Tutorial: Tracing Python Flask requests with OpenTracing

January 15 Bullet_white By Derek Bullet_white Comments Comments

flash and opentracing

A transaction trace is a GPS system for web performance: it paints a rich picture of the flow of a web request through your code.

So, why doesn't everybody trace? I believe there are two reasons:

  1. Complex instrumentation: Adding in-app tracing instrumentation is more involved than calling logger.info() for logging or statsD_client.incr() for metrics.
  2. Vender lockin: You aren't committing to a vendor when you log and record metrics: you can easily swap out different services to aggregate your logs and store your metrics. Even though APM vendor tracing libraries are remarkably similar, there hasn't been a vendor-neutral standard for tracing. Adding complex, vendor-specific instrumentation can feel like a deeper commitment than one desires.

OpenTracing, a vender-neutral tracing API

Enter OpenTracing, a vendor-neutral open standard for distributed tracing. OpenTracing loosens the chains on tracing instrumentation: if we trace our method calls via OpenTracing APIs, we can swap out our tracing vendors just like logging and metrics!

To date, there are a number of services capable of ingesting transaction traces generated via OpenTracing. In this quick tutorial, we'll capture transaction traces from Python Flask app, and view this with Jaeger, an open-source distributed tracing system.

Note: we're developing an OpenTracing-compatible Python agent at Scout. You can signup for early access.

A side: what's "distributed" mean?

You may have noticed that both OpenTracing and Jaeger's tagline mention "distributed". Besides providing a vendor-neutral instrumentation API, OpenTracing also provides a standard way to propagate trace context across microservices. This is crucial for debugging issues if your app is composed of microservices. I like to think of it as distributed backtraces.

I'll go over distributed Python tracing in a later post: we'll focus on instrumenting a single app in this tutorial.

The setup


This tutorial will use the following parts:

Installing Jaeger All-in-one

The Jaeger All-in-one Docker image is designed for quick local testing. It launches the Jaeger UI, collector, query, and agent, with an in memory storage component. The jaeger-client module installed in our Flask app sends traces to the collector on this running Docker container. Traces are then viewable in the web UI, accessible via http://localhost:16686.

You can install and run Jaeger All-in-one via the following command (assuming Docker is installed on your machine):

docker run -d -e COLLECTOR_ZIPKIN_HTTP_PORT=9411 -p5775:5775/udp -p6831:6831/udp -p6832:6832/udp \
  -p5778:5778 -p16686:16686 -p14268:14268 -p9411:9411 jaegertracing/all-in-one:latest

Verify that the Web UI is accessible at http://localhost:16686. Now that Jaeger is running, let's setup our Flask app.

Flask App

We'll create a simple Flask app to see OpenTracing in action. The app displays pull requests for the opentracing-python project. The app:

Running the app:

FLASK_APP=opentracing_flask_step1.py flask run

The app's output:

$ curl http://localhost:5000
OpenTracing Pull Requests: Implement ScopeManager for in-process propagation, Implement global tracer support

Adding the Jaeger Client and Flask-Opentracing

Let's import jaeger-client, Flask-Opentracing, and initialize the tracer:

tracer diff

View the code.

Collect our first trace

Send a web request to http://localhost:5000. Wait a moment, then navigate to http://localhost:16686 and click the "Find Traces" button. You should see your first transaction trace!

first trace

Adding spans

Flask-Opentracing automatically traces web requests to our Flask app. However, it doesn't instrument method calls that occur within a web request. In our app, it'd be nice to instrument the GitHub API call and JSON parsing as these are two distinct areas of logic. To do this, we can wrap a span around each piece of code we'd like to instrument:


View the code

Note how I've added tags. This provides additional context for the span. Jaeger lets us search on these tags as well.

Send a web request to http://localhost:5000, wait a moment, then refresh http://localhost:16686. Inspect your most recent, richer trace:

trace with spans


Note: we're developing an OpenTracing-compatible Python agent at Scout. You can signup for early access.

With a couple of pieces, we've added vender-neutral tracing to a Python Flask app via OpenTracing. We can view these traces with Jaeger, an open-source distributed tracing system. If we'd like to use a tracing system other than Jeager at some point, we can swap out the tracer by changing a single line of code in our Python app.

Up Next: Distributed tracing in Python

OpenTracing not only defines a consistent API for tracing method calls. It also defines a consistent API for propagating transaction traces across requests handled by microservices. Soon I'll show how to trace a request across a Django and Flask app.

Get notified of new posts.

Once a month, we'll deliver a finely-curated selection of optimization tips to your inbox.


comments powered by Disqus