:      "Total number of unexpected claim resolutions against an honest actor",
+		}, []string{
+			"honest_actor_address",
+		}),
 		resolutionStatus: *factory.NewGaugeVec(prometheus.GaugeOpts{
 			Namespace: Namespace,
 			Name:      "resolution_status",
@@ -251,6 +262,10 @@ func (m *Metrics) RecordUp() {
 	m.up.Set(1)
 }
 
+func (m *Metrics) RecordUnexpectedClaimResolution(address common.Address, count int) {
+	m.unexpectedClaimResolutions.WithLabelValues(address.Hex()).Set(float64(count))
+}
+
 func (m *Metrics) RecordGameResolutionStatus(complete bool, maxDurationReached bool, count int) {
 	completion := "complete"
 	if !complete {

op-dispute-mon/metrics/noop.go

@@ -19,6 +19,8 @@ func (*NoopMetricsImpl) RecordUp()           {}
 func (*NoopMetricsImpl) CacheAdd(_ string, _ int, _ bool) {}
 func (*NoopMetricsImpl) CacheGet(_ string, _ bool)        {}
 
+func (*NoopMetricsImpl) RecordUnexpectedClaimResolution(_ common.Address, _ int) {}
+
 func (*NoopMetricsImpl) RecordGameResolutionStatus(_ bool, _ bool, _ int) {}
 
 func (*NoopMetricsImpl) RecordCredit(_ CreditExpectation, _ int) {}

op-dispute-mon/mon/claims.go

@@ -5,6 +5,7 @@ import (
 
 	"github.com/ethereum-optimism/optimism/op-dispute-mon/metrics"
 	"github.com/ethereum-optimism/optimism/op-dispute-mon/mon/types"
+	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/log"
 )
 
@@ -14,35 +15,60 @@ type RClock interface {
 
 type ClaimMetrics interface {
 	RecordClaims(status metrics.ClaimStatus, count int)
+	RecordUnexpectedClaimResolution(address common.Address, count int)
 }
 
 type ClaimMonitor struct {
-	logger  log.Logger
-	clock   RClock
-	metrics ClaimMetrics
+	logger       log.Logger
+	clock        RClock
+	honestActors []common.Address
+	metrics      ClaimMetrics
 }
 
-func NewClaimMonitor(logger log.Logger, clock RClock, metrics ClaimMetrics) *ClaimMonitor {
-	return &ClaimMonitor{logger, clock, metrics}
+func NewClaimMonitor(logger log.Logger, clock RClock, honestActors []common.Address, metrics ClaimMetrics) *ClaimMonitor {
+	return &ClaimMonitor{logger, clock, honestActors, metrics}
 }
 
 func (c *ClaimMonitor) CheckClaims(games []*types.EnrichedGameData) {
 	claimStatus := make(map[metrics.ClaimStatus]int)
+	unexpected := make(map[common.Address]int)
 	for _, game := range games {
-		c.checkGameClaims(game, claimStatus)
+		c.checkGameClaims(game, claimStatus, unexpected)
 	}
 	for status, count := range claimStatus {
 		c.metrics.RecordClaims(status, count)
 	}
+	for address, count := range unexpected {
+		c.metrics.RecordUnexpectedClaimResolution(address, count)
+	}
+}
+
+func (c *ClaimMonitor) checkResolvedAgainstHonestActor(proxy common.Address, claim *types.EnrichedClaim, unexpected map[common.Address]int) {
+	for _, actor := range c.honestActors {
+		if claim.Claimant == actor && claim.CounteredBy != (common.Address{}) {
+			unexpected[actor]++
+			c.logger.Error("Claim resolved against honest actor", "game", proxy, "honest_actor", actor, "countered_by", claim.CounteredBy, "claim_contract_index", claim.ContractIndex)
+			break
+		}
+	}
 }
 
-func (c *ClaimMonitor) checkGameClaims(game *types.EnrichedGameData, claimStatus map[metrics.ClaimStatus]int) {
+func (c *ClaimMonitor) checkGameClaims(
+	game *types.EnrichedGameData,
+	claimStatus map[metrics.ClaimStatus]int,
+	unexpected map[common.Address]int,
+) {
 	// Check if the game is in the first half
 	duration := uint64(c.clock.Now().Unix()) - game.Timestamp
 	firstHalf := duration <= (game.Duration / 2)
 
 	// Iterate over the game's claims
 	for _, claim := range game.Claims {
+		// Check if the claim has resolved against an honest actor
+		if claim.Resolved {
+			c.checkResolvedAgainstHonestActor(game.Proxy, &claim, unexpected)
+		}
+
 		// Check if the clock has expired
 		if firstHalf && claim.Resolved {
 			c.logger.Error("Claim resolved in the first half of the game duration", "game", game.Proxy, "claimContractIndex", claim.ContractIndex)

op-dispute-mon/mon/claims_test.go

@@ -17,30 +17,51 @@ import (
 var frozen = time.Unix(int64(time.Hour.Seconds()), 0)
 
 func TestClaimMonitor_CheckClaims(t *testing.T) {
-	cm, cl, cMetrics := newTestClaimMonitor(t)
-	games := makeMultipleTestGames(uint64(cl.Now().Unix()))
-	cm.CheckClaims(games)
+	t.Run("RecordsClaims", func(t *testing.T) {
+		monitor, cl, cMetrics := newTestClaimMonitor(t)
+		games := makeMultipleTestGames(uint64(cl.Now().Unix()))
+		monitor.CheckClaims(games)
 
