Enable tracing and collect feedback for a flow deployment

After you deploy a generative AI application in production, you might want to enhance your understanding and optimize performance. Trace data for each request, aggregated metrics, and user feedback play critical roles.

In this article, you learn to enable tracing, collect aggregated metrics, and collect user feedback during the inference time of your flow deployment.

Prerequisites

• The Azure CLI and the Azure Machine Learning extension to the Azure CLI.
• An Azure AI Foundry project. If you don't already have a project, you can create one.
• An Application Insights resource. If you don't already have an Application Insights resource, you can create one.
• Azure role-based access controls are used to grant access to operations in Azure Machine Learning. To perform the steps in this article, you must have Owner or Contributor permissions on the selected resource group. For more information, see Role-based access control in the Azure AI Foundry portal.

Deploy a flow for real-time inference

After you test your flow properly (either a flex flow or a DAG flow), you can deploy the flow in production. In this article, we use Deploy a flow for real-time inference as the example. For flex flows, you need to prepare the flow.flex.yaml file instead of flow.dag.yaml.

You can also deploy to other platforms, such as Docker container and Kubernetes cluster.

Use the latest prompt flow base image to deploy the flow so that it supports the tracing and feedback collection API.

Enable trace and collect system metrics for your deployment

If you're using the Azure AI Foundry portal to deploy, select Deployment > Application Insights diagnostics > Advanced settings in the deployment wizard. In this way, the tracing data and system metrics are collected to the project linked to Application Insights.

If you're using the SDK or the CLI, add the app_insights_enabled: true property in the deployment .yaml file that collects data to the project linked to Application Insights.

app_insights_enabled: true

You can also specify other application insights by the environment variable APPLICATIONINSIGHTS_CONNECTION_STRING in the deployment .yaml file. You can find the connection string for Application Insights on the Overview page in the Azure portal.

environment_variables:
  APPLICATIONINSIGHTS_CONNECTION_STRING: <connection_string>

View tracing data in Application Insights

Traces record specific events or the state of an application during execution. It can include data about function calls, variable values, and system events. Traces help to break down an application's components into discrete inputs and outputs. This process is crucial for debugging and understanding an application. To learn more about traces, see this website. The trace data follows the OpenTelemetry specification.

You can view the detailed trace in the application insights that you specified. The following screenshot shows an example of an event of a deployed flow that contains multiple nodes. Select Application Insights > Investigate > Transaction search, and then select each node to view its detailed trace.

The Dependency type event records calls from your deployments. The name of the event is the name of the flow folder. To learn more, see Transaction search and diagnostics in Application Insights.

View system metrics in Application Insights

You can find the workspace default Application Insights metrics on your workspace overview page in the Azure portal.

1. Open Application Insights and select Usage and estimated costs on the left pane. Select Custom metrics (Preview) > With dimensions, and save the change.
2. Select the Metrics tab on the left pane. From Metric Namespace, select promptflow standard metrics. You can explore the metrics from the Metric dropdown list with different aggregation methods.

Collect feedback and send to Application Insights

Prompt flow serving provides a new /feedback API to help customers collect the feedback. The feedback payload can be any JSON format data. Prompt flow serving helps the customer save the feedback data to a trace span. Data is saved to the trace exporter target that the customer configured. Prompt flow serving also supports OpenTelemetry standard trace context propagation. It respects the trace context set in the request header and uses that context as the request parent span context. You can use the distributed tracing functionality to correlate the feedback trace to its chat request trace.

The following sample code shows how to score a flow deployed to a managed endpoint that was enabled for tracing and send the feedback to the same trace span of a scoring request. The flow has the inputs question and chat_history. The output is answer. After the endpoint is scored, feedback is collected and sent to Application Insights as specified when you deploy the flow.

import urllib.request
import json
import os
import ssl
from opentelemetry import trace, context
from opentelemetry.baggage.propagation import W3CBaggagePropagator
from opentelemetry.trace.propagation.tracecontext import TraceContextTextMapPropagator
from opentelemetry.sdk.trace import TracerProvider

# Initialize your tracer.
tracer = trace.get_tracer("my.genai.tracer")
trace.set_tracer_provider(TracerProvider())

# Request data goes here.
# The example below assumes JSON formatting, which might be updated
# depending on the format your endpoint expects.
data = {
    "question": "hello",
    "chat_history": []
}

body = str.encode(json.dumps(data))

url = 'https://basic-chat-endpoint-0506.eastus.inference.ml.azure.com/score'
feedback_url = 'https://basic-chat-endpoint-0506.eastus.inference.ml.azure.com/feedback'
# Replace this with the primary/secondary key, AMLToken, or Microsoft Entra ID token for the endpoint.
api_key = ''
if not api_key:
    raise Exception("A key should be provided to invoke the endpoint")

# The azureml-model-deployment header will force the request to go to a specific deployment.
# Remove this header to have the request observe the endpoint traffic rules.
headers = {'Content-Type':'application/json', 'Authorization':('Bearer '+ api_key), 'azureml-model-deployment': 'basic-chat-deployment' }

try:
    with tracer.start_as_current_span('genai-request') as span:

        ctx = context.get_current()
        TraceContextTextMapPropagator().inject(headers, ctx)
        print(headers)
        print(ctx)
        req = urllib.request.Request(url, body, headers)
        response = urllib.request.urlopen(req)

        result = response.read()
        print(result)

        # Now you can process the answer and collect feedback.
        feedback = "thumbdown"  # Example feedback (modify as needed).

        # Make another request to save the feedback.
        feedback_body = str.encode(json.dumps(feedback))
        feedback_req = urllib.request.Request(feedback_url, feedback_body, headers)
        urllib.request.urlopen(feedback_req)


except urllib.error.HTTPError as error:
    print("The request failed with status code: " + str(error.code))

    # Print the headers - they include the requert ID and the timestamp, which are useful for debugging the failure.
    print(error.info())
    print(error.read().decode("utf8", 'ignore'))

You can view the trace of the request along with feedback in Application Insights.

Advanced usage: Export trace to custom OpenTelemetry collector service

In some cases, you might want to export the trace data to your deployed OpenTelemetry collector service. To enable this service, set OTEL_EXPORTER_OTLP_ENDPOINT. Use this exporter when you want to customize your own span processing logic and your own trace persistent target.

Related content

• Get started building a chat app by using the prompt flow SDK
• Work with projects in Visual Studio Code