I recently listened to Marc Brooker's talk [1] on retries. My key takeaway: You don't want to put more load on your system during stress. At AWS, they do "adaptive retries", which I found interesting.
I second this, retry mechanisms can cause retry storms in vast enough, distributed systems. In Amazon code bases I found the same adaptive retry strategy after a Christmas where once we played whack-a-mole for a service to get back up as its clients kept retrying.