observability

Structured observability patterns for Python APIs using Logfire and OpenTelemetry.

Overview

Use Logfire for fast setup and OpenTelemetry for vendor-neutral exports. Always set service metadata, instrument key frameworks (FastAPI/httpx), and control sampling.

When to Use

• Tracing API requests and dependencies
• Capturing structured logs with context
• Exporting telemetry to OTLP collectors

Quick Start

uv pip install logfire

import logfire
from fastapi import FastAPI

app = FastAPI()
logfire.configure(service_name="backend")
logfire.instrument_fastapi(app)
logfire.instrument_httpx()

Core Patterns

1. Service metadata: set service_name and environment.
2. Automatic instrumentation: FastAPI + httpx for request spans.
3. Structured logs: include request IDs and user context.
4. Sampling: reduce volume for high-traffic services.
5. OTLP exports: keep backends swappable.

OpenTelemetry Export Example

from opentelemetry import trace
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor

resource = Resource(attributes={"service.name": "service"})
trace.set_tracer_provider(TracerProvider(resource=resource))
tracer = trace.get_tracer(__name__)

exporter = OTLPSpanExporter()
trace.get_tracer_provider().add_span_processor(BatchSpanProcessor(exporter))

with tracer.start_as_current_span("healthcheck"):
    pass

Troubleshooting

• Missing service name: spans hard to find in UI
• High-cardinality attributes: explode storage
• No spans: instrumentation called too late

References

• https://logfire.pydantic.dev/docs/
• https://opentelemetry.io/docs/instrumentation/python/