Distributed Tracing with OpenTelemetry

Overview

OpenTelemetry is the CNCF standard for distributed tracing. It provides a SDK to create spans (units of work) and traces (a tree of spans across services). When a request flows through multiple microservices, OpenTelemetry propagates the trace context via headers, so you can see the full path in a UI like Jaeger or Zipkin.

When to Use

• Microservices where you need to trace a request across service boundaries
• Identifying latency bottlenecks in a distributed system (which service is slow?)
• Debugging errors that span multiple services (where did the request fail?)
• Measuring the time spent in each part of a request (database, external API, queue)
• Correlating logs across services via trace IDs

When NOT to Use

• Monolithic applications — use a profiler (py-spy, cProfile) instead
• Simple scripts or CLI tools — tracing adds overhead and configuration
• Applications without a trace backend (Jaeger, Zipkin, Tempo) — spans are useless without visualization

Solution

Setup

pip install opentelemetry-api opentelemetry-sdk
pip install opentelemetry-exporter-jaeger
pip install opentelemetry-instrumentation-flask
pip install opentelemetry-instrumentation-requests

Basic span creation

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.jaeger.thrift import JaegerExporter
from opentelemetry.sdk.resources import Resource

# Configure tracer
resource = Resource.create({"service.name": "user-service"})
provider = TracerProvider(resource=resource)
jaeger_exporter = JaegerExporter(
    agent_host_name="localhost",
    agent_port=6831,
)
provider.add_span_processor(BatchSpanProcessor(jaeger_exporter))
trace.set_tracer_provider(provider)

tracer = trace.get_tracer(__name__)

# Create spans
with tracer.start_as_current_span("process_order") as span:
    span.set_attribute("order.id", "ord-123")
    span.set_attribute("order.total", 99.99)

    with tracer.start_as_current_span("validate_payment"):
        span.set_attribute("payment.method", "credit_card")
        # ... validation logic ...

    with tracer.start_as_current_span("save_to_database"):
        span.set_attribute("db.operation", "INSERT")
        # ... database save ...

Auto-instrumentation for Flask

from flask import Flask
from opentelemetry.instrumentation.flask import FlaskInstrumentor
from opentelemetry.instrumentation.requests import RequestsInstrumentor

app = Flask(__name__)
FlaskInstrumentor().instrument_app(app)
RequestsInstrumentor().instrument()

@app.route("/api/users/<user_id>")
def get_user(user_id):
    # Flask auto-instrumentation creates a span for this request
    # with method, path, status code, and duration
    response = requests.get(f"https://api.example.com/users/{user_id}")
    # requests auto-instrumentation creates a child span for the HTTP call
    return response.json()

Manual context propagation across services

import requests
from opentelemetry import trace, propagate

tracer = trace.get_tracer(__name__)

def call_external_service(url: str, data: dict):
    with tracer.start_as_current_span("call_external") as span:
        headers = {"Content-Type": "application/json"}

        # Inject trace context into headers
        propagate.inject(headers)

        span.set_attribute("http.url", url)
        response = requests.post(url, json=data, headers=headers)
        span.set_attribute("http.status_code", response.status_code)

        return response.json()

The receiving service extracts the context:

from flask import Flask, request
from opentelemetry import propagate, trace

app = Flask(__name__)

@app.before_request
def extract_context():
    ctx = propagate.extract(dict(request.headers))
    token = trace.set_span_in_context(trace.get_tracer(__name__).start_span("handle_request"))
    # Context is now linked to the incoming trace

Adding span events and exceptions

from opentelemetry import trace
from opentelemetry.trace import Status, StatusCode

tracer = trace.get_tracer(__name__)

def process_payment(order_id: str):
    with tracer.start_as_current_span("process_payment") as span:
        span.set_attribute("order.id", order_id)

        try:
            result = charge_card(order_id)
            span.add_event("payment_succeeded", {"amount": result.amount})
            return result
        except PaymentError as e:
            span.record_exception(e)
            span.set_status(Status(StatusCode.ERROR, str(e)))
            raise

