feat: Conditionally emit metrics based on enablement

CHANGELOG.md

@@ -59,6 +59,10 @@ Every module contains its own CHANGELOG.md. Please refer to the module you are i
 
 ### Improvements
 
+* (telemetry) [#19903](https://github.com/cosmos/cosmos-sdk/pull/19903) Conditionally emit metrics based on enablement.
+  * **Introduction of `Now` Function**: Added a new function called `Now` to the telemetry package. It returns the current system time if telemetry is enabled, or a zero time if telemetry is not enabled.
+  * **Atomic Global Variable**: Implemented an atomic global variable to manage the state of telemetry's enablement. This ensures thread safety for the telemetry state.
+  * **Conditional Telemetry Emission**: All telemetry functions have been updated to emit metrics only when telemetry is enabled. They perform a check with `isTelemetryEnabled()` and return early if telemetry is disabled, minimizing unnecessary operations and overhead.
 * (types) [#19869](https://github.com/cosmos/cosmos-sdk/pull/19869) Removed `Any` type from `codec/types` and replaced it with an alias for `cosmos/gogoproto/types/any`.
 * (server) [#19854](https://github.com/cosmos/cosmos-sdk/pull/19854) Add customizability to start command.
     * Add `StartCmdOptions` in `server.AddCommands` instead of `servertypes.ModuleInitFlags`. To set custom flags set them in the `StartCmdOptions` struct on the `AddFlags` field.

baseapp/abci.go

@@ -177,7 +177,8 @@ func (app *BaseApp) Query(_ context.Context, req *abci.RequestQuery) (resp *abci
 
 	telemetry.IncrCounter(1, "query", "count")
 	telemetry.IncrCounter(1, "query", req.Path)
-	defer telemetry.MeasureSince(time.Now(), req.Path)
+	start := telemetry.Now()
+	defer telemetry.MeasureSince(start, req.Path)
 
 	if req.Path == QueryPathBroadcastTx {
 		return sdkerrors.QueryResult(errorsmod.Wrap(sdkerrors.ErrInvalidRequest, "can't route a broadcast tx message"), app.trace), nil

telemetry/metrics.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/http"
+	"sync/atomic"
 	"time"
 
 	"github.com/hashicorp/go-metrics"
@@ -14,6 +15,21 @@ import (
 	"github.com/prometheus/common/expfmt"
 )
 
+// globalTelemetryEnabled is a private variable that stores the telemetry enabled state.
+// It is set on initialization and does not change for the lifetime of the program.
+var globalTelemetryEnabled atomic.Bool
+
+// initTelemetry sets the global variable based on the configuration.
+// It is called only once, at startup, to set the telemetry enabled state.
+func initTelemetry(enabled bool) {
+	globalTelemetryEnabled.Store(enabled)
+}
+
+// isTelemetryEnabled provides controlled access to check if telemetry is enabled.
+func isTelemetryEnabled() bool {
+	return globalTelemetryEnabled.Load()
+}
+
 // globalLabels defines the set of global labels that will be applied to all
 // metrics emitted using the telemetry package function wrappers.
 var globalLabels = []metrics.Label{}
@@ -95,6 +111,7 @@ type GatherResponse struct {
 
 // New creates a new instance of Metrics
 func New(cfg Config) (_ *Metrics, rerr error) {
+	initTelemetry(cfg.Enabled)
 	if !cfg.Enabled {
 		return nil, nil
 	}

telemetry/wrapper.go

@@ -24,6 +24,10 @@ func NewLabel(name, value string) metrics.Label {
 // metric for a module with a given set of keys. If any global labels are defined,
 // they will be added to the module label.
 func ModuleMeasureSince(module string, start time.Time, keys ...string) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.MeasureSinceWithLabels(
 		keys,
 		start.UTC(),
@@ -35,6 +39,10 @@ func ModuleMeasureSince(module string, start time.Time, keys ...string) {
 // module with a given set of keys. If any global labels are defined, they will
 // be added to the module label.
 func ModuleSetGauge(module string, val float32, keys ...string) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.SetGaugeWithLabels(
 		keys,
 		val,
@@ -45,29 +53,58 @@ func ModuleSetGauge(module string, val float32, keys ...string) {
 // IncrCounter provides a wrapper functionality for emitting a counter metric with
 // global labels (if any).
 func IncrCounter(val float32, keys ...string) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.IncrCounterWithLabels(keys, val, globalLabels)
 }
 
 // IncrCounterWithLabels provides a wrapper functionality for emitting a counter
 // metric with global labels (if any) along with the provided labels.
 func IncrCounterWithLabels(keys []string, val float32, labels []metrics.Label) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.IncrCounterWithLabels(keys, val, append(labels, globalLabels...))
 }
 
