feat(metrics): migrate sei-cosmos to OpenTelemetry (PLT-353)

evmrpc/state.go

@@ -136,7 +136,7 @@ func (a *StateAPI) GetProof(ctx context.Context, address common.Address, storage
 	for _, key := range storageKeys {
 		paddedKey := common.BytesToHash([]byte(key))
 		formattedKey := append(types.StateKey(address), paddedKey[:]...)
-		qres := queryStore.Query(abci.RequestQuery{
+		qres := queryStore.Query(ctx, abci.RequestQuery{
 			Path:   "/key",
 			Data:   formattedKey,
 			Height: block.Block.Height,

sei-cosmos/baseapp/abci.go

@@ -23,6 +23,8 @@ import (
 	"github.com/sei-protocol/sei-chain/sei-cosmos/types/legacytm"
 	abci "github.com/sei-protocol/sei-chain/sei-tendermint/abci/types"
 	tmproto "github.com/sei-protocol/sei-chain/sei-tendermint/proto/tendermint/types"
+	"go.opentelemetry.io/otel/attribute"
+	otelmetric "go.opentelemetry.io/otel/metric"
 	"google.golang.org/grpc/codes"
 	grpcstatus "google.golang.org/grpc/status"
 )
@@ -102,7 +104,12 @@ func (app *BaseApp) Info(ctx context.Context, req *abci.RequestInfo) (*abci.Resp
 }
 
 func (app *BaseApp) MidBlock(ctx sdk.Context, height int64) (events []abci.Event) {
-	defer telemetry.MeasureSince(time.Now(), "abci", "mid_block")
+	start := time.Now()
+	defer func() {
+		baseappMetrics.midBlockDuration.Record(ctx.Context(), time.Since(start).Seconds())
+		// TODO(PLT-353): remove once baseapp_mid_block_duration verified
+		telemetry.MeasureSince(start, "abci", "mid_block")
+	}()
 
 	if app.midBlocker != nil {
 		midBlockEvents := app.midBlocker(ctx, height)
@@ -114,7 +121,12 @@ func (app *BaseApp) MidBlock(ctx sdk.Context, height int64) (events []abci.Event
 
 // EndBlock implements the ABCI interface.
 func (app *BaseApp) EndBlock(ctx sdk.Context, req abci.RequestEndBlock) (res abci.ResponseEndBlock) {
-	defer telemetry.MeasureSince(time.Now(), "abci", "end_block")
+	start := time.Now()
+	defer func() {
+		baseappMetrics.endBlockDuration.Record(ctx.Context(), time.Since(start).Seconds())
+		// TODO(PLT-353): remove once baseapp_end_block_duration verified
+		telemetry.MeasureSince(start, "abci", "end_block")
+	}()
 
 	if app.endBlocker != nil {
 		res = app.endBlocker(ctx, req)
@@ -166,15 +178,25 @@ func (app *BaseApp) DeliverTxBatch(ctx sdk.Context, req sdk.DeliverTxBatchReques
 // Regardless of tx execution outcome, the ResponseDeliverTx will contain relevant
 // gas execution context.
 func (app *BaseApp) DeliverTx(ctx sdk.Context, req abci.RequestDeliverTxV2, tx sdk.Tx, checksum [32]byte) (res abci.ResponseDeliverTx) {
-	defer telemetry.MeasureSince(time.Now(), "abci", "deliver_tx")
-
+	deliverTxStart := time.Now()
 	gInfo := sdk.GasInfo{}
 	resultStr := "successful"
 
 	defer func() {
+		baseappMetrics.deliverTxDuration.Record(ctx.Context(), time.Since(deliverTxStart).Seconds())
+		// TODO(PLT-353): remove once baseapp_deliver_tx_duration verified
+		telemetry.MeasureSince(deliverTxStart, "abci", "deliver_tx")
+		baseappMetrics.txCount.Add(ctx.Context(), 1)
+		// TODO(PLT-353): remove once baseapp_tx_count verified
 		telemetry.IncrCounter(1, "tx", "count")
+		baseappMetrics.txResult.Add(ctx.Context(), 1, otelmetric.WithAttributes(attribute.String("result", resultStr)))
+		// TODO(PLT-353): remove once baseapp_tx_result verified
 		telemetry.IncrCounter(1, "tx", resultStr)
+		baseappMetrics.txGasUsed.Record(ctx.Context(), int64(gInfo.GasUsed)) //nolint:gosec
+		// TODO(PLT-353): remove once baseapp_tx_gas_used verified
 		telemetry.SetGauge(float32(gInfo.GasUsed), "tx", "gas", "used")
+		baseappMetrics.txGasWanted.Record(ctx.Context(), int64(gInfo.GasWanted)) //nolint:gosec
+		// TODO(PLT-353): remove once baseapp_tx_gas_wanted verified
 		telemetry.SetGauge(float32(gInfo.GasWanted), "tx", "gas", "wanted")
 	}()
 
