OpenTelemetry Bedrock Instrumentation

Warnings

• breaking The OpenTelemetry Generative AI (GenAI) semantic conventions have undergone significant changes, particularly in attribute naming and structure. Recent versions of `opentelemetry-instrumentation-bedrock` (e.g., 0.53.4, 0.54.0, 0.55.0, 0.57.0) reflect these updates to conform to the latest OTel GenAI semantic conventions (e.g., 0.5.0).
  Fix: Review existing dashboards, alerts, and querying logic that rely on OpenTelemetry span attributes. Update them to use the new semantic convention attribute names. For a transition period, you might use the `OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental` environment variable to explicitly enable the latest experimental conventions or facilitate dual emission, though direct adaptation is recommended for stable setups. Consult the OpenTelemetry semantic conventions documentation for specific changes.
• gotcha By default, this instrumentation logs prompts, completions, and embeddings to span attributes, which can contain sensitive user data. This is done to provide rich observability, but may pose privacy concerns.
  Fix: To disable logging of this content, set the environment variable `TRACELOOP_TRACE_CONTENT` to `false`.
• gotcha Ensure your `boto3` and `botocore` library versions are compatible with the specific AWS Bedrock features you intend to use. Older versions might not support newer Bedrock APIs (e.g., `botocore >= 1.34.116` is required for the `converse` API).
  Fix: Check the AWS SDK documentation for `boto3` and `botocore` for version requirements related to specific Bedrock functionalities you are using. Upgrade `boto3` and `botocore` if necessary: `pip install --upgrade boto3 botocore`.

Install

• pip install opentelemetry-instrumentation-bedrock boto3 Install library and AWS Boto3

Imports

• BedrockInstrumentor

  from opentelemetry.instrumentation.bedrock import BedrockInstrumentor

Quickstart

This quickstart demonstrates how to set up the OpenTelemetry SDK with a console exporter, instrument the Bedrock client, and then invoke a Bedrock model. The instrumentation automatically creates spans for the Bedrock interaction, which are then printed to the console by the `ConsoleSpanExporter`.

import os
import boto3
from opentelemetry import trace
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor
from opentelemetry.instrumentation.bedrock import BedrockInstrumentor

# Configure OpenTelemetry SDK
resource = Resource.create({"service.name": "bedrock-app"})
tracer_provider = TracerProvider(resource=resource)
span_processor = SimpleSpanProcessor(ConsoleSpanExporter())
tracer_provider.add_span_processor(span_processor)
trace.set_tracer_provider(tracer_provider)

# Instrument Bedrock
BedrockInstrumentor().instrument()

# Ensure AWS credentials are set up (e.g., via environment variables or AWS CLI config)
# For a runnable example, ensure AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION are set or mock boto3.
# For this quickstart, we'll use os.environ.get for region, assuming credentials are in place.
aws_region = os.environ.get('AWS_REGION', 'us-east-1')

def invoke_bedrock_model():
    try:
        bedrock_runtime = boto3.client(
            service_name='bedrock-runtime',
            region_name=aws_region
        )
        
        model_id = 'amazon.titan-text-express-v1'
        body = {
            'inputText': 'Explain the importance of OpenTelemetry.',
            'textGenerationConfig': {
                'maxTokenCount': 50,
                'stopSequences': [],
                'temperature': 0.7,
                'topP': 0.9
            }
        }

        response = bedrock_runtime.invoke_model(
            body=json.dumps(body),
            modelId=model_id,
            accept='application/json',
            contentType='application/json'
        )

        response_body = json.loads(response.get('body').read())
        print(f"Bedrock Model Response: {response_body['results'][0]['outputText'].strip()}")
    except Exception as e:
        print(f"Error invoking Bedrock model: {e}")

import json
invoke_bedrock_model()