OpenTelemetry in Production Microservices

Distributed tracing only helps when every hop shares a trace ID. In NestJS microservice fleets, consistent OpenTelemetry instrumentation turns vague "it's slow" reports into span waterfalls you can act on.

Instrumentation baseline

• Auto-instrument HTTP, gRPC, and database drivers
• Propagate traceparent headers through the API gateway
• Attach service.name, deployment.environment, and request.id attributes
• Export to OTLP collector—avoid vendor-specific SDK lock-in at the edge

NestJS integration pattern

import { trace } from '@opentelemetry/api';

const tracer = trace.getTracer('blog-service');
const span = tracer.startSpan('posts.publish', { attributes: { postId } });
try {
  await this.outbox.enqueue(event);
} finally {
  span.end();
}

Wrap outbound HTTP calls and Prisma queries in child spans. Missing database spans hide 80% of latency.

Logs and metrics correlation

Sampling strategy

Head-based 100% sampling collapses under load. Use tail sampling in the collector for errors and high-latency traces. Keep always-on sampling for payment and auth paths.

Observability is production infrastructure—invest in correlation IDs before the next cross-service outage.