We need to be careful here. The approach described here does not solve the issue when you have more than one kernel. The problem then still is that the KernelTestCase class keeps a reference to the previously created kernel in its static $kernel property. Thus, if your functional tests do not run in isolated processes a later run test for a different kernel will reuse the previously created instance which points to a different kernel.

Thanks for the explanation! You are right and this is critical. What would be the best (and simplest) solution to this problem?

It should be sufficient to always reset the static $class property inside the tearDown() method:

protected function tearDown()
{
    parent::tearDown();

    static::$class = null;
}

The parent::tearDown() call is important to make sure that the kernel is shut down after the test was run.

Thanks!! I've fixed the code and reused your original explanation because it's great.