• September 29, 2026
• 4 min read

Jaeger — Distributed Tracing for OpenTelemetry¶

Open-source (Apache 2.0, CNCF graduated) backend for distributed traces: it receives OTLP spans, stores them and shows the path of one request through all services. Jaeger v2 is built on the OpenTelemetry Collector.

Where Jaeger Fits¶

flowchart LR
  subgraph SUT["System under test"]
    API["orders-api<br/>FastAPI + OTel SDK"] --> PAY["payments-api<br/>OTel SDK"]
    API --> DB[("PostgreSQL")]
  end
  TESTS["pytest suite<br/>OTel SDK, traceparent"] -- "HTTP + traceparent" --> API
  TESTS -- "OTLP" --> J
  API -- "OTLP gRPC :4317<br/>OTLP HTTP :4318" --> J["Jaeger v2<br/>receivers → storage"]
  PAY -- "OTLP" --> J
  J --> UI["Jaeger UI / API :16686"]
  J --> ST[("memory / Badger /<br/>Elasticsearch / OpenSearch / Cassandra")]

• OpenTelemetry — produces the spans. Instrumentation, SDK setup and the Collector are covered in OpenTelemetry — Python Observability.
• Jaeger — stores spans, searches traces, renders the timeline, compares traces, builds the service dependency graph.
• Tests — start the trace and pass traceparent to the system under test, so one Jaeger trace shows the test and every backend call it caused.

Section Map¶

Minimal Setup¶

docker run --rm --name jaeger \
  -p 16686:16686 -p 4317:4317 -p 4318:4318 \
  jaegertracing/jaeger:latest        # pin a version in CI

# uv add opentelemetry-sdk opentelemetry-exporter-otlp-proto-grpc
from opentelemetry import trace
from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor

provider = TracerProvider(resource=Resource.create({"service.name": "orders-api"}))
provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter(endpoint="http://localhost:4317", insecure=True)))
trace.set_tracer_provider(provider)

with trace.get_tracer(__name__).start_as_current_span("checkout") as span:
    span.set_attribute("order.id", "ord-42")

provider.shutdown()   # flush before the script exits
# Open http://localhost:16686 -> Service "orders-api" -> Find Traces

Quick Commands¶

Ports¶

Jaeger vs Phoenix, Langfuse and Tempo¶

Rule of thumb: Jaeger answers which service was slow or failed; Phoenix and Langfuse answer what the LLM saw and said. Both can receive the same OTLP stream through one Collector.

Quick Rules¶

1. Use OTLP only — the old Jaeger client libraries and opentelemetry-exporter-jaeger are retired; every current SDK speaks OTLP to Jaeger.
2. Use the v2 image jaegertracing/jaeger — jaegertracing/all-in-one and the separate v1 images are Jaeger v1, which is end-of-life.
3. Always set service.name — otherwise the service list shows unknown_service.
4. Memory storage is for local runs and CI only — it is lost on restart and capped by max_traces.
5. Start traces in the tests and propagate traceparent, so a failed test points to exactly one trace.
6. Flush spans (provider.shutdown()) at the end of test sessions and short-lived scripts.
7. Pin the image version in CI and upgrade on purpose.
8. Put a Collector in front in shared environments — tail sampling and redaction happen there, not in Jaeger.

See also¶

• Digital Garden: Knowledge Base
• Tools — Practical Reference Guides
• OpenTelemetry — Python Observability
• OpenTelemetry — Collector & Backends
• Arize Phoenix — LLM Tracing & Evaluation
• Langfuse — LLM Tracing, Prompts & Evals
• MLflow — Experiment Tracking, LLM Tracing & Evaluation
• Client–Server: Observability