uniform

pkg/telemetry/logs.go

@@ -12,22 +12,42 @@ import (
 	"go.opentelemetry.io/otel/sdk/log"
 )
 
-// NewLoggerProvider builds the OTLP log provider from the environment, or returns
-// (nil, nil) when logs are disabled — in which case the caller MUST skip
+// LoggerConfig controls the optional OTLP log pipeline.
+// The zero value disables logging (Enabled = false, SampleRatio = 0).
+type LoggerConfig struct {
+	Enabled     bool
+	SampleRatio float64 // 0.0–1.0, clamped by NewLoggerProvider
+}
+
+// LoggerConfigFromEnv reads OTEL_LOGS_ENABLED and OTEL_LOGS_SAMPLE_RATIO from
+// the environment. Call this from main(); library consumers that want explicit
+// control should construct LoggerConfig directly.
+func LoggerConfigFromEnv() LoggerConfig {
+	var enabled bool
+	switch os.Getenv("OTEL_LOGS_ENABLED") {
+	case "1", "true", "TRUE", "True", "yes", "on":
+		enabled = true
+	}
+	ratio := 1.0
+	if v := os.Getenv("OTEL_LOGS_SAMPLE_RATIO"); v != "" {
+		if r, err := strconv.ParseFloat(v, 64); err == nil && r >= 0 && r <= 1 {
+			ratio = r
+		}
+	}
+	return LoggerConfig{Enabled: enabled, SampleRatio: ratio}
+}
+
+// NewLoggerProvider builds the OTLP log provider from cfg, or returns
+// (nil, nil) when cfg.Enabled is false — in which case the caller MUST skip
 // global.SetLoggerProvider and the shutdown registration.
 //
-// Env:
-//
-//	OTEL_LOGS_ENABLED       "1"/"true"/"yes"/"on" to enable the OTLP log pipeline.
-//	                        Default OFF: the batch processor clones its record ring
-//	                        on every export, which profiling showed to be the
-//	                        dominant heap allocator in these services. Traces and
-//	                        metrics are always on regardless of this flag.
-//	OTEL_LOGS_SAMPLE_RATIO  fraction of records to export, 0.0–1.0 (default 1.0).
-//	                        e.g. 0.1 keeps ~10% of logs. Dropped records never reach
-//	                        the batch ring, so allocation churn falls ~proportionally.
-func NewLoggerProvider(ctx context.Context) (*log.LoggerProvider, error) {
-	if !logsEnabled() {
+// Use LoggerConfigFromEnv() to read the conventional OTEL_LOGS_ENABLED /
+// OTEL_LOGS_SAMPLE_RATIO env vars, or construct LoggerConfig directly for
+// explicit control (e.g. tests). Default OFF: the batch processor clones its
+// record ring on every export, which profiling showed to be the dominant heap
+// allocator in these services. Traces and metrics are always on regardless.
+func NewLoggerProvider(ctx context.Context, cfg LoggerConfig) (*log.LoggerProvider, error) {
+	if !cfg.Enabled {
 		return nil, nil
 	}
 	exporter, err := otlploggrpc.New(ctx)
@@ -35,36 +55,12 @@ func NewLoggerProvider(ctx context.Context) (*log.LoggerProvider, error) {
 		return nil, err
 	}
 	var proc log.Processor = log.NewBatchProcessor(exporter)
-	if ratio := sampleRatio(); ratio < 1.0 {
-		proc = &samplingProcessor{next: proc, ratio: ratio}
+	if cfg.SampleRatio < 1.0 {
+		proc = &samplingProcessor{next: proc, ratio: cfg.SampleRatio}
 	}
 	return log.NewLoggerProvider(log.WithProcessor(proc)), nil
 }
 
-func logsEnabled() bool {
-	switch os.Getenv("OTEL_LOGS_ENABLED") {
-	case "1", "true", "TRUE", "True", "yes", "on":
-		return true
-	default:
-		return false
-	}
-}
-
-func sampleRatio() float64 {
-	v := os.Getenv("OTEL_LOGS_SAMPLE_RATIO")
-	if v == "" {
-		return 1.0
-	}
-	r, err := strconv.ParseFloat(v, 64)
-	if err != nil || r < 0 {
-		return 1.0
-	}
-	if r > 1 {
-		return 1.0
-	}
-	return r
-}
-
 // samplingProcessor forwards a random `ratio` fraction of records to the wrapped
 // processor and drops the rest before they reach the (allocation-heavy) batch
 // ring. This is head sampling: cheap, stateless, and per-record independent.

pkg/telemetry/otel.go

@@ -15,11 +15,11 @@ import (
 )
 
 // SetupOTelSDK bootstraps the OpenTelemetry pipeline (propagator + traces +
-// metrics, and an opt-in logger — see NewLoggerProvider). It is shared by every
+// metrics, and an opt-in logger controlled by logCfg). It is shared by every
 // service main package; the service name comes from OTEL_RESOURCE_ATTRIBUTES, so
 // the same setup works everywhere. If it returns no error, call the returned
 // shutdown for proper cleanup.
-func SetupOTelSDK(ctx context.Context) (func(context.Context) error, error) {
+func SetupOTelSDK(ctx context.Context, logCfg LoggerConfig) (func(context.Context) error, error) {
 	var shutdownFuncs []func(context.Context) error
 	var err error
 
@@ -58,7 +58,7 @@ func SetupOTelSDK(ctx context.Context) (func(context.Context) error, error) {
 
 	// Logger provider is opt-in via OTEL_LOGS_ENABLED; nil means logs are off.
 	// Traces + metrics above are always on.
-	loggerProvider, err := NewLoggerProvider(ctx)
+	loggerProvider, err := NewLoggerProvider(ctx, logCfg)
 	if err != nil {
 		handleErr(err)
 		return shutdown, err