Micrometer Tracing (replaces Sleuth)

OpenTelemetry integration, trace propagation and log correlation.

Advanced11 min read

Read these first

Structured Logs and Metrics

Why this matters

A slow checkout might spend 400 ms in inventory, 2 seconds in payment, and 50 ms in your API — logs alone cannot show where the time went across service boundaries.
Micrometer Tracing replaced Spring Cloud Sleuth in Boot 3.x; it uses OpenTelemetry under the hood and propagates trace context through HTTP headers and message metadata.
Correlating traces with logs (same trace ID in MDC) turns a Grafana trace view into a jump-off point for the exact log lines on each hop.

Tracing stack in Boot 3.4

Micrometer Tracing — abstraction layer; bridges to OpenTelemetry or Brave exporters.
OpenTelemetry SDK — vendor-neutral instrumentation and export to Jaeger, Zipkin, or OTLP collectors.
traceId / spanId — identifiers propagated across services; inject into MDC for log correlation.
ObservationRegistry — creates spans around HTTP requests, @Scheduled jobs, and custom code.
OTLP exporter — sends spans to Grafana Tempo, Honeycomb, or any OpenTelemetry-compatible backend.

Logs

Logback + MDCStructured JSON

Metrics

MicrometerCounters, timers

Traces

OpenTelemetryTrace propagation

Logs, metrics and traces — the three pillars wired through Actuator and Micrometer.

Dependencies and configuration

Java

<dependency>
    <groupId>io.micrometer</groupId>
    <artifactId>micrometer-tracing-bridge-otel</artifactId>
</dependency>
<dependency>
    <groupId>io.opentelemetry</groupId>
    <artifactId>opentelemetry-exporter-otlp</artifactId>
</dependency>

Java

management:
  tracing:
    sampling:
      probability: 1.0  # reduce in production, e.g. 0.1 for 10%
  otlp:
    tracing:
      endpoint: http://otel-collector:4318/v1/traces

logging:
  pattern:
    level: "%5p [${spring.application.name:},%X{traceId:-},%X{spanId:-}]"

Automatic instrumentation

Spring Boot auto-instruments:

Auto-instrumented components

Incoming HTTP requests via ServerHttpObservationFilter
Outgoing RestClient / WebClient calls (trace headers injected)
JDBC queries when the datasource is observed
Kafka producers and consumers when Spring Kafka tracing is enabled

Java

@Service
public class CheckoutOrchestrator {

    private final RestClient inventoryClient;
    private final RestClient paymentClient;
    private final ObservationRegistry observations;

    public CheckoutOrchestrator(RestClient.Builder builder,
                                ObservationRegistry observations) {
        this.inventoryClient = builder.baseUrl("http://inventory-service").build();
        this.paymentClient = builder.baseUrl("http://payment-service").build();
        this.observations = observations;
    }

    public CheckoutResult checkout(CheckoutRequest request) {
        return Observation.createNotStarted("bookstore.checkout", observations)
                .lowCardinalityKeyValue("customer.id", request.customerId())
                .observe(() -> {
                    reserveInventory(request);
                    chargePayment(request);
                    return persistOrder(request);
                });
    }
}

Kafka trace propagation

Java

spring:
  kafka:
    template:
      observation-enabled: true
    listener:
      observation-enabled: true

Java

@KafkaListener(topics = "bookstore.orders.placed", groupId = "bookstore-analytics")
public void onOrderPlaced(OrderPlacedMessage message) {
    log.info("Processing order {} in analytics pipeline", message.orderId());
    analyticsService.record(message);
}

The consumer span links to the producer span automatically when observation is enabled on both sides.

Quick recall

Everything you need if you only revisit this box.

Micrometer Tracing + OpenTelemetry bridge replaces Sleuth in Spring Boot 3.x.
Auto-instrumentation covers HTTP in/out, JDBC, and Kafka when observation flags are enabled.
Put traceId and spanId in log patterns via MDC for trace-to-log correlation.
Use Observation.createNotStarted(...).observe() for custom business spans like full checkout.
Tune management.tracing.sampling.probability — full sampling is for dev, not production.

Test yourself

Answer these before moving on — recall is what makes it stick.