What is Jaeger?

Jaeger was created by Uber Engineering in 2015 to trace requests across their massive microservices architecture. Open-sourced in 2017, donated to CNCF, and graduated in 2019, it’s now one of the most widely deployed distributed tracing systems.

Jaeger answers: “What happened to request X as it traveled through 15 microservices and took 3 seconds?”

Quick Start

# All-in-one Docker (for local development)
docker run -d --name jaeger \
  -e COLLECTOR_OTLP_ENABLED=true \
  -p 16686:16686 \  # UI
  -p 4317:4317 \    # OTLP gRPC
  -p 4318:4318 \    # OTLP HTTP
  jaegertracing/all-in-one:latest

# Open UI
open http://localhost:16686

Sending Traces to Jaeger

With OpenTelemetry SDK:

const { OTLPTraceExporter } = require('@opentelemetry/exporter-trace-otlp-grpc');

const traceExporter = new OTLPTraceExporter({
  url: 'http://jaeger:4317',  // OTLP gRPC
});

# Environment variable approach
OTEL_EXPORTER_OTLP_ENDPOINT=http://jaeger:4317 \
OTEL_SERVICE_NAME=payments-service \
node server.js

Kubernetes Deployment

# Install Jaeger Operator
kubectl create namespace observability
kubectl apply -f https://github.com/jaegertracing/jaeger-operator/releases/latest/download/jaeger-operator.yaml -n observability

# Deploy Jaeger instance (simple in-memory, for development)
cat <<EOF | kubectl apply -f -
apiVersion: jaegertracing.io/v1
kind: Jaeger
metadata:
  name: simplest
spec:
  strategy: allInOne
  allInOne:
    options:
      log-level: debug
  storage:
    type: memory
    options:
      memory:
        max-traces: 100000
EOF

For production with Elasticsearch:

apiVersion: jaegertracing.io/v1
kind: Jaeger
metadata:
  name: production-jaeger
spec:
  strategy: production
  storage:
    type: elasticsearch
    options:
      es:
        server-urls: https://elasticsearch:9200
        index-prefix: jaeger
  ingress:
    enabled: true

Trace Sampling

Jaeger supports multiple sampling strategies:

# sampling-strategies.json
{
  "service_strategies": [
    {
      "service": "payments-service",
      "type": "probabilistic",
      "param": 1.0  # 100% - always sample (low volume critical service)
    },
    {
      "service": "user-service",
      "type": "probabilistic",
      "param": 0.1  # 10% - high volume service
    }
  ],
  "default_strategy": {
    "type": "probabilistic",
    "param": 0.05  # 5% default
  }
}

Adaptive sampling adjusts rates dynamically:

# Adaptive sampling configuration
sampling:
  strategies-file: /etc/jaeger/sampling.json
  strategies-max-ttl: 30s

Service Dependency Map

Jaeger automatically builds a service dependency map from traces. Access it via the UI at /dependencies — it shows:

Which services call which services
Call rates between services
Services with the highest error rates

This is invaluable for understanding blast radius during incidents and for architecture review.

Comparing Traces

One of Jaeger’s most useful features: compare two traces side-by-side:

Find a slow trace for a request
Find a fast trace for the same operation
Use “Compare Traces” to see exactly which span accounts for the difference

This is the most efficient way to diagnose performance regressions.

Jaeger vs Zipkin vs Tempo

Feature	Jaeger	Zipkin	Grafana Tempo
OTLP support	✅	Partial	✅
Storage options	Multiple	Multiple	Object storage
UI	Excellent	Good	Via Grafana
Scalability	Good	Limited	Very high
Cloud integration	Self-hosted	Self-hosted	Grafana Cloud
Best for	On-prem K8s	Simple setups	Grafana stack

Use Jaeger when you want a standalone distributed tracing system with an excellent UI and don’t want to commit to the full Grafana stack.

Jaeger — Distributed Tracing for Microservices

✅ Pros

⚠️ Cons