This repository is organized such that each commit adds more sophistication to the zipkin-tracing from the Envoy repository. Each step is tagged, so you can set the repository to a given step with

git checkout step<n>, e.g. git checkout step1.

The Envoy documentation provides a good overview of how to run the example.

We'll also use wrk and curl to drive load against our services.

We highly recommend you start with step1.

In the last step we used header based routing to deploy a new service version before releasing it. In this step we'll execute an incremental release. We can leave our header-based routing in place, but add a new match rule that will activate 25% of the time. The runtime object in the route match tells Envoy to roll a 100 sided die, and if the result is less than the value of the runtime key (we default it to 25 here), then activate the match. By routing to a different cluster in this match, we can send a percentage of traffic to our new version.

                     retry_on: 5xx
                     num_retries: 3
                     per_try_timeout: 0.300s
+              - match:
+                  prefix: "/"
+                  runtime:
+                    default_value: 25
+                    runtime_key: routing.traffic_shift.helloworld
+                route:
+                  cluster: service1a
+                  retry_policy:
+                    retry_on: 5xx
+                    num_retries: 3
+                    per_try_timeout: 0.300s
               - match:
                   prefix: "/"
                 route:

Shut down your example, if needed by running

docker-compose down --remove-orphans

in the zipkin-tracing directory, and then start your example again by running

docker-compose up --build -d

Now if we make a request to our service with no headers we should see responses from service 1a about 25% of the time.

This touches on a number of Envoy traffic control features, but you can go much deeper. Our product, Houston provides a UI for configuring routes, along with a dashboard that lets you ensure that your changes haven't impacted customers.

There's also video of a talk I gave on these examples, as well as slides. If you have any questions, let me know, [email protected], or @mccv on Twitter.