As Yunong mentioned, we were doing some upgrades to the metadata tier, and the errors are expected for that kind of upgrade. Sorry for the disruption, and especially for not communicating that ahead of time. We're working internally on improving that. (Literally — we're talking about how best to handle this now.) The vast majority of our upgrades aren't noticeable from the outside, and most of the time when it is, the period of disruption has been so been transient that people haven't really noticed, but obviously we can and should do better around communicating these.
The other piece of this is that we pitch 500s instead of blocking when things are transitioning internally. We do this to provide better visibility into the underlying service's state — IME, it's much worse to have requests black-hole for minutes at a time than to fail frankly and fast. But to deal with this, it's recommended that clients use retries with backoff (especially 503s).