Consumer-driven contracts

We all know the problems, when two (not opposing) teams (A and B) work on specific services and the expectation here is, that after a successful deployment both services work together happily every after.

So the first deployment is done and everything is working fine. The next milestone has to be reached and one team wants to add a new shiny feature, which unfortunately breaks the public API and violates the contract between both services?

For the ease of the discussion, let us say team A is responsible for the consumer of the public API and team B takes care of the producer.

This scenario has lots of problems, but let us talk about two in detail:

  1. The consumer cannot verify its assumptions about the public facing API independently.

  2. The producer cannot easily verify, if any change of the API breaks any existing contract.


This whole problem isn’t new and we have a perfect hammer for this particular nail: We just create a mock or even better a fake from the producer and test the consumer against it:

How does this help to validate assumptions? Or even worse: Who is responsible for updating the mock or fake, when the API really changes?

There has to be another way..

Consumer-driven Contracts

Consumer-driven Contracts (CDC) is a way to help both teams with their problems: It allows the consumer to state, what it expects from a producer in a well-defined format. This format (or contract) includes all the interaction of the two services with the results.

In other words:

A contract between a consuming service and a providing service, stating what the consumer wants from a providing service, in a defined format.

Alas, consumer-driven contracts are not a silver bullet:

  • They cannot test business logic.

  • They still succeed, when something underlying changes, but the format stays the same.

  • They are not a Service Level Agreement (SLA).

  • They cannot validate external API responses.

That out of the way, let us focus on a examples.

There are some frameworks available to support this, like Spring Cloud Contract and Pact - we will focus on the latter one and see shortly, what it really can do for us.


Pact provides a complete set of supporting tools for maven and a broker, to ease the collaboration and sharing of contracts and the results.



Pact comes with a JSON DSL, which allows to easily describe the resulting JSON.ยด A complete explanation of the markup can be found at the website, but here is a short example from the showcase:

@Pact(consumer = "Todo-Consumer")
public RequestResponsePact createPact(PactDslWithProvider builder) {
    return builder (1)
            .given("a running server")
            .uponReceiving("POST /todo")
            .body(new PactDslJsonBody()
                .stringValue("title", "string")
                .stringValue("description", "string")
                .booleanType("done", true)
                    .stringValue("due", "2021-05-07")
                    .stringValue("start", "2021-05-07")
1 The builder from the DSL is used to build the request and describe the expected result.

JUnit5 integration

Pact comes with a proper JUnit5 integration, so specifying the actual expected results can be done pretty easy with the corresponding extension:

void todoCheckPactTest(MockServer mockServer) throws IOException {
    ObjectMapper mapper = new ObjectMapper();
    String endpoint = mockServer.getUrl() + "/todo";

    HttpResponse response = Request.Post(endpoint)
            .bodyString(mapper.writeValueAsString(createTodo()), ContentType.APPLICATION_JSON)

    assertThat(response.getStatusLine().getStatusCode()).isEqualTo(201); (1)
1 And the actual verification takes place here.


The broker allows to uncouple consumer and producer, so that instead of talking directly to each other during tests the communication looks like this:

This makes it possible for both sides to work independently on their services without interrupting other teams.

Test flow

Define the expected result on consumer side

During the test execution, Pact creates a mock server with provider stubs in place. Then, the requests are sent and the results are compared with the actual definition.

Once this succeeds, a contract file is created and stored in targets/pacts.

So to sum this up, the actual contracts are defined as code, can therefore be reproduced and are easy to understand for developers.

Share the generated contract

Sharing the generated contract is also pretty easy:

The example of the showcase is configured to use a Pact broker running inside of a docker container and can be reached under http://localhost:9292, once it has been started via make docker.

And with a call of mvn pact:publish or make pact-publish the contract should be visible in the broker:

pact broker publish

Test the provider

Moving to the provider side, it is time to verify the contract against the actual implementation now.

@PactBroker(valueResolver = AbstractPactTest.PactValueResolver.class)
public class TodoResourcePactProvider extends AbstractPactTest {
    void pactVerificationTestTemplate(PactVerificationContext context) {

    void before(PactVerificationContext context) {
        context.setTarget(new HttpTestTarget("localhost", 8081, "/")); (1)

    static void setUp() {
        startApplication(); (2)

    @State("a running server") (3)
    public void runningState() {
        /* All preparations done? */
1 In this first step the test target is set to the testing configuration of quarkus.
2 Before the first test run, the Quarkus application has to be started manually.
3 This defines a state, which can be used for different setups.

There are multiple ways to start this verification step, the most convenient way is to just execute mvn test and then let Pact upload the result to the broker.

Another option is to execute the aptly named mvn pact:verify or docker pact-verify.

When the test runs successfully, the output should look like this:

Verifying a pact between Todo-Consumer (0.1) and Todo-Provider

    1) The pact at http://localhost:9292/pacts/provider/Todo-Provider/consumer/Todo-Consumer/pact-version/dd4742201f8511b7f05c31f5038c319b2deec46d is being verified because it matches the following configured selection criterion: latest pact between a consumer and Todo-Provider

  [from Pact Broker http://localhost:9292/pacts/provider/Todo-Provider/consumer/Todo-Consumer/pact-version/dd4742201f8511b7f05c31f5038c319b2deec46d/metadata/c1tdW2xdPXRydWUmc1tdW2N2bl09MC4x]
  Given a running server
         WARNING: State Change ignored as there is no stateChange URL
  POST /todo
    returns a response which
      has status code 201 (OK)
      has a matching body (OK)

And the verification result should also be visible in an updated listing:

pact broker verify


Connection to invalid SSL certificates

The maven part of Pact runs inside of another JVM, so adding flags to maven to bypass any SSL issues like -Dmaven.wagon.http.ssl.insecure=true doesn’t help here.

We ultimately got rid of this problem by adding the certificate to the matching JVM:

$ curl -o RootCA.crt
$ keytool -import -alias RootCA -cacerts -file RootCA.crt -storepass changeit -noprompt

A colleague also opened a feature request and gladly they accepted and added it:


Pact takes good care of the bulk work of the consumer-driven contract flow, so it is quite easy go get started with it. In general, adding this to CI/CD can still be a challenge, especially if many stages test or dev may contain different versions of the services.

My showcase can be found here: