> "Hubs SHOULD retry notifications repeatedly until successful". Where > successful is defined as "an HTTP success (2xx) code". > > If a client is not coping for some reason, perhaps returning 503s or > not even able to respond at all, the recommended course of action is > to flood their callback with even more data.
You dropped the end of that sentence, James: "Hubs SHOULD retry notifications repeatedly until successful (up to some reasonable maximum over a reasonable time period)." Note that SHOULD also implies that the Hub can use it's own policy to determine how it wants to do retries. I could make this language stronger by saying hubs SHOULD also use error-rate detection to prevent hammering callback URLs that are obviously broken. Does that sound like useful guidance? I wonder if that should be part of the spec or just part of the best practices, non-normative part of the document. -Brett