@@ -245,8 +267,12 @@ func (app *BaseApp) SetDeliverStateToCommit() {
 // against that height and gracefully halt if it matches the latest committed
 // height.
 func (app *BaseApp) Commit(ctx context.Context) (res *abci.ResponseCommit, err error) {
-	defer telemetry.MeasureSince(time.Now(), "abci", "commit")
 	commitStart := time.Now()
+	defer func() {
+		baseappMetrics.commitDuration.Record(ctx, time.Since(commitStart).Seconds())
+		// TODO(PLT-353): remove once baseapp_commit_duration verified
+		telemetry.MeasureSince(commitStart, "abci", "commit")
+	}()
 	app.commitLock.Lock()
 	defer app.commitLock.Unlock()
 
@@ -388,7 +414,13 @@ func (app *BaseApp) Snapshot(height int64) {
 // Query implements the ABCI interface. It delegates to CommitMultiStore if it
 // implements Queryable.
 func (app *BaseApp) Query(ctx context.Context, req *abci.RequestQuery) (res *abci.ResponseQuery, err error) {
-	defer telemetry.MeasureSinceWithLabels([]string{"abci", "query"}, time.Now(), []metrics.Label{{Name: "path", Value: req.Path}})
+	queryStart := time.Now()
+	defer func() {
+		route := app.abciQueryMetricRoute(req.Path)
+		baseappMetrics.abciQueryDuration.Record(ctx, time.Since(queryStart).Seconds(), otelmetric.WithAttributes(attribute.String(abciQueryMetricRouteLabel, route)))
+		// TODO(PLT-353): remove once baseapp_abci_query_duration verified
+		telemetry.MeasureSinceWithLabels([]string{"abci", "query"}, queryStart, []metrics.Label{{Name: "path", Value: req.Path}})
+	}()
 
 	// Add panic recovery for all queries.
 	// ref: https://github.com/cosmos/cosmos-sdk/pull/8039
@@ -425,7 +457,7 @@ func (app *BaseApp) Query(ctx context.Context, req *abci.RequestQuery) (res *abc
 		resp = handleQueryApp(app, path, *req)
 
 	case "store":
-		resp = handleQueryStore(app, path, *req)
+		resp = handleQueryStore(ctx, app, path, *req)
 
 	case "custom":
 		resp = handleQueryCustom(app, path, *req)
@@ -846,7 +878,7 @@ func handleQueryApp(app *BaseApp, path []string, req abci.RequestQuery) abci.Res
 		), app.trace)
 }
 
-func handleQueryStore(app *BaseApp, path []string, req abci.RequestQuery) abci.ResponseQuery {
+func handleQueryStore(ctx context.Context, app *BaseApp, path []string, req abci.RequestQuery) abci.ResponseQuery {
 	var (
 		queryable sdk.Queryable
 		ok        bool
@@ -875,7 +907,7 @@ func handleQueryStore(app *BaseApp, path []string, req abci.RequestQuery) abci.R
 			), app.trace)
 	}
 
-	resp := queryable.Query(req)
+	resp := queryable.Query(ctx, req)
 	resp.Height = req.Height
 
 	return resp
@@ -934,7 +966,12 @@ func splitPath(requestPath string) (path []string) {
 
 // ABCI++
 func (app *BaseApp) ProcessProposal(ctx context.Context, req *abci.RequestProcessProposal) (resp *abci.ResponseProcessProposal, err error) {
-	defer telemetry.MeasureSince(time.Now(), "abci", "process_proposal")
+	processProposalStart := time.Now()
+	defer func() {
+		baseappMetrics.processProposalDuration.Record(ctx, time.Since(processProposalStart).Seconds())
+		// TODO(PLT-353): remove once baseapp_process_proposal_duration verified
+		telemetry.MeasureSince(processProposalStart, "abci", "process_proposal")
+	}()
 	ppStart := time.Now()
 	defer func() { app.execProcessProposalMs = time.Since(ppStart).Milliseconds() }()
 	if app.ChainID != req.Header.ChainID {
@@ -997,7 +1034,12 @@ func (app *BaseApp) ProcessProposal(ctx context.Context, req *abci.RequestProces
 }
 
