DEV Community

Daniel Werner
Daniel Werner

Posted on • Originally published at 42coders.com on

Under the hood: How RefreshDatabase works in Laravel tests

Introduction

When writing test for any application it is crucial to every test run independently without affecting each other. If your tests use database, every test should run with a clean database. By clean I mean a known state, fresh migrated or seeded, not necessarily an empty one. Laravel offers an easy way to handle this using the RefreshDatabase trait. For more information about the usage of the trait please check the documentation here.

A test’s life cycle in Laravel

Before we dive into the details how refresh database works, let’s take a look at the test’s life cycle. Laravel ships with PHPUnit, so the life cycle of the tests is very similar to the PHPUnit tests life cycle. I won’t cover the details of the whole test running process, just stick to the points which are interesting for the current topic:

  • Before every test the setUp method is running:
    • It creates/refreshes the laravel application
    • Sets up traits
    • Calls the afterApplicationCreatedCallbacks, sets up events, clears facade’s resolved instances, etc.
  • The actual test is running
  • After every test the tearDown method is running:
    • It calls the beforeApplicationDestroyedCallbacks
    • Closes Mockery, resets variables etc.

The RefreshDatabase trait

If the test uses the RefreshDatabase trait, the setUpTraits calls the refreshDatabase() method from the trait, and the interesting part starts here. In tests you can use in memory and regular databases, depending on how you’ve set up the test environment, it will refresh the database accordingly.

The in memory database

In case of in memory database things are quite simple just migrates the database, it runs very fast, so it can be done before every test. Easy enough.

    protected function refreshInMemoryDatabase()
    {
        $this->artisan('migrate');

        $this->app[Kernel::class]->setArtisan(null);
    }

Regular database

In other cases it only migrates the database, if it has not been migrated e.g. before running the first test.

        if (! RefreshDatabaseState::$migrated) {
            $this->artisan('migrate:fresh', [
                '--drop-views' => $this->shouldDropViews(),
                '--drop-types' => $this->shouldDropTypes(),
            ]);

            $this->app[Kernel::class]->setArtisan(null);

            RefreshDatabaseState::$migrated = true;
        }

When the database has been migrated it starts a database transaction:

        foreach ($this->connectionsToTransact() as $name) {
            $connection = $database->connection($name);
            $dispatcher = $connection->getEventDispatcher();

            $connection->unsetEventDispatcher();
            $connection->beginTransaction();
            $connection->setEventDispatcher($dispatcher);
        }

and registers a beforeApplicationDestroyed callback where the transaction is rolled back. These callbacks are running in the tearDown method, so after every test. The rollback ensures that the database isn’t changed by the test, and the next test can run on a clean database.

        $this->beforeApplicationDestroyed(function () use ($database) {
            foreach ($this->connectionsToTransact() as $name) {
                $connection = $database->connection($name);
                $dispatcher = $connection->getEventDispatcher();

                $connection->unsetEventDispatcher();
                $connection->rollback();
                $connection->setEventDispatcher($dispatcher);
                $connection->disconnect();
            }
        });

Potential pitfalls

While the method described above works fine in most cases, I had some cases when I spent a couple of hours searching, debugging why my tests are failing and the database doesn’t get refreshed correctly.

As you might presume, the solution is NOT to commit the opened database transaction in tests, and avoid using statements which causes implicit commit in mysql. I made the second mistake by running an sql dump in one of my seeders, which created a table, and therefore implicitly committed the transaction. You can find out more about implicit commits in the mysql documentation. Self defense on: I know the database tables should be created with migrations, and seeders should be written with factories, but sometimes we need to do non optimal things :-).

Conclusion

The RefreshDatabase is a very useful feature when writing tests for Laravel application, just need to be careful with transaction commits in your tests.

Hope this article was useful, if you have any additions, remarks, please let me know in the comments section.

The post Under the hood: How RefreshDatabase works in Laravel tests appeared first on 42 Coders.

Top comments (5)

Collapse
 
uplink03 profile image
Uplink03

Implicit commits? Try this: I'm using transactions as explicitly. Look at my fancy code... I figured out why my seeds complain about duplicate entries by using Xdebug and walking through the damn trait myself. This should be documented in big bold letters, ffs.

Collapse
 
moay profile image
moay

Was just looking for a good wrap up of how the trait does the migrations, seeding and transactions for our trainees. This one is it. Thanks @daniel_werner

Collapse
 
daniel_werner profile image
Daniel Werner

I am glad you found my article useful!

Collapse
 
ericknyoto profile image
Erick Nyoto

I didn't understand. What do you mean by "avoid using statements which causes implicit commit in mysql"
Could you please give an example?

Thank you!

Collapse
 
benalidjamel profile image
benali djamel

awesome thank you