Protecting provider builds from changed pacts
Those familiar with automated testing frameworks have probably come across the term "pending tests". These are tests that have a particular flag on them that causes them (depending on the framework) to either be skipped, or if executed, to not fail the test suite.
When using Pact with API Hub for Contract Testing, we have a similar concept of a "pending pact". A pending pact is one which does not yet have a successful verification result published. Pending pacts can fail verification without causing the overall verification task (and hence, the overall provider build) to fail.
Once a pact has a successful verification result published, it is now considered to be an accepted/supported contract, and any subsequent failure can only be as a result of the provider itself changing. Once a pact has left pending state, verification failures will cause the provider build to fail.
Something important to note about the pending calculation is that it is based on the branch for the provider version. For example, once a pact has a successful verification from a provider version with branch master, it will cease to be pending for subsequent master versions, but would still be pending for a feat/x branch of the provider (until the first successful feat/x verification was published, etc.)
For further reading: https://docs.pact.io/pending
Change the pact
Let's add a new field to the expectations we have for the product API. We're going to make the change on master, and it's going to break our consumer build, as it won't be compatible with our provider. This is an exercise in "what not to do, and why".
In your consumer codebase, open up
src/api.pact.spec.js
, scroll down to the first test, and in theexpectedProduct definition
, add a new field e.g.color: "red"
.Run
make test
. These tests should be green - the consumer code is consistent with its own expectations.Commit and push the changes.
Note
👉 Note that in the
Makefile
file, the consumer version is associated with the name of the git branch.publish_pacts: .env @echo "\n========== STAGE: publish pacts ==========\n" @"${PACT_CLI}" publish ${PWD}/pacts --consumer-app-version ${GIT_COMMIT} --branch ${GIT_BRANCH}
Open up the the consumer build in Github Actions. It will generate and publish the pact successfully, then wait for a period of time and then fail on the
can-i-deploy
step, as there is no successful verification from the provider.Open up the provider build in Github Actions. The changed pact will have triggered a pact verification build of the provider project. This will have failed, as the new field does not exist in the API. This particular failed build is expected, and not a problem, as the pact verification build is generally separate from the provider's normal pipeline. For a more detailed explanation of this see https://github.com/pactflow/example-provider#pact-verifications.
Check the pact's status in API Hub for Contract Testing
Open up the pact in API Hub for Contract Testing. You'll see that there is a pact associated with branch master with a failed verification result.
Click on "VIEW PACT" and you'll see that each interaction has a status next to it.
Expand the failing interaction, and you'll see the field level mismatches.
Expected state by the end of this step
A consumer build that is failing at the
can-i-deploy
step in Github Actions.A provider build that is failing during verification in Github Actions.
A pact associated with the consumers main branch
master
in API Hub for Contract Testing that has a failed verification result.
Run the provider main branch build
When a provider build runs, it is configured to verify the latest pact of any registered consumers main branch, so publishing a pact with a new expectation and associating its consumer version as master
(the consumers main branch) will report a (correctly) failed verification result, which stopped our consumer from deploying. Our provider is still compatible with its deployed consumer(s) and therefore is safe to deploy. We use the Pending pacts
feature of Pact to aid us.
You can demonstrate this by running a provider build in Github (Actions
-> Under Workflows
, select Build -> Run workflow
-> Run workflow
).
👉 The test output indicates that the master pact is in pending mode, so even though its verification fails, the overall build still passes.
👉 The provider is still compatible with pact deployed to the prod
environment , so the deploy step is able to proceed without issue.
Look at src/product/product.providerChange.pact.test.js,
enablePending: true
is set in the options for the dynamically fetched pacts.Locally you can run
make test
- note that this passes ✅👉 Whenever we run the verification step with the pending feature enabled, we have to make sure we've set the branch correctly (which we do in this codebase by providing the environment variable
GIT_BRANCH
toproviderVersionBranch
in our testproduct.providerChange.pact.test.js
), because the pending status for each pact depends on it.In API Hub for Contract Testing, refresh the pact page.
👉 Even though the provider build is passing, the
master
pact has a (correctly) failed verification result, which tells the consumer team that they cannot deploy the code associated with this pact yet.
Disable 'pending pacts' for the provider
Note
This step is optional and for demonstration purposes to show what happens without the Pending pacts
feature
In src/product/product.providerChange.pact.test.js, set enablePending: false in the options for the dynamically fetched pacts.
Run
make test
- note that this now fails ❌👉 Without the
Pending pacts
feature, the teams would now be at a stale-mate.The
master
pact has a (correctly) failed verification result, which tells the consumer team that they cannot deploy the code associated with this pact yet.The provider is verifying the code from its main branch to deploy, against the consumers
master
branch, and any deployed versions, and the verification job now fails, meaning the provider is blocked from deployed to production through no fault of their own. We will show how the consumer can safely deliver this change later in the workshop
Caution
The real problem is that the provider is now unable to deploy from their master branch 😧. This is because the provider is configured to verify the latest pact of any registered consumers main branch, so publishing a pact with a new expectation and associating its consumer version as master
(the consumers main branch) causes the verification step to fail, breaking the provider's build through no fault of its own.
Make a breaking change to the provider
The consumer is protected from deploying a change incompatible with the master
and deployed
provider, and the provider is protected from incoming changes from the consumer.
Let's see what happens if the provider makes a change that would break its contract with the production version of the consumer. The pact deployed to the prod
environment is not in pending mode, as it already has a successful verification result published. This means that a failure for the prod
pact verification will fail the overall verification task.
Make a change to the provider that would cause the production pact to fail.,
An easy way to do this is to open up
product/product.js
and changethis.id
tothis.uuid
.
Run
GIT_BRANCH=master make test
- note that this now fails ❌Discard your changes.
Expected state by the end of this step
A provider build that is passing and still deploying to production, with a failing master
pact in pending state.
Conclusion
Making changes to the pact on the consumers main branch can break both consumer and provider builds, and may stop both projects from being able to deploy.
Enabling the pending pacts feature for verifications stops changed pacts from breaking provider builds, allowing the provider to deploy to production if it is still compatible with the prod pact, while ensuring that the provider cannot make changes that break existing consumers.