pulseboard-tracer (Python)

Auto-instrumenting OpenTelemetry tracer for Python, wired for PulseBoard. One install, one CLI flag, no code change.

pip install pulseboard-tracer
pulseboard-bootstrap                 # one-time: pip-installs OTel instrumentations
                                     # for whatever libs are in your venv
PULSE_API_KEY=sk_... pulseboard-run python app.py

Within ~30s your service shows up in the PulseBoard service catalog with HTTP / DB / cache / queue spans flowing.

What you get

• Distributed traces for every popular Python library that has an upstream OpenTelemetry instrumentation: requests, urllib, urllib3, httpx, aiohttp, flask, django, fastapi, starlette, psycopg, psycopg2, pymongo, redis, sqlalchemy, celery, boto3, botocore, grpc, kafka-python, pika, elasticsearch, and more.
• Resource attributes auto-detected for Docker, Kubernetes (downward API env vars), AWS Lambda and GCP Cloud Run.
• OTLP/HTTP exporter wired to your PulseBoard tenant with Authorization: Bearer $PULSE_API_KEY.
• ParentBased(TraceIdRatio) sampling, batched span export — same defaults as the OpenTelemetry contrib distribution.

This package is a thin shell over the upstream opentelemetry-sdk and opentelemetry-distro. We do not fork OTel; you get every patch and instrumentation the upstream community ships.

How it boots

pulseboard-run python app.py

1. Translates PULSE_* env vars into the matching OTEL_* env vars the upstream SDK understands.
2. Execs opentelemetry-instrument python app.py, which discovers every installed opentelemetry-instrumentation-* package via its entry points and patches the matching libraries before your code imports them.

pulseboard-run is the recommended path. The two alternatives below also work because we register ourselves as an OpenTelemetry distro via the opentelemetry_distro entry point:

# Equivalent to pulseboard-run:
opentelemetry-instrument python app.py

# Or programmatic:
python -c "from pulseboard_tracer import start; start(); ..."

Configuration

Every setting is environment-driven. Programmatic overrides are also supported via start(...).

Programmatic use

from pulseboard_tracer import start, shutdown

handle = start(service_name="checkout", sample_ratio=0.1)
try:
    ...
finally:
    shutdown()

Acceptance smoke

PULSE_API_KEY=sk_... PULSE_DEBUG=1 \
    pulseboard-run python examples/smoke.py

You should see [pulseboard-tracer debug] env translated → … on stderr and the in-process HTTP span in your PulseBoard tenant.

Scope (this release)

Shipped in 0.1:

• Traces (OTLP/HTTP, BatchSpanProcessor)
• Metrics + logs export plumbing (PulseBoard-specific exemplar wiring lands with the LLM tracer slice)
• Auto-instrumentation discovery via the upstream opentelemetry_instrumentor entry-point machinery
• Env-based detection for AWS Lambda / GCP Cloud Run / k8s downward API
• pulseboard-run / pulseboard-bootstrap CLI wrappers
• Programmatic start() / shutdown() API

Explicitly not in 0.1:

• Continuous CPU/heap profiler hook (planned with the agent-side profiler service)
• Error / exception capture beyond what OTel spans already record
• LLM request shape capture (planned with the LLM observability slice)
• Framework-specific extensions beyond the upstream auto-instrumentations (FastAPI/Celery/LangChain extras land in a later slice)
• Source-map / debug-symbol upload helper

Development

python -m venv .venv
source .venv/bin/activate
pip install -e ".[test]"
pytest -q

CI runs the same suite against Python 3.9 / 3.10 / 3.11 / 3.12.

License

MIT — see LICENSE.