dimredtracer 0.1.0

OpenTelemetry tracer for DimRed observability platform

dimredtracer

A minimal Python SDK that wraps OpenTelemetry and OpenInference for seamless LLM observability.

Features

• Simple API: Just three methods: set_attribute(), start_span(), and force_flush()
• Environment-driven configuration: No hardcoded endpoints
• OpenInference integration: Pass any OpenInference instrumentor during initialization
• Works with async/await: Full support for async OpenAI clients
• Production-ready: Idempotent initialization, fail-soft error handling

Installation

From source

pip install -r requirements.txt

From GitHub

pip install "git+https://github.com/DimRedLabs/dimred-python-sdk.git@main"
pip install -r requirements.txt

Quick Start

1. Set environment variables

export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT="http://your-collector:4318/v1/traces"
export OTEL_SERVICE_NAME="my-service"
export DIMRED_TENANT_ID="your-tenant-id"  # optional
export OPENAI_API_KEY="sk-..."  # for OpenAI examples

2. Basic usage

from dimredtracer import Tracer
from openinference.instrumentation.openai import OpenAIInstrumentor
from openai import AsyncOpenAI

# Initialize tracer with OpenInference instrumentation
tracer = Tracer(OpenAIInstrumentor())

client = AsyncOpenAI()

# Set attributes on the current span
tracer.set_attribute("experiment", "demo-1")
tracer.set_attribute("user.id", "user-123")

# Make LLM call (auto-instrumented by OpenInference)
response = await client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=[{"role": "user", "content": "Hello!"}],
)

# Add post-call attributes
tracer.set_attribute("response.length", len(response.choices[0].message.content))

# Ensure spans are exported
tracer.force_flush()

3. Custom spans

# Create custom spans for non-LLM operations
with tracer.start_span("preprocessing"):
    tracer.set_attribute("chunks_processed", 42)
    # ... your preprocessing logic ...

with tracer.start_span("postprocessing"):
    tracer.set_attribute("validation_passed", True)
    # ... your postprocessing logic ...

API Reference

Tracer(instrumentation=None, name="dimred.tracer")

Initialize the tracer with optional OpenInference instrumentation.

Parameters:

• instrumentation (optional): An OpenInference instrumentor instance (e.g., OpenAIInstrumentor())
• name (str): Logical name for the OTEL tracer

tracer.set_attribute(key: str, value: Any)

Attach an attribute to the current active span.

tracer.start_span(name: str)

Context manager that creates a custom span and makes it current.

tracer.force_flush(timeout_millis: int = 30000)

Force flush buffered spans to the collector.

Environment Variables

Variable	Required	Description	Example
OTEL_EXPORTER_OTLP_TRACES_ENDPOINT	Yes	Full OTLP HTTP endpoint	http://your-collector:4318/v1/traces
OTEL_SERVICE_NAME	No	Service identifier	my-llm-service (default: dimredtracer-service)
DIMRED_TENANT_ID	No	Tenant ID for multi-tenancy	acme-corp

Examples

See the examples/ directory for complete examples:

• openai_async_example.py: Async OpenAI with OpenInference instrumentation

Requirements

• Python 3.10+
• OpenTelemetry SDK 1.22.0+
• (Optional) OpenInference instrumentation packages