OpenInference OpenAI Instrumentation

Warnings

• gotcha Compatibility with OpenAI SDK versions: The instrumentation might require specific OpenAI SDK versions to correctly handle new output formats or features. For example, `openai>=1.26` is required to capture token counts when using streaming completions with `stream_options={'include_usage': True}`.
  Fix: Ensure you are using a compatible `openai` SDK version as specified in the `openinference-instrumentation-openai` documentation or releases. Upgrade both if experiencing unexpected trace data.
• gotcha OpenTelemetry `suppress_instrumentation` context flag is not fully respected by `openinference-instrumentation-openai`.
  Fix: Directly reading `_SUPPRESS_INSTRUMENTATION_KEY` for span creation, rather than using OpenTelemetry's `is_instrumentation_suppressed()` utility, can lead to spans being created even when suppression is intended. Manual intervention or alternative span filtering might be necessary if strict suppression is required.
• gotcha In certain environments like Google Colab, `openinference-instrumentation-openai` might fail to instrument correctly immediately after `pip install`. A session restart or explicit dependency check bypass might be needed.
  Fix: After `pip install`, restart the Colab runtime. Alternatively, you can try `OpenAIInstrumentor().instrument(skip_dep_check=True)` as a workaround, though it's generally recommended to restart the runtime.

Install

• pip install openinference-instrumentation-openai "openai>=1.26" opentelemetry-sdk opentelemetry-exporter-otlp arize-phoenix Full Quickstart Dependencies
• pip install openinference-instrumentation-openai Minimal Installation

Imports

• OpenAIInstrumentor

  from openinference.instrumentation.openai import OpenAIInstrumentor

• OpenAI Client

  import openai
  client = openai.OpenAI()

• OTLPSpanExporter

  from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter

• TracerProvider

  from opentelemetry.sdk import trace as trace_sdk

• SimpleSpanProcessor, ConsoleSpanExporter

  from opentelemetry.sdk.trace.export import SimpleSpanProcessor, ConsoleSpanExporter

Quickstart

This quickstart demonstrates how to instrument OpenAI API calls using `openinference-instrumentation-openai` and send the resulting traces to an OpenTelemetry collector. It sets up a `TracerProvider` to export traces via HTTP OTLP and then instruments the OpenAI client. You should ensure an OpenTelemetry collector (like Arize Phoenix, running `python -m phoenix.server.main serve`) is running to receive traces. Remember to set your `OPENAI_API_KEY` environment variable.

import os
import openai
from openinference.instrumentation.openai import OpenAIInstrumentor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk import trace as trace_sdk
from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor

# Set your OpenAI API key from environment variables
os.environ['OPENAI_API_KEY'] = os.environ.get('OPENAI_API_KEY', 'sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx')

# Configure OpenTelemetry Tracer Provider to send traces to a collector (e.g., Phoenix)
endpoint = "http://127.0.0.1:6006/v1/traces" # Default Phoenix endpoint
tracer_provider = trace_sdk.TracerProvider()
tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
# Optionally, also print spans to the console for debugging
tracer_provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))

# Instrument the OpenAI SDK
OpenAIInstrumentor().instrument(tracer_provider=tracer_provider)

if __name__ == "__main__":
    client = openai.OpenAI()
    try:
        response = client.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[{"role": "user", "content": "Write a haiku about observability."}],
            max_tokens=20,
            stream=False # Set to True and add stream_options={'include_usage': True} for streaming with token counts
        )
        print("OpenAI API call successful.")
        print(f"Response: {response.choices[0].message.content}")
    except openai.AuthenticationError:
        print("Error: OpenAI API key is missing or invalid. Please set OPENAI_API_KEY.")
    except Exception as e:
        print(f"An unexpected error occurred: {e}")