 // SetGauge provides a wrapper functionality for emitting a gauge metric with
 // global labels (if any).
 func SetGauge(val float32, keys ...string) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.SetGaugeWithLabels(keys, val, globalLabels)
 }
 
 // SetGaugeWithLabels provides a wrapper functionality for emitting a gauge
 // metric with global labels (if any) along with the provided labels.
 func SetGaugeWithLabels(keys []string, val float32, labels []metrics.Label) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.SetGaugeWithLabels(keys, val, append(labels, globalLabels...))
 }
 
 // MeasureSince provides a wrapper functionality for emitting a a time measure
 // metric with global labels (if any).
 func MeasureSince(start time.Time, keys ...string) {
+	if !isTelemetryEnabled() {
+		return
+	}
+
 	metrics.MeasureSinceWithLabels(keys, start.UTC(), globalLabels)
 }
+
+// Now return the current time if telemetry is enabled or a zero time if it's not
+func Now() time.Time {
+	if !isTelemetryEnabled() {
+		return time.Time{}
+	}
+
+	return time.Now()
+}

telemetry/wrapper_test.go

@@ -0,0 +1,54 @@
+package telemetry
+
+import (
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/suite"
+)
+
+// TelemetrySuite is a struct that holds the setup for the telemetry tests.
+// It includes a mutex to ensure that tests that depend on the global state
+// do not run in parallel, which can cause race conditions and unpredictable results.
+type TelemetrySuite struct {
+	suite.Suite
+	mu sync.Mutex
+}
+
+// SetupTest is called before each test to reset the global state to a known disabled state.
+// This ensures each test starts with the telemetry disabled
+func (suite *TelemetrySuite) SetupTest() {
+	initTelemetry(false)
+}
+
+// TestNow tests the Now function when telemetry is enabled and disabled.
+func (suite *TelemetrySuite) TestNow() {
+	suite.mu.Lock()
+	defer suite.mu.Unlock()
+
+	initTelemetry(true)
+	telemetryTime := Now()
+	suite.NotEqual(time.Time{}, telemetryTime, "Now() should not return zero time when telemetry is enabled")
+
+	initTelemetry(false)
+	telemetryTime = Now()
+	suite.Equal(time.Time{}, telemetryTime, "Now() should return zero time when telemetry is disabled")
+}
+
+// TestIsTelemetryEnabled tests the isTelemetryEnabled function.
+func (suite *TelemetrySuite) TestIsTelemetryEnabled() {
+	suite.mu.Lock()
+	defer suite.mu.Unlock()
+
+	initTelemetry(true)
+	suite.True(isTelemetryEnabled(), "isTelemetryEnabled() should return true when globalTelemetryEnabled is set to true")
+
+	initTelemetry(false)
+	suite.False(isTelemetryEnabled(), "isTelemetryEnabled() should return false when globalTelemetryEnabled is set to false")
+}
+
+// TestTelemetrySuite initiates the test suite.
+func TestTelemetrySuite(t *testing.T) {
+	suite.Run(t, new(TelemetrySuite))
+}

x/circuit/module.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"time"
 
 	gwruntime "github.com/grpc-ecosystem/grpc-gateway/runtime"
 	"google.golang.org/grpc"
@@ -93,7 +92,7 @@ func (am AppModule) ValidateGenesis(bz json.RawMessage) error {
 
 // InitGenesis performs genesis initialization for the circuit module.
 func (am AppModule) InitGenesis(ctx context.Context, data json.RawMessage) error {
-	start := time.Now()
+	start := telemetry.Now()
 	var genesisState types.GenesisState
 	if err := am.cdc.UnmarshalJSON(data, &genesisState); err != nil {
 		return err

x/crisis/abci.go

@@ -2,7 +2,6 @@ package crisis
 
 import (
 	"context"
-	"time"
 
 	"github.com/cosmos/cosmos-sdk/telemetry"
 	sdk "github.com/cosmos/cosmos-sdk/types"
@@ -12,7 +11,8 @@ import (
 
 // check all registered invariants
 func EndBlocker(ctx context.Context, k keeper.Keeper) {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyEndBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyEndBlocker)
 
 	sdkCtx := sdk.UnwrapSDKContext(ctx)
 	if k.InvCheckPeriod() == 0 || sdkCtx.BlockHeight()%int64(k.InvCheckPeriod()) != 0 {

x/crisis/module.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"time"
 
 	"github.com/spf13/cobra"
 	"google.golang.org/grpc"
@@ -118,7 +117,7 @@ func (am AppModule) ValidateGenesis(bz json.RawMessage) error {
 