 func (app *BaseApp) FinalizeBlock(ctx context.Context, req *abci.RequestFinalizeBlock) (*abci.ResponseFinalizeBlock, error) {
-	defer telemetry.MeasureSince(time.Now(), "abci", "finalize_block")
+	finalizeBlockStart := time.Now()
+	defer func() {
+		baseappMetrics.finalizeBlockDuration.Record(ctx, time.Since(finalizeBlockStart).Seconds())
+		// TODO(PLT-353): remove once baseapp_finalize_block_duration verified
+		telemetry.MeasureSince(finalizeBlockStart, "abci", "finalize_block")
+	}()
 	fbStart := time.Now()
 	app.execBlockTxCount = len(req.Txs)
 	defer func() { app.execFinalizeBlockMs = time.Since(fbStart).Milliseconds() }()
@@ -1040,7 +1082,7 @@ func (app *BaseApp) FinalizeBlock(ctx context.Context, req *abci.RequestFinalize
 	}
 }
 
-func (app *BaseApp) GetTxPriorityHint(_ context.Context, req *abci.RequestGetTxPriorityHintV2) (_resp *abci.ResponseGetTxPriorityHint, _err error) {
+func (app *BaseApp) GetTxPriorityHint(ctx context.Context, req *abci.RequestGetTxPriorityHintV2) (_resp *abci.ResponseGetTxPriorityHint, _err error) {
 	defer func() {
 		if r := recover(); r != nil {
 			// Fall back to no-op priority if we panic for any reason. This is to avoid DoS
@@ -1056,7 +1098,12 @@ func (app *BaseApp) GetTxPriorityHint(_ context.Context, req *abci.RequestGetTxP
 		}
 	}()
 
-	defer telemetry.MeasureSince(time.Now(), "abci", "get_tx_priority_hint")
+	priorityHintStart := time.Now()
+	defer func() {
+		baseappMetrics.getTxPriorityHintDuration.Record(ctx, time.Since(priorityHintStart).Seconds())
+		// TODO(PLT-353): remove once baseapp_get_tx_priority_hint_duration verified
+		telemetry.MeasureSince(priorityHintStart, "abci", "get_tx_priority_hint")
+	}()
 
 	tx, err := app.txDecoder(req.Tx)
 	if err != nil {

sei-cosmos/baseapp/abci_query_metrics.go

@@ -0,0 +1,85 @@
+package baseapp
+
+import storetypes "github.com/sei-protocol/sei-chain/sei-cosmos/store/types"
+
+// storeByNameLookup is implemented by CommitMultiStore backends (rootmulti, storev2).
+type storeByNameLookup interface {
+	GetStoreByName(name string) storetypes.Store
+}
+
+// abciStoreQuerySubpaths are the only subpath segments used by mounted store
+// Query implementations (/key, /subspace). See store/rootmulti and storev2/commitment.
+var abciStoreQuerySubpaths = map[string]struct{}{
+	"key":       {},
+	"subspace":  {},
+}
+
+// abciQueryMetricRoute returns a bounded label for ABCI query metrics.
+// Raw client paths are not used directly to avoid unbounded metric cardinality.
+//
+// Rules:
+//   - Registered gRPC query paths are returned as-is (finite set at startup).
+//   - Legacy paths use a fixed prefix + registered segment shape.
+//   - Everything else is "other".
+func (app *BaseApp) abciQueryMetricRoute(reqPath string) string {
+	if app.grpcQueryRouter != nil && app.grpcQueryRouter.Route(reqPath) != nil {
+		return reqPath
+	}
+
+	parts := splitPath(reqPath)
+	if len(parts) == 0 {
+		return "other"
+	}
+
+	switch parts[0] {
+	case "app":
+		if len(parts) >= 2 {
+			switch parts[1] {
+			case "simulate", "version", "snapshots":
+				return "app/" + parts[1]
+			}
+		}
+		return "app/unknown"
+
+	case "store":
+		return app.abciStoreQueryMetricRoute(parts)
+
+	case "custom":
+		if len(parts) >= 2 && parts[1] != "" && app.queryRouter != nil && app.queryRouter.Route(parts[1]) != nil {
+			return "custom/" + parts[1]
+		}
+		return "custom/unknown"
+
+	default:
+		return "other"
+	}
+}
+
+func (app *BaseApp) abciStoreQueryMetricRoute(parts []string) string {
+	if len(parts) < 3 {
+		return "store/unknown"
+	}
+	storeName, subpath := parts[1], parts[2]
+	if _, ok := abciStoreQuerySubpaths[subpath]; !ok {
+		return "store/unknown"
+	}
+	if !app.storeRegisteredForQuery(storeName) {
+		return "store/unknown"
+	}
+	return "store/" + storeName + "/" + subpath
+}
+
+func (app *BaseApp) storeRegisteredForQuery(name string) bool {
+	if lookup, ok := app.cms.(storeByNameLookup); ok && lookup.GetStoreByName(name) != nil {
+		return true
+	}
+	if app.qms != nil {
+		if lookup, ok := app.qms.(storeByNameLookup); ok && lookup.GetStoreByName(name) != nil {
+			return true
+		}
+	}
+	return false
+}
+
+// abciQueryMetricRouteLabel is kept as "path" for compatibility with existing dashboards.
+const abciQueryMetricRouteLabel = "path"

