Iโm not certain, but I think this is a combination of two things. (Evidence another client gives the actual error which is timeout waiting for a slot)
The instance having difficulty coping with the API calls when busy and the API not having a graceful fallback.