Custom span attributes and resources

from opentelemetry.sdk.resources import Resource, SERVICE_NAME, SERVICE_VERSION

resource = Resource.create({
    SERVICE_NAME: "order-service",
    SERVICE_VERSION: "1.2.0",
    "deployment.environment": "production",
    "host.name": "order-api-01",
})

provider = TracerProvider(resource=resource)
trace.set_tracer_provider(provider)

# Every span from this service includes these resource attributes
with tracer.start_as_current_span("create_order") as span:
    span.set_attribute("order.customer_id", 42)
    span.set_attribute("order.items", 3)
    span.set_attribute("order.total", 149.98)

Using OTLP exporter instead of Jaeger

from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter

otlp_exporter = OTLPSpanExporter(
    endpoint="http://localhost:4317",
    insecure=True,
)
provider.add_span_processor(BatchSpanProcessor(otlp_exporter))

Instrumenting SQLAlchemy queries

from opentelemetry.instrumentation.sqlalchemy import SQLAlchemyInstrumentor
from sqlalchemy import create_engine

engine = create_engine("postgresql://user:pass@localhost/mydb")
SQLAlchemyInstrumentor().instrument(engine=engine)

# Every query now creates a span with the SQL statement and duration

Instrumenting Redis

from opentelemetry.instrumentation.redis import RedisInstrumentor
import redis

RedisInstrumentor().instrument()
client = redis.Redis(host="localhost", port=6379)

# Every Redis command creates a span
client.set("key", "value")
value = client.get("key")

Variants

Using OpenTelemetry with FastAPI

from fastapi import FastAPI
from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor

app = FastAPI()
FastAPIInstrumentor.instrument_app(app)

@app.get("/api/users/{user_id}")
async def get_user(user_id: str):
    return {"id": user_id, "name": "Alice"}

Using OpenTelemetry with Celery

from opentelemetry.instrumentation.celery import CeleryInstrumentor
from celery import Celery

CeleryInstrumentor().instrument()

app = Celery("tasks", broker="redis://localhost:6379")

@app.task
def process_order(order_id: str):
    # Each task execution creates a span
    with tracer.start_as_current_span("process_order_task"):
        # ... process ...
        pass

Sampling configuration

from opentelemetry.sdk.trace.sampling import TraceIdRatioBased

# Sample 10% of traces in production
sampler = TraceIdRatioBased(0.1)
provider = TracerProvider(resource=resource, sampler=sampler)

Best Practices

• For a deeper guide, see Distributed Tracing: OpenTelemetry, Jaeger, Zipkin.

• Use auto-instrumentation first — it covers Flask, requests, SQLAlchemy, Redis without code changes

• Add manual spans for business logic that auto-instrumentation can’t see (e.g., “validate_order”, “calculate_pricing”)

• Set service.name in the resource — it identifies your service in Jaeger

• Use span.record_exception() and span.set_status(ERROR) in catch blocks — errors appear in the trace UI

• Keep span attributes low-cardinality — don’t use user IDs or request IDs as attributes

• Use BatchSpanProcessor in production — it batches and sends spans asynchronously

• Use TraceIdRatioBased(0.1) sampling in high-traffic production — 100% sampling is for development

Common Mistakes

• Not propagating context: if you don’t inject/extract trace headers, spans from different services won’t link into a single trace.
• Creating too many spans: a span per loop iteration creates thousands of spans. Wrap the loop, not each iteration.
• Not setting error status: catching an exception without span.set_status(ERROR) hides failures in the trace UI.
• Using AlwaysOn sampler in production: 100% sampling at high traffic overwhelms the trace backend. Use ratio-based sampling.
• Forgetting to instrument outgoing HTTP calls: without RequestsInstrumentor, calls to other services don’t create child spans.