honeycomb-pycpp 0.1.14

Honeycomb OpenTelemetry distro which embeds the C++ SDK

honeycomb-pycpp

Python bindings for the OpenTelemetry C++ SDK. Provides high-performance tracing via a Pythonic interface, and ships as an OpenTelemetry distro for drop-in use with auto-instrumentation.

This library is experimental.

Installation

pip install honeycomb-pycpp

The wheel bundles the OpenTelemetry C++ SDK — no system-level dependencies required.

Configuration

The SDK is configured via a YAML file following the OpenTelemetry Configuration File Format. A default config is embedded in the package and used when no override is provided.

Environment variable	Description
OTEL_CONFIG_FILE	Path to a custom configuration YAML. Overrides the embedded default.
OTEL_EXPORTER_OTLP_ENDPOINT	OTLP endpoint (default: http://localhost:4318)
OTEL_EXPORTER_OTLP_HEADERS	Headers to send with OTLP requests
OTEL_EXPORTER_OTLP_CERTIFICATE	Configure certificate used to verify a server's TLS credentials.
OTEL_EXPORTER_OTLP_CLIENT_KEY	Configure mTLS client certificate.
OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE	Configure mTLS private client key.
OTEL_RESOURCE_ATTRIBUTES	Comma-separated resource attributes
OTEL_SERVICE_NAME	Service name

Usage

As a distro (auto-instrumentation)

opentelemetry-instrument --service-name my-service python app.py

The distro registers itself automatically via entry points — no code changes required.

Tracing

import honeycomb_pycpp as otel

# Initialize from config file (or uses embedded default)
provider = otel.TracerProvider("path/to/otel.yaml")

tracer = provider.get_tracer("my-tracer")

with tracer.start_as_current_span("my-span") as span:
    span.set_attribute("key", "value")
    # ... do work ...

Metrics

Install the system metrics instrumentor:

pip install opentelemetry-instrumentation-system-metrics

Then activate it programmatically after initializing the MeterProvider:

import honeycomb_pycpp as otel
from opentelemetry import metrics
from opentelemetry.instrumentation.system_metrics import SystemMetricsInstrumentor

provider = otel.MeterProvider("path/to/otel.yaml")
metrics.set_meter_provider(provider)

SystemMetricsInstrumentor().instrument()

This collects CPU, memory, network, and other host metrics and exports them via the C++ SDK.

Logs

import honeycomb_pycpp as otel

provider = otel.LoggerProvider("path/to/otel.yaml")

logger = provider.get_logger("my-logger")

# Emit using keyword arguments
logger.emit(body="something happened", severity_number=otel.SeverityNumber.INFO)

# Emit using a LogRecord object
record = otel.LogRecord()
record.body = "request failed"
record.severity_number = otel.SeverityNumber.ERROR
record.attributes = {"user.id": "u-123", "http.status_code": 500}
logger.emit(record)

# Attach an exception
try:
    1 / 0
except ZeroDivisionError as exc:
    record = otel.LogRecord()
    record.body = "unhandled exception"
    record.severity_number = otel.SeverityNumber.ERROR
    record.exception = exc
    logger.emit(record)

provider.shutdown()

The LoggerProvider can also be used via bridges such that opentelemetry-instrumentation-logging work:

import logging
import honeycomb_pycpp as otel
from opentelemetry.instrumentation.logging.handler import LoggingHandler
from opentelemetry._logs import get_logger_provider

handler = LoggingHandler(logger_provider=otel.LoggerProvider("path/to/otel.yaml"))
logging.getLogger().addHandler(handler)

logging.getLogger("myapp").warning("watch out")

Deploying alongside the standard OpenTelemetry Python distro

Both honeycomb-pycpp and the standard opentelemetry-distro can be installed at the same time. When both are present, opentelemetry-instrument may pick up either distro. Use OTEL_PYTHON_DISTRO and OTEL_PYTHON_CONFIGURATOR to explicitly select which one runs.

pip install honeycomb-pycpp opentelemetry-distro

To use this C++ distro:

OTEL_PYTHON_DISTRO=cpp_distro \
OTEL_PYTHON_CONFIGURATOR=cpp_configurator \
opentelemetry-instrument python app.py

To use the standard Python SDK distro:

OTEL_PYTHON_DISTRO=distro \
OTEL_PYTHON_CONFIGURATOR=configurator \
opentelemetry-instrument python app.py

If neither variable is set and both distros are installed, the one selected is non-deterministic — always set them explicitly in multi-distro environments.

Current limitations

• OTLP gRPC exporter not included in published wheels. Only OTLP HTTP (otlp_http) and console exporters are bundled. Use http/protobuf as your OTLP protocol. To build with gRPC support, install gRPC and its dependencies, then build from source with WITH_OTLP_GRPC=ON:

  WITH_OTLP_GRPC=ON pip install -e .
  

• OpenTelemetry C++ ABI v2 not yet enabled, any features relying on it (i.e. links) are not supported

Building from source

Requirements: Python >= 3.10, CMake >= 3.15, C++17 compiler.

git clone https://github.com/honeycombio/honeycomb-pycpp
cd honeycomb-pycpp
pip install -r requirements-dev.txt
pip install -e .

To rebuild after C++ changes:

pip install -e . --force-reinstall --no-deps

To clean up cmake artifacts:

rm -rf CMakeCache.txt CMakeFiles/ cmake_install.cmake build/ dist/ *.egg-info/ *.so

Shared vs static libraries

Published wheels link the OpenTelemetry C++ core libraries dynamically. When multiple worker processes load the same .so, the OS shares the OTel code pages — reducing per-process memory overhead in multi-worker deployments (e.g. gunicorn, uWSGI).

To build with static libraries instead (suitable for single-process or embedded deployments):

WITH_SHARED_LIBS=OFF pip install -e .

Note: the OTLP HTTP exporter builder libraries are always linked statically regardless of this flag.

License

Apache License 2.0