Tests can be run on a CI server as part of a build in the same way as you would run other tests. However, there are a few things to consider:
The application must be deployed to a server and the server started.
The URL used in the test must match the URL of the deployed application.
The browsers you want to run on must be available.
If run in parallel, tests should be truly independent of each other.
You need to install a license file on the CI server.
Deployment of the application can be done in several ways, depending on the setup and your preferences. The important thing is that the application is deployed and ready to accept requests before the tests are started.
For applications without external dependencies, it’s often handy to start a test as part of the build. If you’re building with Maven, see Running Tests with Maven for information on how you can start and stop the server as part of the build.
If you have an external server on which you deploy the application, you would typically copy the result of the build to that server in one build step. You would then wait for the deployment to finish by querying the server or polling the URL where the application should be. The following build step then executes the TestBench tests using the predefined URL.
In tests, you typically use a URL like
http://localhost:8080/ when running on a local machine. On a build server, this is usually fine if you’re running the server and tests on the same build agent.
If only the server is running on the build agent, and the browsers are running on a separate machine or on a cloud-based browser provider, you might need to define and use a public IP of the build agent. You either need to pass the IP address to the build in some way and use it in your test, or you have to use the provided
IPAddress.findSiteLocalAddress() helper in your test.
Below is an example of how you might use the helper:
getDriver().get("http://" + IPAddress.findSiteLocalAddress() + ":8080/");
If you’re deploying on another host name, you need to pass that information to the tests in a suitable way. For example, you might do this as a system property or environment variable that’s read in the test code.
If you’re not using site local addresses (i.e., 10.x.y.z, 172.16.x.y or 192.168.x.y), you can use instead
When running the tests on your local machine, you need to have a suitable browser installed. For instance, if the test creates a
ChromeDriver instance, you need to have Chrome installed, and so on.
The same goes for the CI server, the build agent, if you’re running tests directly on a local browser — as opposed to a test cluster, described in Running Tests on Multiple Browsers.
In addition to installing the browsers on the build agent, you must consider the fact that browsers typically require a GUI to be run. This isn’t available directly on your typical build system.
The options for running on such a system are:
Run Chrome or Firefox in headless mode; then no GUI is needed.
If it’s a Linux-based system, start
xvfb, which provides a GUI environment for the browser without actually showing the GUI environment anywhere.
To run Chrome in headless mode, you need to pass the
--headless parameter — and
--disable-gpu on Windows — to the
ChromeDriver when starting the browser. The parameters can be defined using
ChromeOptions like so:
ChromeOptions options = new ChromeOptions(); options.addArguments("--headless", "--disable-gpu"); setDriver(new ChromeDriver(options));
Similarly, to run Firefox in headless mode, you need to pass the
--headless parameter to the
FirefoxOptions options = new FirefoxOptions(); options.addArguments("--headless", "--disable-gpu"); setDriver(new FirefoxDriver(options));
|Previously, PhantomJS was recommended as a good way to do headless testing. You should no longer use PhantomJS, as it has fallen far behind the latest browser versions and likely won’t work with Vaadin.|
The easiest way to ensure that tests don’t interfere with each other is to have a separate test database initialized for each test. How you do this is typically dependent on the stack you’re using. If you’re using, for example, Spring Boot, you can use the
SpringRunner and set your test to rollback any transaction at the end of the test. Other environments might have different options for this.
If you aren’t resetting the database for each test, you should typically not run the tests in parallel, as it’s difficult to understand where something went wrong when several tests fail suddenly. Even though it’s not good practice, when running tests in sequence, you can consider the way in which the previous test modified the data set.
A better approach typically is to try not to alter the global state in the test — or at least include the data needed by the test in the test itself. An example would be that when you test a CRUD view, you should start by creating an entity instead of selecting an existing entity at random. You can then delete your test entity at the end of the test.
Both of these approaches, however, cause several tests to fail if one test fails at the beginning of the set. You always need to search for the initial problem and then rerun the whole set to find any additional errors.