OpenTelemetry Go-Stack: Tempo Prometheus Loki

# OpenTelemetry-Stack für Go bereitstellen: Tracing, Metriken und Logs

Der OpenTelemetry-Stack für Go-Anwendungen sammelt Traces in Tempo, Metriken in Prometheus und Logs in Loki. Die Client-App app-1 sendet Anfragen an die Server-App app-2 über otel-collector. Grafana visualisiert alle Daten. Der gesamte Stack läuft mit docker-compose und minimaler Konfiguration.

Docker-Compose-Konfiguration

Wichtige Dienste: zwei Go-Anwendungen, otel-collector als Telemetrie-Receiver, Backends Tempo/Prometheus/Loki und Grafana. Das Observability-Netzwerk isoliert den Traffic.

Wichtige Ports:

• app-2: 8080
• otel-collector: 4317 (gRPC), 4318 (HTTP), 8889 (metrics)
• Prometheus: 9090
• Tempo: 3200
• Loki: 3100
• Grafana: 3000

services:
  app-1:
    build: ./app_1
    depends_on:
      otel-collector:
        condition: service_started
    networks:
      - observability
  app-2:
    build: ./app_2
    ports:
      - "8080:8080"
    networks:
      - observability
  otel-collector:
    image: otel/opentelemetry-collector-contrib:latest
    ports:
      - "4318:4318"
      - "4317:4317"
      - "8889:8889"
    volumes:
      - ./configs/otel-collector-config.yaml:/etc/otelcol/config.yaml:ro
    networks:
      - observability
  prometheus:
    image: prom/prometheus:latest
    ports:
      - "9090:9090"
    volumes:
      - ./configs/prometheus.yml:/etc/prometheus/prometheus.yml:ro
    networks:
      - observability
  tempo:
    image: grafana/tempo:2.7.0
    ports:
      - "3200:3200"
    volumes:
      - ./configs/tempo-config.yaml:/etc/tempo/tempo.yaml
    networks:
      - observability
  loki:
    image: grafana/loki:3.3.2
    ports:
      - "3100:3100"
    volumes:
      - ./configs/loki-config.yaml:/etc/loki/local-config.yaml:ro
    networks:
      - observability
  grafana:
    image: grafana/grafana:latest
    ports:
      - "3000:3000"
    volumes:
      - ./configs/grafana/datasources.yaml:/etc/grafana/provisioning/datasources/datasources.yaml:ro
    networks:
      - observability
networks:
  observability:
    driver: bridge

Tempo-Konfiguration für Tracing

Tempo empfängt OTLP über gRPC (4317) und HTTP (4318). Trace-Blöcke werden mit 30m Retention kompaktioniert. trace_idle_period von 5s sorgt für schnelle Sichtbarkeit in der UI.

server:
  http_listen_port: 3200
  grpc_listen_port: 9095
distributor:
  receivers:
    otlp:
      protocols:
        grpc:
          endpoint: "0.0.0.0:4317"
ingester:
  trace_idle_period: 5s
  max_block_duration: 30m
storage:
  trace:
    backend: local
    local:
      path: /var/tempo/traces
compactor:
  compaction:
    block_retention: 30m

Prometheus und Alerting-Regeln

Metriken werden von otel-collector:8889 alle 60s gescraped. Alert-Regeln überwachen P95-Latenz > 2000ms und Error-Rate > 5%.

global:
  scrape_interval: 60s
scrape_configs:
  - job_name: "otel-collector"
    static_configs:
      - targets: ["otel-collector:8889"]

Beispielregeln:

groups:
  - name: app1-service-alerts
    rules:
      - alert: App1LatencyHigh
        expr: histogram_quantile(0.95, sum by (le) (rate(app_1_requests_latency_ms_bucket[5m]))) > 2000
        for: 3m
        labels:
          severity: warning
  - name: app2-service-alerts
    rules:
      - alert: App2HighErrorRate
        expr: rate(app_2_requests_total{status="failed"}[5m]) / rate(app_2_requests_total[5m]) > 0.05
        for: 2m
        labels:
          severity: critical

Loki-Konfiguration für Logs

Logs werden im Dateisystem mit TSDB-Schema v13 gespeichert. Indizierung in 24h-Perioden. Strukturierte Metadaten aktiviert.

auth_enabled: false
server:
  http_listen_port: 3100
common:
  path_prefix: /tmp/loki
  storage:
    filesystem:
      chunks_directory: /tmp/loki/chunks
schema_config:
  configs:
    - from: 2020-10-24
      store: tsdb
      schema: v13
      index:
        period: 24h
limits_config:
  allow_structured_metadata: true

Integration in Grafana

Datasources verknüpfen Metriken→Traces→Logs:

• Prometheus exemplarTraceIdDestinations
• Loki derivedFields by traceID
• Tempo serviceMap und tracesToLogsV2

OTel in der Go-Client initialisieren

Die InitOTel-Funktion erstellt Provider für alle Signale mit Resource-Attributen. 10% Sampling für Traces, Batching für Export.

func InitOTel(ctx context.Context, serviceName string) (*log.LoggerProvider, func(context.Context) error) {
    ctx, cancel := context.WithTimeout(ctx, 10*time.Second)
    defer cancel()

    res := resource.NewWithAttributes(
       semconv.SchemaURL,
       semconv.ServiceName(serviceName),
       semconv.ServiceVersion("1.0.0"),
    )

    // Tracing
    traceExp, err := otlptracegrpc.New(ctx,
       otlptracegrpc.WithEndpoint("otel-collector:4317"),
       otlptracegrpc.WithInsecure(),
    )
    tracerProvider := sdktrace.NewTracerProvider(
       sdktrace.WithSampler(sdktrace.ParentBased(sdktrace.TraceIDRatioBased(0.1))),
       sdktrace.WithBatcher(traceExp),
       sdktrace.WithResource(res),
    )
    otel.SetTracerProvider(tracerProvider)

    // Metrics
    metricExp, err := otlpmetricgrpc.New(ctx,
       otlpmetricgrpc.WithEndpoint("otel-collector:4317"),
       otlpmetricgrpc.WithInsecure(),
    )
    meterProvider := sdkmetric.NewMeterProvider(
       sdkmetric.WithReader(sdkmetric.NewPeriodicReader(metricExp, sdkmetric.WithInterval(10*time.Second))),
       sdkmetric.WithResource(res),
    )
    otel.SetMeterProvider(meterProvider)

    // Logs
    logExp, _ := otlploggrpc.New(ctx,
       otlploggrpc.WithEndpoint("otel-collector:4317"),
       otlploggrpc.WithInsecure(),
    )
    logProvider := log.NewLoggerProvider(
       log.WithProcessor(log.NewBatchProcessor(logExp)),
       log.WithResource(res),
    )

    return logProvider, func(ctx context.Context) error {
       // shutdown logic
       return nil
    }
}

Wichtige Punkte

• Vollständiger OTel-Stack: Tracing (Tempo), Metriken (Prometheus), Logs (Loki) über otel-collector integriert
• Go-Instrumentation: otelhttp.Transport, otelslog.Handler, Counter/Histogramm-Metriken
• Native Grafana-Integration: Übergänge von Metriken→Traces→Logs per traceID
• Performance: trace_idle_period 5s, 10% Sampling, 10s Periodic Reader
• Alerting: P95-Latenz > 2000ms, Error-Rate > 5%

— Editorial Team