EVM-850 Validator rootchain balance metrics

command/server/config/config.go

@@ -39,6 +39,8 @@ type Config struct {
 
 	ConcurrentRequestsDebug uint64 `json:"concurrent_requests_debug" yaml:"concurrent_requests_debug"`
 	WebSocketReadLimit      uint64 `json:"web_socket_read_limit" yaml:"web_socket_read_limit"`
+
+	MetricsInterval time.Duration `json:"metrics_interval" yaml:"metrics_interval"`
 }
 
 // Telemetry holds the config details for metric services.
@@ -96,6 +98,10 @@ const (
 	// DefaultRelayerTrackerPollInterval specifies time interval after which relayer node's event tracker
 	// polls child chain to get the latest block
 	DefaultRelayerTrackerPollInterval time.Duration = time.Second
+
+	// DefaultMetricsInterval specifies the time interval after which Prometheus metrics will be generated.
+	// A value of 0 means the metrics are disabled.
+	DefaultMetricsInterval time.Duration = time.Second * 8
 )
 
 // DefaultConfig returns the default server configuration
@@ -136,6 +142,7 @@ func DefaultConfig() *Config {
 		ConcurrentRequestsDebug:    DefaultConcurrentRequestsDebug,
 		WebSocketReadLimit:         DefaultWebSocketReadLimit,
 		RelayerTrackerPollInterval: DefaultRelayerTrackerPollInterval,
+		MetricsInterval:            DefaultMetricsInterval,
 	}
 }
 

command/server/params.go

@@ -45,6 +45,8 @@ const (
 	webSocketReadLimitFlag      = "websocket-read-limit"
 
 	relayerTrackerPollIntervalFlag = "relayer-poll-interval"
+
+	metricsIntervalFlag = "metrics-interval"
 )
 
 // Flags that are deprecated, but need to be preserved for
@@ -190,5 +192,6 @@ func (p *serverParams) generateConfig() *server.Config {
 		Relayer:                    p.relayer,
 		NumBlockConfirmations:      p.rawConfig.NumBlockConfirmations,
 		RelayerTrackerPollInterval: p.rawConfig.RelayerTrackerPollInterval,
+		MetricsInterval:            p.rawConfig.MetricsInterval,
 	}
 }

command/server/server.go

@@ -249,6 +249,13 @@ func setFlags(cmd *cobra.Command) {
 		"interval (number of seconds) at which relayer's tracker polls for latest block at childchain",
 	)
 
+	cmd.Flags().DurationVar(
+		&params.rawConfig.MetricsInterval,
+		metricsIntervalFlag,
+		defaultConfig.MetricsInterval,
+		"interval (number of seconds) at which special metrics is generated. zero means the metrics is disabled",
+	)
+
 	setLegacyFlags(cmd)
 
 	setDevFlags(cmd)

consensus/consensus.go

@@ -3,6 +3,7 @@ package consensus
 import (
 	"context"
 	"log"
+	"time"
 
 	"github.com/0xPolygon/polygon-edge/blockchain"
 	"github.com/0xPolygon/polygon-edge/chain"
@@ -78,6 +79,7 @@ type Params struct {
 	BlockTime      uint64
 
 	NumBlockConfirmations uint64
+	MetricsInterval       time.Duration
 }
 
 // Factory is the factory function to create a discovery consensus

consensus/polybft/polybft.go

@@ -527,6 +527,12 @@ func (p *Polybft) Start() error {
 	// start state DB process
 	go p.state.startStatsReleasing()
 
+	// additional polybft metrics
+	go polybftMetrics(
+		p.consensusConfig.Bridge.JSONRPCEndpoint,
+		p.key.Address(), p.closeCh,
+		p.logger, p.config.MetricsInterval)
+
 	return nil
 }
 

consensus/polybft/state_stats.go

@@ -1,10 +1,14 @@
 package polybft
 
 import (
+	"math/big"
 	"time"
 
 	"github.com/armon/go-metrics"
+	"github.com/hashicorp/go-hclog"
 	"github.com/prometheus/client_golang/prometheus"
+	"github.com/umbracle/ethgo"
+	"github.com/umbracle/ethgo/jsonrpc"
 )
 
 // startStatsReleasing starts the process that releases BoltDB stats into prometheus periodically.
@@ -154,3 +158,44 @@ func (s *State) startStatsReleasing() {
 		prev = stats
 	}
 }
+
+func polybftMetrics(rootnodeURL string,
+	validatorAddress ethgo.Address,
+	closeCh <-chan struct{},
+	logger hclog.Logger,
+	interval time.Duration) {
+	// zero means metrics are disabled
+	if interval <= 0 {
+		return
+	}
+
+	gweiPerWei := new(big.Int).Exp(big.NewInt(10), big.NewInt(9), nil) // 10^9
+
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+
+	rpcClient, err := jsonrpc.NewClient(rootnodeURL)
+	if err != nil {
+		logger.Error("metrics - connection to root node failed", "err", err)
+
+		return
+	}
+
+	for {
+		select {
+		case <-closeCh:
+			return
+		case <-ticker.C:
+			balance, err := rpcClient.Eth().GetBalance(validatorAddress, ethgo.Latest)
+			if err != nil {
+				logger.Error("metrics get balance call failed", "err", err)
+
+				continue
+			}
+
+			balanceInGwei := new(big.Int).Div(balance, gweiPerWei).Uint64()
+			metrics.SetGauge([]string{"bridge", "validator_balance_gwei", validatorAddress.String()},
+				float32(balanceInGwei))
+		}
+	}
+}

server/config.go

@@ -46,6 +46,7 @@ type Config struct {
 
 	NumBlockConfirmations      uint64
 	RelayerTrackerPollInterval time.Duration
+	MetricsInterval            time.Duration
 }
 
 // Telemetry holds the config details for metric services

server/server.go

@@ -585,6 +585,7 @@ func (s *Server) setupConsensus() error {
 			SecretsManager:        s.secretsManager,
 			BlockTime:             uint64(blockTime.Seconds()),
 			NumBlockConfirmations: s.config.NumBlockConfirmations,
+			MetricsInterval:       s.config.MetricsInterval,
 		},
 	)