- sync retries with wait_time_between_attempts
- make max_retries interpret the initial attempt not as a retry
- dont retry on 429, do retry on 503
I think this makes sense, especially since these gateway timeout errors are not going to resolve themselves within 10 seconds.
Do you know how to check if the underlying cause is correct? I couldn't figure it out when I was writing the test. Might be good to add here.