Retrying a failed request once is smart. When every client retries instantly and all at once, a brief hiccup becomes a long outage.