Generate traces and metrics with Python

This document describes how to modify a Python app to collect trace and metric data using the open source OpenTelemetry framework, and how to write structured JSON logs to standard out. This document also provides information about a sample Python app that you can install and run. The app uses the Flask web framework and is configured to generate metrics, traces, and logs.

To learn more about instrumentation, see the following documents:

About manual and automatic instrumentation

For this language, OpenTelemetry defines automatic instrumentation as the practice of collecting telemetry from libraries and frameworks without making code changes. However, you do have install modules and set environment variables.

This document doesn't describe automatic instrumentation. For information about that topic, see Automatic Instrumentation for Python.

For general information, see OpenTelemetry Instrumentation for Python.

Before you begin

Enable the Cloud Logging, Cloud Monitoring, and Cloud Trace APIs.

Enable the APIs

Instrument your app to collect traces, metrics, and logs

To instrument your app to collect trace and metric data and to write structured JSON to standard out, perform the following steps as described in subsequent sections of this document:

  1. Configure OpenTelemetry
  2. Configure structured logging

Configure OpenTelemetry

This example app is configured to use the OpenTelemetry Python SDK to export traces and metrics by using the OTLP protocol. By default, the OpenTelemetry Python SDK uses the W3C Trace Context format for propagating trace context, which ensures that spans have the correct parent-child relationship within a trace.

The following code sample illustrates a Python module to setup OpenTelemetry. To view the full sample, click More, and then select View on GitHub.

resource = Resource.create(attributes={
    # Use the PID as the to avoid duplicate timeseries
    # from different Gunicorn worker processes.
    SERVICE_INSTANCE_ID: f"worker-{os.getpid()}",

traceProvider = TracerProvider(resource=resource)
processor = BatchSpanProcessor(OTLPSpanExporter())

reader = PeriodicExportingMetricReader(
meterProvider = MeterProvider(metric_readers=[reader], resource=resource)

The Flask app relies on Gunicorn to serve HTTP requests following the recommendations in Flask's Deploying to Production guide. Gunicorn starts multiple copies of your app running in independent worker processes to increase throughput. To ensure that metrics from the worker processes don't conflict with each other, we recommend that each worker process sets a unique value for the resource attribute. One way to do this is by including the process ID in the For more information, see Time-series collisions.

For more information and configuration options, see OpenTelemetry Python instrumentation.

Configure structured logging

To write structured logs that are linked to traces, configure your app to output JSON-formatted logs to standard out with keys containing trace information. The following code sample illustrates how to configure the standard logging library to output JSON structured logs using the python-json-logger library, and how to use the opentelemetry-instrumentation-logging package to include trace information.


logHandler = logging.StreamHandler()
formatter = jsonlogger.JsonFormatter(
    "%(asctime)s %(levelname)s %(message)s %(otelTraceID)s %(otelSpanID)s %(otelTraceSampled)s",
        "levelname": "severity",
        "asctime": "timestamp",
        "otelTraceID": "",
        "otelSpanID": "",
        "otelTraceSampled": "",

The previous configuration extracts information about the active span from the log message, and then adds that information as attributes to the JSON structured log. These attributes can then be used to correlate a log with a trace:

  • Resource name of the trace associated with the log entry.
  • The span ID with the trace that is associated with the log entry.
  • The value of this field must be true or false.

For more information about these fields, see the LogEntry structure.

Run a sample app configured to collect telemetry

The example app uses vendor-neutral formats, including JSON for logs and OTLP for metrics and traces. Telemetry from the app is routed to Google Cloud using the OpenTelemetry Collector configured with Google exporters. It uses Flask to serve HTTP requests, and the requests library for making HTTP requests. To generate metrics and traces for the HTTP client and server, the example app installs the opentelemetry-instrumentation-flask and opentelemetry-instrumentation-requests instrumentation libraries:

logger = logging.getLogger(__name__)

app = Flask(__name__)

The app has two endpoints:

  • The /multi endpoint is handled by the multi function. The load generator in the app issues requests to the /multi endpoint. When this endpoint receives a request, it sends between three and seven requests to the /single endpoint on the local server.