 // InitGenesis performs genesis initialization for the crisis module.
 func (am AppModule) InitGenesis(ctx context.Context, data json.RawMessage) error {
-	start := time.Now()
+	start := telemetry.Now()
 	var genesisState types.GenesisState
 	if err := am.cdc.UnmarshalJSON(data, &genesisState); err != nil {
 		return err

x/distribution/keeper/abci.go

@@ -1,8 +1,6 @@
 package keeper
 
 import (
-	"time"
-
 	"cosmossdk.io/x/distribution/types"
 
 	"github.com/cosmos/cosmos-sdk/telemetry"
@@ -13,7 +11,8 @@ import (
 // and distribute rewards for the previous block.
 // TODO: use context.Context after including the comet service
 func (k Keeper) BeginBlocker(ctx sdk.Context) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyBeginBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyBeginBlocker)
 
 	// determine the total power signing the block
 	var previousTotalPower int64

x/evidence/keeper/abci.go

@@ -3,7 +3,6 @@ package keeper
 import (
 	"context"
 	"fmt"
-	"time"
 
 	"cosmossdk.io/core/comet"
 	"cosmossdk.io/x/evidence/types"
@@ -15,7 +14,8 @@ import (
 // BeginBlocker iterates through and handles any newly discovered evidence of
 // misbehavior submitted by CometBFT. Currently, only equivocation is handled.
 func (k Keeper) BeginBlocker(ctx context.Context) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyBeginBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyBeginBlocker)
 
 	bi := sdk.UnwrapSDKContext(ctx).CometInfo()
 

x/gov/keeper/abci.go

@@ -21,7 +21,8 @@ import (
 
 // EndBlocker is called every block.
 func (k Keeper) EndBlocker(ctx context.Context) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyEndBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyEndBlocker)
 
 	logger := k.Logger()
 	// delete dead proposals from store and returns theirs deposits.

x/mint/keeper/abci.go

@@ -2,7 +2,6 @@ package keeper
 
 import (
 	"context"
-	"time"
 
 	"cosmossdk.io/core/event"
 	"cosmossdk.io/x/mint/types"
@@ -13,7 +12,8 @@ import (
 
 // BeginBlocker mints new tokens for the previous block.
 func (k Keeper) BeginBlocker(ctx context.Context, ic types.InflationCalculationFn) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyBeginBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyBeginBlocker)
 
 	// fetch stored minter & params
 	minter, err := k.Minter.Get(ctx)

x/slashing/abci.go

@@ -2,7 +2,6 @@ package slashing
 
 import (
 	"context"
-	"time"
 
 	"cosmossdk.io/x/slashing/keeper"
 	"cosmossdk.io/x/slashing/types"
@@ -14,7 +13,8 @@ import (
 // BeginBlocker check for infraction evidence or downtime of validators
 // on every begin block
 func BeginBlocker(ctx context.Context, k keeper.Keeper) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyBeginBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyBeginBlocker)
 
 	// Iterate over all the validators which *should* have signed this block
 	// store whether or not they have actually signed it and slash/unbond any

x/staking/keeper/abci.go

@@ -2,7 +2,6 @@ package keeper
 
 import (
 	"context"
-	"time"
 
 	"cosmossdk.io/core/appmodule"
 	"cosmossdk.io/x/staking/types"
@@ -13,12 +12,14 @@ import (
 // BeginBlocker will persist the current header and validator set as a historical entry
 // and prune the oldest entry based on the HistoricalEntries parameter
 func (k *Keeper) BeginBlocker(ctx context.Context) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyBeginBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyBeginBlocker)
 	return k.TrackHistoricalInfo(ctx)
 }
 
 // EndBlocker called at every block, update validator set
 func (k *Keeper) EndBlocker(ctx context.Context) ([]appmodule.ValidatorUpdate, error) {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyEndBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyEndBlocker)
 	return k.BlockValidatorUpdates(ctx)
 }

x/upgrade/keeper/abci.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"errors"
 	"fmt"
-	"time"
 
 	storetypes "cosmossdk.io/store/types"
 	"cosmossdk.io/x/upgrade/types"
@@ -22,7 +21,8 @@ import (
 // a migration to be executed if needed upon this switch (migration defined in the new binary)
 // skipUpgradeHeightArray is a set of block heights for which the upgrade must be skipped
 func (k Keeper) PreBlocker(ctx context.Context) error {
-	defer telemetry.ModuleMeasureSince(types.ModuleName, time.Now(), telemetry.MetricKeyBeginBlocker)
+	start := telemetry.Now()
+	defer telemetry.ModuleMeasureSince(types.ModuleName, start, telemetry.MetricKeyBeginBlocker)
 
 	blockHeight := k.environment.HeaderService.GetHeaderInfo(ctx).Height
 	plan, err := k.GetUpgradePlan(ctx)