For the API testing, we are using an independent API server for each test because:

• We want tests to be isolated.
• Sometimes, we need a specific tracker configuration for the test.

We assign a random port for each API server and test with a function like this:

fn random_port() -> u16 {
    // todo: this may produce random test failures because two tests can try to bind the same port.
    // We could create a pool of available ports (with read/write lock)
    let mut rng = thread_rng();
    rng.gen_range(49152..65535)
}

We have reached a high number of parallel running tests that produces this conflict too often.

We could create a centralised free-of-race-conditions service which assigns the free port numbers.

We could also reuse API server for tests that use the same configuration and do not conflict, but I would like to avoid that because it can generate hidden coupling between tests.