I would add some random jitter to the timing of the backoff. If the fail is due to two such processes getting themselves into some sort of deadlock and they bot

I would add some random jitter to the timing of the backoff. If the fail is due to two such processes getting themselves into some sort of deadlock and they both fail at the same time, they might retry at the same time if at the same part of the retry cycle and fail again for the same reason.