-	require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfExpiredResolved])
-	require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfExpiredUnresolved])
-	require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfNotExpiredResolved])
-	require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfNotExpiredUnresolved])
+		require.Equal(t, 2, cMetrics.calls[metrics.FirstHalfExpiredResolved])
+		require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfExpiredUnresolved])
+		require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfNotExpiredResolved])
+		require.Equal(t, 1, cMetrics.calls[metrics.FirstHalfNotExpiredUnresolved])
 
-	require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfExpiredResolved])
-	require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfExpiredUnresolved])
-	require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfNotExpiredResolved])
-	require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfNotExpiredUnresolved])
+		require.Equal(t, 2, cMetrics.calls[metrics.SecondHalfExpiredResolved])
+		require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfExpiredUnresolved])
+		require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfNotExpiredResolved])
+		require.Equal(t, 1, cMetrics.calls[metrics.SecondHalfNotExpiredUnresolved])
+	})
+
+	t.Run("RecordsUnexpectedClaimResolution", func(t *testing.T) {
+		monitor, cl, cMetrics := newTestClaimMonitor(t)
+		games := makeMultipleTestGames(uint64(cl.Now().Unix()))
+		monitor.CheckClaims(games)
+
+		// Our honest actors 0x01 has claims resolved against them (1 per game)
+		require.Equal(t, 2, cMetrics.unexpected[common.Address{0x01}])
+		require.Equal(t, 0, cMetrics.unexpected[common.Address{0x02}])
+
+		// The other actors should not be metriced
+		require.Equal(t, 0, cMetrics.unexpected[common.Address{0x03}])
+		require.Equal(t, 0, cMetrics.unexpected[common.Address{0x04}])
+	})
 }
 
 func newTestClaimMonitor(t *testing.T) (*ClaimMonitor, *clock.DeterministicClock, *stubClaimMetrics) {
 	logger := testlog.Logger(t, log.LvlInfo)
 	cl := clock.NewDeterministicClock(frozen)
 	metrics := &stubClaimMetrics{}
-	return NewClaimMonitor(logger, cl, metrics), cl, metrics
+	honestActors := []common.Address{
+		common.Address{0x01},
+		common.Address{0x02},
+	}
+	return NewClaimMonitor(logger, cl, honestActors, metrics), cl, metrics
 }
 
 type stubClaimMetrics struct {
-	calls map[metrics.ClaimStatus]int
+	calls      map[metrics.ClaimStatus]int
+	unexpected map[common.Address]int
 }
 
 func (s *stubClaimMetrics) RecordClaims(status metrics.ClaimStatus, count int) {
@@ -50,6 +71,13 @@ func (s *stubClaimMetrics) RecordClaims(status metrics.ClaimStatus, count int) {
 	s.calls[status] += count
 }
 
+func (s *stubClaimMetrics) RecordUnexpectedClaimResolution(address common.Address, count int) {
+	if s.unexpected == nil {
+		s.unexpected = make(map[common.Address]int)
+	}
+	s.unexpected[address] += count
+}
+
 func makeMultipleTestGames(duration uint64) []*types.EnrichedGameData {
 	return []*types.EnrichedGameData{
 		makeTestGame(duration),      // first half
@@ -68,21 +96,36 @@ func makeTestGame(duration uint64) *types.EnrichedGameData {
 		Claims: []types.EnrichedClaim{
 			{
 				Claim: faultTypes.Claim{
-					Clock: faultTypes.NewClock(time.Duration(0), frozen),
+					Clock:    faultTypes.NewClock(time.Duration(0), frozen),
+					Claimant: common.Address{0x02},
+				},
+				Resolved: true,
+			},
+			{
+				Claim: faultTypes.Claim{
+					Claimant:    common.Address{0x01},
+					CounteredBy: common.Address{0x03},
 				},
 				Resolved: true,
 			},
 			{
-				Claim:    faultTypes.Claim{},
+				Claim: faultTypes.Claim{
+					Claimant:    common.Address{0x04},
+					CounteredBy: common.Address{0x02},
+				},
 				Resolved: true,
 			},
 			{
 				Claim: faultTypes.Claim{
-					Clock: faultTypes.NewClock(time.Duration(0), frozen),
+					Claimant:    common.Address{0x04},
+					CounteredBy: common.Address{0x02},
+					Clock:       faultTypes.NewClock(time.Duration(0), frozen),
 				},
 			},
 			{
-				Claim: faultTypes.Claim{},
+				Claim: faultTypes.Claim{
+					Claimant: common.Address{0x01},
+				},
 			},
 		},
 	}

op-dispute-mon/mon/service.go

@@ -88,8 +88,8 @@ func (s *Service) initFromConfig(ctx context.Context, cfg *config.Config) error
 		return fmt.Errorf("failed to init rollup client: %w", err)
 	}
 
+	s.initClaimMonitor(cfg)
 	s.initResolutionMonitor()
-	s.initClaimMonitor()
 	s.initWithdrawalMonitor()
 
 	s.initOutputValidator()   // Must be called before initForecast
@@ -109,12 +109,12 @@ func (s *Service) initFromConfig(ctx context.Context, cfg *config.Config) error
 	return nil
 }
 
-func (s *Service) initResolutionMonitor() {
-	s.resolutions = NewResolutionMonitor(s.logger, s.metrics, s.cl)
+func (s *Service) initClaimMonitor(cfg *config.Config) {
+	s.claims = NewClaimMonitor(s.logger, s.cl, cfg.HonestActors, s.metrics)
 }
 
-func (s *Service) initClaimMonitor() {
-	s.claims = NewClaimMonitor(s.logger, s.cl, s.metrics)
+func (s *Service) initResolutionMonitor() {
+	s.resolutions = NewResolutionMonitor(s.logger, s.metrics, s.cl)
 }
 
 func (s *Service) initWithdrawalMonitor() {