Let nginx do retry on other upstream if fail or not?

Is it good in general to let nginx do retry on other upstream if one fail, timeout and etc?
There is a setting "proxy_next_upstream" and "proxy_next_upstream_tries" for configuring that, and even if we can configure the behavior for "non_idempotent" to not be retried, but is it good to assume that no one will use a GET request for non_idempotent behavior?