We were alerted to spiking error responses in the registry as a result of database connectivity, the platform failed over to a high availability standby and resumed normal operation in ~10 minutes.

The CDN cache should retain stale versions of pages when error responses occur, but some uncached requests would have experience errors during this time.