zerotel 0.2.0

Zero-config observability SDK for Python services — auto-instruments FastAPI with traces, metrics, and logs via OpenTelemetry

zerotel

Zero-config observability SDK for Python services.

Add one line. Get distributed traces, Prometheus metrics, and structured JSON logs — all correlated by trace_id and ready to query in Grafana.

from fastapi import FastAPI
from zerotel import Zerotel

app = FastAPI()
Zerotel(app, service_name="my-api")   # done

---

What you get

Signal	What's captured
Traces	Root span per request — method, route, status, latency, error
Traces	@trace child spans, nested correctly under the request
Traces	SQLAlchemy async queries as child spans (with sanitised SQL)
Metrics	zerotel_requests_total — counter by method/route/status
Metrics	zerotel_request_duration_seconds — histogram
Metrics	zerotel_requests_in_flight — gauge
Logs	Structured JSON with trace_id, span_id, service auto-injected

---

Installation

pip install zerotel

Optional extras:

pip install "zerotel[sqlalchemy]"   # async SQLAlchemy query tracing
pip install "zerotel[flask]"        # Flask WSGI adapter
pip install "zerotel[all]"          # everything

---

Quickstart

Minimal

from fastapi import FastAPI
from zerotel import Zerotel

app = FastAPI()
Zerotel(app, service_name="payments-api")

Full configuration

from fastapi import FastAPI
from zerotel import Zerotel, ZerotelConfig

app = FastAPI()

Zerotel(app, config=ZerotelConfig(
    service_name="payments-api",
    service_version="2.1.0",
    otlp_endpoint="http://otel-collector:4317",
    enable_traces=True,
    enable_metrics=True,
    enable_logging=True,
    exclude_paths=["/health", "/metrics"],  # skip these from tracing
    log_request_body=False,                 # keep off in prod (PII risk)
    trace_sample_rate=1.0,                  # lower in high-volume prod
))

@trace decorator

Add a named child span to any function — works on both async def and def:

from zerotel import trace

@trace(name="send-email")
async def send_email(user_id: int) -> None:
    ...  # this entire function becomes a child span under the request span

@trace
def compute_score(data: list[float]) -> float:
    ...  # span name defaults to "module.compute_score"

Reading the current trace ID

from zerotel import get_trace_id, get_span_id

def my_helper() -> None:
    tid = get_trace_id()   # 32-char hex or "0" * 32 if outside a request
    sid = get_span_id()    # 16-char hex

FastAPI dependency injection

from fastapi import FastAPI
from zerotel.integrations.fastapi import TraceIdDep, RequestContextDep

app = FastAPI()

@app.get("/profile")
async def get_profile(trace_id: TraceIdDep) -> dict:
    return {"trace_id": trace_id}

SQLAlchemy async query tracing

from sqlalchemy.ext.asyncio import create_async_engine
from zerotel.integrations.sqlalchemy import instrument_sqlalchemy

engine = create_async_engine("postgresql+asyncpg://user:pw@localhost/mydb")
instrument_sqlalchemy(engine)
# Every query is now a child span with sanitised SQL as an attribute

---

Local observability stack

The docker/ folder contains a ready-to-use Docker Compose stack:

Service	URL	Purpose
OTel Collector	—	Receives OTLP, fans out
Grafana Tempo	http://localhost:3200	Trace storage
Prometheus	http://localhost:9090	Metrics storage
Grafana Loki	http://localhost:3100	Log aggregation
Grafana	http://localhost:3000	Unified UI

cd docker/
docker compose up -d

# Run your service
OTLP_ENDPOINT=http://localhost:4317 uvicorn myapp:app --reload

# Open Grafana (admin / admin)
open http://localhost:3000

---

Configuration reference

Field	Type	Default	Description
service_name	str	"unknown-service"	Appears on every trace, metric, and log
service_version	str	"0.0.0"	Attached to trace resources
otlp_endpoint	str	"http://localhost:4317"	OTLP gRPC collector address
enable_traces	bool	True	Export OpenTelemetry spans
enable_metrics	bool	True	Expose Prometheus /metrics
enable_logging	bool	True	Configure structlog JSON
exclude_paths	list[str]	["/health", "/metrics"]	Paths to skip from instrumentation
log_request_body	bool	False	Capture request body on span (PII risk)
trace_sample_rate	float	1.0	Fraction of requests to sample (0.0–1.0)
metrics_endpoint	str	"/metrics"	Path for Prometheus scrape endpoint

---

Development

git clone https://github.com/Kamalesh-Kavin/zerotel
cd zerotel
uv sync --extra dev

# Run tests
uv run pytest

# Lint
uv run ruff check src/
uv run ruff format src/

See CONTRIBUTING.md for full guidelines.

---

Philosophy

"If you cannot explain the code without the AI, you haven't learned it yet."

Every file in this project is heavily commented. The goal is for you to be able to read any file cold and understand exactly what it does and why.

---

License

MIT — see LICENSE.