2025-02-03 17:30:50 +00:00
|
|
|
package observability
|
|
|
|
|
|
2025-02-03 19:01:37 +00:00
|
|
|
import (
|
|
|
|
|
"context"
|
|
|
|
|
"log/slog"
|
2025-03-06 22:33:39 +00:00
|
|
|
|
2025-03-11 18:15:49 +00:00
|
|
|
"queryorchestration/internal/serviceconfig/build"
|
2025-03-05 16:59:01 +00:00
|
|
|
"queryorchestration/internal/serviceconfig/observability/prometheus"
|
2025-02-03 19:01:37 +00:00
|
|
|
|
|
|
|
|
"go.opentelemetry.io/otel"
|
|
|
|
|
"go.opentelemetry.io/otel/exporters/otlp/otlptrace/otlptracegrpc"
|
|
|
|
|
sdktrace "go.opentelemetry.io/otel/sdk/trace"
|
|
|
|
|
)
|
|
|
|
|
|
2025-02-03 17:30:50 +00:00
|
|
|
type ConfigProvider interface {
|
|
|
|
|
IsOtelEnabled() bool
|
2025-02-03 19:01:37 +00:00
|
|
|
SetOtel(context.Context) func() error
|
2025-03-05 16:59:01 +00:00
|
|
|
GetMetrics() *prometheus.Metrics
|
|
|
|
|
SetMetrics(metrics *prometheus.Metrics)
|
2025-02-03 17:30:50 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
type ObsConfig struct {
|
|
|
|
|
EnableOtel bool `env:"ENABLE_OTEL" envDefault:"false"`
|
2025-03-05 16:59:01 +00:00
|
|
|
metrics *prometheus.Metrics
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func (o *ObsConfig) GetMetrics() *prometheus.Metrics {
|
|
|
|
|
return o.metrics
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func (o *ObsConfig) SetMetrics(metrics *prometheus.Metrics) {
|
|
|
|
|
o.metrics = metrics
|
2025-02-03 17:30:50 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func (o *ObsConfig) IsOtelEnabled() bool {
|
|
|
|
|
return o.EnableOtel
|
|
|
|
|
}
|
2025-02-03 19:01:37 +00:00
|
|
|
|
|
|
|
|
func (o *ObsConfig) SetOtel(ctx context.Context) func() error {
|
|
|
|
|
if !o.EnableOtel {
|
|
|
|
|
slog.Warn("OpenTelemetry is disabled. Set ENABLE_OTEL to true to enable.")
|
|
|
|
|
return func() error { return nil }
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
exporter, err := otlptracegrpc.New(ctx)
|
|
|
|
|
if err != nil {
|
|
|
|
|
return func() error { return nil }
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
tp := sdktrace.NewTracerProvider(
|
|
|
|
|
sdktrace.WithBatcher(exporter),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
otel.SetTracerProvider(tp)
|
|
|
|
|
|
|
|
|
|
return func() error {
|
|
|
|
|
err := tp.Shutdown(ctx)
|
|
|
|
|
if err != nil {
|
|
|
|
|
return err
|
|
|
|
|
}
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
}
|
2025-03-05 16:59:01 +00:00
|
|
|
|
|
|
|
|
const (
|
|
|
|
|
SubSystem = "RunnerService"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
func InitializeMetrics(ctx context.Context, cfg ConfigProvider) error {
|
|
|
|
|
standardDefinitions := prometheus.NewStandardMetricDefinitions()
|
|
|
|
|
config := prometheus.MetricsConfig{
|
|
|
|
|
Port: 8080,
|
2025-03-11 18:15:49 +00:00
|
|
|
Namespace: build.SystemName,
|
2025-03-05 16:59:01 +00:00
|
|
|
Subsystem: SubSystem,
|
|
|
|
|
ServeHttp: true,
|
|
|
|
|
Definitions: standardDefinitions,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
metrics, err := prometheus.New(ctx, config)
|
|
|
|
|
if err != nil {
|
|
|
|
|
slog.Error("Failed to create metrics object", "error", err)
|
|
|
|
|
return err
|
|
|
|
|
}
|
|
|
|
|
slog.Info("Metrics server started on port 8080")
|
|
|
|
|
cfg.SetMetrics(metrics)
|
|
|
|
|
|
|
|
|
|
err = cfg.GetMetrics().RecordLifecycleEvent(1.0, "startup", "main")
|
|
|
|
|
if err != nil {
|
|
|
|
|
slog.Error("Failed to create metrics at startup", "error", err)
|
|
|
|
|
return err
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return nil
|
|
|
|
|
}
|