sei-cosmos/baseapp/abci_query_metrics_test.go

@@ -0,0 +1,74 @@
+package baseapp
+
+import (
+	"testing"
+
+	abci "github.com/sei-protocol/sei-chain/sei-tendermint/abci/types"
+	"github.com/sei-protocol/sei-chain/sei-cosmos/testutil"
+	"github.com/sei-protocol/sei-chain/sei-cosmos/testutil/testdata"
+	sdk "github.com/sei-protocol/sei-chain/sei-cosmos/types"
+	"github.com/stretchr/testify/require"
+	dbm "github.com/tendermint/tm-db"
+)
+
+func TestAbciQueryMetricRoute(t *testing.T) {
+	db := dbm.NewMemDB()
+	app := NewBaseApp(t.Name(), db, nil, nil, &testutil.TestAppOpts{})
+
+	grpcQR := app.GRPCQueryRouter()
+	grpcQR.SetInterfaceRegistry(testdata.NewTestInterfaceRegistry())
+	testdata.RegisterQueryServer(grpcQR, testdata.QueryImpl{})
+
+	app.QueryRouter().AddRoute("bank", func(_ sdk.Context, _ []string, _ abci.RequestQuery) ([]byte, error) {
+		return nil, nil
+	})
+
+	grpcEcho := "/testdata.Query/Echo"
+
+	tests := map[string]struct {
+		reqPath  string
+		expected string
+	}{
+		// Production gRPC paths (3-month sample); registered via testdata.Query.
+		"grpc registered": {
+			reqPath:  grpcEcho,
+			expected: grpcEcho,
+		},
+		"grpc unregistered": {
+			reqPath:  "/cosmos.bank.v1beta1.Query/TotalSupply",
+			expected: "other",
+		},
+
+		// Legacy app paths.
+		"app snapshots": {reqPath: "app/snapshots", expected: "app/snapshots"},
+		"app version":   {reqPath: "app/version", expected: "app/version"},
+		"app unknown":   {reqPath: "app/unknown-action", expected: "app/unknown"},
+
+		// Legacy store paths (unknown store name or subpath → single bucket).
+		"store ibc key unregistered": {reqPath: "store/ibc/key", expected: "store/unknown"},
+		"store random attack":        {reqPath: "store/random1/random2", expected: "store/unknown"},
+		"store bad subpath":          {reqPath: "store/key1/evil", expected: "store/unknown"},
+		"store short":                {reqPath: "store/bank", expected: "store/unknown"},
+
+		// Legacy custom paths; subpath segments are not part of the metric label.
+		"custom bank":        {reqPath: "custom/bank/all_balances", expected: "custom/bank"},
+		"custom unregistered": {reqPath: "custom/unknown/foo", expected: "custom/unknown"},
+
+		// Garbage / attack paths.
+		"empty":        {reqPath: "", expected: "other"},
+		"random":       {reqPath: "/totally/made/up", expected: "other"},
+		"leading slash app": {reqPath: "/app/snapshots", expected: "app/snapshots"},
+	}
+
+	for name, tc := range tests {
+		t.Run(name, func(t *testing.T) {
+			require.Equal(t, tc.expected, app.abciQueryMetricRoute(tc.reqPath))
+		})
+	}
+}
+
+func TestAbciQueryMetricRoute_RegisteredStore(t *testing.T) {
+	app := setupBaseApp(t)
+	require.Equal(t, "store/key1/key", app.abciQueryMetricRoute("store/key1/key"))
+	require.Equal(t, "store/unknown", app.abciQueryMetricRoute("store/random1/key"))
+}

sei-cosmos/baseapp/abci_test.go

@@ -216,7 +216,7 @@ func TestHandleQueryStore_NonQueryableMultistore(t *testing.T) {
 		Height: 1,
 	}
 
-	resp := handleQueryStore(app, path, req)
+	resp := handleQueryStore(context.Background(), app, path, req)
 	require.True(t, resp.IsErr())
 	require.Contains(t, resp.Log, "multistore doesn't support queries")
 }