feat(relay): add OCR2 on-chain view gauges for median (DF-21989)

Author: cl-efornaciari

core/services/relay/evm/evm.go

@@ -955,7 +955,7 @@ func (r *Relayer) NewMedianProvider(ctx context.Context, rargs commontypes.Relay
 		return nil, err
 	}
 
-	medianContract, err := newMedianContract(configWatcher.ContractConfigTracker(), configWatcher.contractAddress, configWatcher.chain, rargs.JobID, rargs.OracleSpecID, r.ds, lggr)
+	medianContract, err := newMedianContract(configWatcher.ContractConfigTracker(), configWatcher.contractAddress, configWatcher.chain, rargs.JobID, rargs.OracleSpecID, r.ds, lggr, pargs.TransmitterID)
 	if err != nil {
 		return nil, err
 	}

core/services/relay/evm/median.go

@@ -30,9 +30,10 @@ type medianContract struct {
 	configTracker       types.ContractConfigTracker
 	contractCaller      *ocr2aggregator.OCR2AggregatorCaller
 	requestRoundTracker *round.RequestRoundTracker
+	onchainViewMetrics  *medianOnchainViewMetrics
 }
 
-func newMedianContract(configTracker types.ContractConfigTracker, contractAddress common.Address, chain legacyevm.Chain, jobID int32, oracleSpecID int32, ds sqlutil.DataSource, lggr logger.Logger) (*medianContract, error) {
+func newMedianContract(configTracker types.ContractConfigTracker, contractAddress common.Address, chain legacyevm.Chain, jobID int32, oracleSpecID int32, ds sqlutil.DataSource, lggr logger.Logger, transmitterID string) (*medianContract, error) {
 	lggr = logger.Named(lggr, "MedianContract")
 	contract, err := offchain_aggregator_wrapper.NewOffchainAggregator(contractAddress, chain.Client())
 	if err != nil {
@@ -50,9 +51,10 @@ func newMedianContract(configTracker types.ContractConfigTracker, contractAddres
 	}
 
 	return &medianContract{
-		lggr:           lggr,
-		configTracker:  configTracker,
-		contractCaller: contractCaller,
+		lggr:               lggr,
+		configTracker:      configTracker,
+		contractCaller:     contractCaller,
+		onchainViewMetrics: newMedianOnchainViewMetrics(chain.ID().String(), contractAddress.Hex(), transmitterID),
 		requestRoundTracker: round.NewRequestRoundTracker(
 			contract,
 			contractFilterer,
@@ -87,7 +89,14 @@ func (oc *medianContract) HealthReport() map[string]error {
 func (oc *medianContract) LatestTransmissionDetails(ctx context.Context) (ocrtypes.ConfigDigest, uint32, uint8, *big.Int, time.Time, error) {
 	opts := bind.CallOpts{Context: ctx, Pending: false}
 	result, err := oc.contractCaller.LatestTransmissionDetails(&opts)
-	return result.ConfigDigest, result.Epoch, result.Round, result.LatestAnswer, time.Unix(int64(result.LatestTimestamp), 0), errors.Wrap(err, "error getting LatestTransmissionDetails")
+	if err != nil {
+		return ocrtypes.ConfigDigest{}, 0, 0, nil, time.Time{}, errors.Wrap(err, "error getting LatestTransmissionDetails")
+	}
+	updatedAt := time.Unix(int64(result.LatestTimestamp), 0)
+	if oc.onchainViewMetrics != nil {
+		oc.onchainViewMetrics.record(result.LatestAnswer, result.Epoch, result.Round, updatedAt)
+	}
+	return result.ConfigDigest, result.Epoch, result.Round, result.LatestAnswer, updatedAt, nil
 }
 
 // LatestRoundRequested returns the configDigest, epoch, and round from the latest

core/services/relay/evm/median_onchain_view_metrics.go

@@ -0,0 +1,63 @@
+package evm
+
+import (
+	"math/big"
+	"time"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+// Prometheus gauges for the node's view of on-chain OCR2 median aggregator state (DF-22676).
+// latest_answer is exported as float64; values above ~2^53 may lose integer precision (documented in HELP).
+var (
+	promOCR2OnchainLatestAnswer = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "ocr2_onchain_transmission_latest_answer",
+		Help: "Latest median answer from the node's on-chain read of LatestTransmissionDetails (OCR2 aggregator). Float64 may not represent full int256 precision.",
+	}, []string{"chain_id", "contract_address", "transmitter_id"})
+	promOCR2OnchainLatestTimestampUnix = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "ocr2_onchain_transmission_latest_timestamp_unix",
+		Help: "Unix seconds of latest on-chain transmission timestamp from the node's LatestTransmissionDetails read.",
+	}, []string{"chain_id", "contract_address", "transmitter_id"})
+	promOCR2OnchainEpoch = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "ocr2_onchain_transmission_epoch",
+		Help: "Epoch from the node's on-chain LatestTransmissionDetails read.",
+	}, []string{"chain_id", "contract_address", "transmitter_id"})
+	promOCR2OnchainRound = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "ocr2_onchain_transmission_round",
+		Help: "Round from the node's on-chain LatestTransmissionDetails read.",
+	}, []string{"chain_id", "contract_address", "transmitter_id"})
+)
+
+type medianOnchainViewMetrics struct {
+	chainID       string
+	contract      string
+	transmitterID string
+}
+
+func newMedianOnchainViewMetrics(chainID, contractAddress, transmitterID string) *medianOnchainViewMetrics {
+	if transmitterID == "" {
+		transmitterID = "unknown"
+	}
+	return &medianOnchainViewMetrics{
+		chainID:       chainID,
+		contract:      contractAddress,
+		transmitterID: transmitterID,
+	}
+}
+
+func (m *medianOnchainViewMetrics) record(latestAnswer *big.Int, epoch uint32, round uint8, updatedAt time.Time) {
+	lv := bigIntToPromGaugeValue(latestAnswer)
+	promOCR2OnchainLatestAnswer.WithLabelValues(m.chainID, m.contract, m.transmitterID).Set(lv)
+	promOCR2OnchainLatestTimestampUnix.WithLabelValues(m.chainID, m.contract, m.transmitterID).Set(float64(updatedAt.Unix()))
+	promOCR2OnchainEpoch.WithLabelValues(m.chainID, m.contract, m.transmitterID).Set(float64(epoch))
+	promOCR2OnchainRound.WithLabelValues(m.chainID, m.contract, m.transmitterID).Set(float64(round))
+}
+
+func bigIntToPromGaugeValue(i *big.Int) float64 {
+	if i == nil {
+		return 0
+	}
+	f, _ := new(big.Float).SetInt(i).Float64()
+	return f
+}