How would you handle a scenario where a test environment is frequently unstable?