The instance manager was not actually checking whether an instance
  was actually running before trying to stop it. Now it checks first.