datarobot-opentelemetry

OpenTelemetry semantic conventions and utilities for DataRobot telemetry integration.

Installation

# Install base package with semantic conventions only
pip install datarobot-opentelemetry

# Install with integration module for automatic OpenTelemetry setup
pip install datarobot-opentelemetry[integrations]

# Install with FastAPI instrumentation on top of the integration module
pip install datarobot-opentelemetry[fastapi]

The [integrations] extra includes all OpenTelemetry dependencies needed for the configure() function. The [fastapi] extra additionally includes FastAPI/httpx/requests/SQLAlchemy auto-instrumentation and pulls in [integrations] automatically.

Usage

from datarobot_opentelemetry.semconv import SpanAttributes

# Use constants as span attribute keys
span.set_attribute(SpanAttributes.GEN_AI_REQUEST_MODEL, "gpt-4o")
span.set_attribute(SpanAttributes.GEN_AI_USAGE_INPUT_TOKENS, 128)
span.set_attribute(SpanAttributes.GEN_AI_USAGE_OUTPUT_TOKENS, 64)

# DataRobot-specific attributes
span.set_attribute(SpanAttributes.DATAROBOT_TRACE_NAME, "my-agent-trace")
span.set_attribute(SpanAttributes.DATAROBOT_SESSION_ID, session_id)

Available attribute groups

All constants live in datarobot_opentelemetry.semconv.SpanAttributes.

How to use integration

The integration module provides automatic configuration of OpenTelemetry tracing, metrics, and logging to send telemetry data to DataRobot backends.

Basic setup

from datarobot_opentelemetry.integrations import configure

# Configure the OpenTelemetry integration
result = configure(
    endpoint="https://your-telemetry-endpoint.example.com",  # optional if OTEL_EXPORTER_OTLP_ENDPOINT is set
    entity_type="deployment",  # optional if DATAROBOT_ENTITY_TYPE is set
    entity_id="your-entity-id",  # optional if DATAROBOT_ENTITY_ID is set
    api_key="your-api-key",  # optional if DATAROBOT_API_TOKEN is set
)

# Check configuration results
print(f"Tracing configured: {result.tracing_configured}")
print(f"Metrics configured: {result.metrics_configured}")
print(f"Logging configured: {result.logger_configured}")

You can also configure entirely from environment variables:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://your-telemetry-endpoint.example.com"
export DATAROBOT_ENTITY_TYPE="deployment"
export DATAROBOT_ENTITY_ID="your-entity-id"
export DATAROBOT_API_TOKEN="your-api-key"

or using OTEL specific environment variables that take precedence:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://your-telemetry-endpoint.example.com"
export OTEL_EXPORTER_OTLP_HEADERS="X-DataRobot-Entity-Id=deployment-<your-entity-id>,X-DataRobot-Api-Key=<your-api-key>"

from datarobot_opentelemetry.integrations import configure

result = configure()

Configuration parameters

• endpoint (optional argument, required value): OTLP HTTP endpoint URL for telemetry data. If not passed, uses OTEL_EXPORTER_OTLP_ENDPOINT.
• entity_type (optional argument, required value): Type of entity being monitored, e.g. EntityType.DEPLOYMENT or EntityType.WORKLOAD (datarobot_opentelemetry.enums.EntityType). Accepts any string too, since the platform can introduce entity kinds before this enum is updated. If not passed, uses DATAROBOT_ENTITY_TYPE.
• entity_id (optional argument, required value): Unique identifier for the entity. If not passed, uses DATAROBOT_ENTITY_ID.
• api_key (optional argument, required value): API key for authentication. If not passed, uses DATAROBOT_API_TOKEN.
• log_level (optional): Logging level for the integration (default: logging.INFO)
• metrics_export_interval (optional): Interval in milliseconds for exporting metrics (default: 60000)

Argument values take precedence over environment variables.

Advanced usage

After calling configure(), standard OpenTelemetry APIs work automatically:

from opentelemetry import trace, metrics

# Get tracer and record spans
tracer = trace.get_tracer(__name__)
with tracer.start_as_current_span("my-operation") as span:
    span.set_attribute("custom.attribute", "value")
    # Your code here

# Get meter and record metrics  
meter = metrics.get_meter(__name__)
counter = meter.create_counter("my.counter")
counter.add(1)

How to use FastAPI instrumentation

datarobot_opentelemetry.instrumentations.fastapi layers FastAPI/httpx/requests/SQLAlchemy auto-instrumentation and redacted log export on top of configure(), instead of re-implementing provider setup. configure() remains the single place that builds Trace/Log/Metric providers; this module adds what's specific to FastAPI applications.

Basic setup

from fastapi import FastAPI

from datarobot_opentelemetry.enums import EntityType
from datarobot_opentelemetry.instrumentations.fastapi import OTel

otel = OTel(entity_type=EntityType.CUSTOM_APPLICATION, entity_id="your-entity-id")

# config only needs otel_exporter_otlp_endpoint / otel_exporter_otlp_headers /
# otel_sdk_disabled attributes - any app Settings/Config class works, no inheritance required
result = otel.configure(config)

app = FastAPI()
otel.instrument_fastapi_app(app)

Available methods

• configure(config): applies OTel settings from app config. Call once during startup; a second call is a no-op.
• instrument_fastapi_app(app): adds FastAPI/httpx/requests/SQLAlchemy auto-instrumentation.
• trace / meter / meter_and_trace: decorators for tracing and recording call-count metrics on a function.
• span(name) / time(name): context managers for a manual span or a duration metric.
• get_logger(name) / get_tracer(name) / get_meter(name): pass-through accessors for the configured providers.
• log_application_start(application_name): logs a structured startup line.
• shutdown(): flushes and shuts down the configured providers.

OTel is a singleton, since only one set of providers can be configured per process.

How to use uvicorn logging

datarobot_opentelemetry.instrumentations.uvicorn routes uvicorn's access/error loggers through the same formatters and redaction as the rest of the app, and filters out health check request noise.

from datarobot_opentelemetry.instrumentations.uvicorn import configure_uvicorn_logging

configure_uvicorn_logging(log_format="json", log_level="INFO")

Requirements

• Python 3.10+

Release History

• See CHANGELOG.md for version-by-version release notes.