I mean, sometimes handling an API failure correctly is failing over. If an API call actually matters, then when the API stops responding, throw an error. Even if you're probably catching that and giving out an error message, I'd still call that a production outage.