    def multi():
        """Handle an http request by making 3-7 http requests to the /single endpoint."""
        subRequests = randint(3, 7)"handle /multi request", extra={'subRequests': subRequests})
        for _ in range(subRequests):
            requests.get(url_for('single', _external=True))
        return 'ok'
  • The /single endpoint is handled by the single function. When this endpoint receives a request, it sleeps for a short delay and then responds with a string.

    def single():
        """Handle an http request by sleeping for 100-200 ms, and write the number of seconds slept as the response."""
        duration = uniform(0.1, 0.2)
        return f'slept {duration} seconds'

Download and deploy the app

To run the sample, do the following:

  1. In the Google Cloud console, activate Cloud Shell.

    Activate Cloud Shell

    At the bottom of the Google Cloud console, a Cloud Shell session starts and displays a command-line prompt. Cloud Shell is a shell environment with the Google Cloud CLI already installed and with values already set for your current project. It can take a few seconds for the session to initialize.

  2. Clone the repository:

    git clone
  3. Go to the sample directory:

    cd opentelemetry-operations-python/samples/instrumentation-quickstart
  4. Build and run the sample:

    docker compose up --abort-on-container-exit

    If you aren't running on Cloud Shell, then run the application with the GOOGLE_APPLICATION_CREDENTIALS environment variable pointing to a credentials file. Application Default Credentials provides a credentials file at $HOME/.config/gcloud/application_default_credentials.json.

    # Set environment variables
    export GOOGLE_APPLICATION_CREDENTIALS="$HOME/.config/gcloud/application_default_credentials.json"
    export USERID="$(id -u)"
    # Run
    docker compose -f docker-compose.yaml -f docker-compose.creds.yaml up --abort-on-container-exit

View your metrics

The OpenTelemetry instrumentation in the sample app generates Prometheus metrics that you can view by using the Metrics Explorer:

  • Prometheus/http_server_duration_milliseconds/histogram records the duration of server requests and stores the results in a histogram.

  • Prometheus/http_client_duration_milliseconds/histogram records the duration of client requests and stores the results in a histogram.

To view the metrics generated by the sample app, do the following:
  1. In the Google Cloud console, go to the  Metrics explorer page:

    Go to Metrics explorer

    If you use the search bar to find this page, then select the result whose subheading is Monitoring.

  2. In the Metric element, expand the Select a metric menu, enter http_server in the filter bar, and then use the submenus to select a specific resource type and metric:
    1. In the Active resources menu, select Prometheus Target.
    2. In the Active metric categories menu, select Http.
    3. In the Active metrics menu, select a metric.
    4. Click Apply.
  3. Configure how the data is viewed.

    When the measurements for a metric are cumulative, Metrics Explorer automatically normalizes the measured data by the alignment period, which which results in the chart displaying a rate. For more information, see Kinds, types, and conversions.

    When integer or double values are measured, such as with the two counter metrics, Metrics Explorer automatically sums all time series. To view the data for the /multi and /single HTTP routes, set the first menu of the Aggregation entry to None.

    For more information about configuring a chart, see Select metrics when using Metrics Explorer.

View your traces

To view your trace data, do the following:

  1. In the Google Cloud console, go to the Trace explorer page:

    Go to Trace explorer

    You can also find this page by using the search bar.

  2. In the scatter plot, select a trace with the URI of /multi.
  3. In the Gantt chart on the Trace details panel, select the span labeled /multi.

    A panel opens that displays information about the HTTP request. These details include the method, status code, number of bytes, and the user agent of the caller.

  4. To view the logs associated with this trace, select the Logs & Events tab.

    The tab shows individual logs. To view the details of the log entry, expand the log entry. You can also click View Logs and view the log by using the Logs Explorer.

For more information about using the Cloud Trace explorer, see Find and explore traces.

View your logs

From the Logs Explorer, you can inspect your logs, and you can also view associated traces, when they exist.

  1. In the Google Cloud console, go to the Logs Explorer page:

    Go to Logs Explorer

    If you use the search bar to find this page, then select the result whose subheading is Logging.

  2. Locate a log with the description of handle /multi request.

    To view the details of the log, expand the log entry.

  3. Click Traces on a log entry with the "handle /multi request" message, and then select View trace details.

    A Trace details panel opens and displays the selected trace.

For more information about using the Logs Explorer, see View logs by using the Logs Explorer.

What's next