In seatbelt, all of the fallback handlers act on the output. I have a scenario in which my fallback behavior is to re-issue the failing request to a different endpoint. To do that, I need a clone of the original request.
At the moment, I am having to work around this with some ugly acrobatics in my service (copy each request to each output, for use in fallback).
Ideally, the request cloning would be limited to cases where an input-centric fallback action is in play. Output-centric fallback actions wouldn't incur that overhead.