Currently if MusicBrainz is unavailable, HTTP 503 is returned and log messages are recorded and sent via email to subscribers. I think this case should be handled a bit differently.

We could catch errors that are returned when MusicBrainz is unavailable, return custom error message. Email notifications about that shouldn't be sent. There's a good chance that a lot of requests will fit in the time frame of unavailability and you wouldn't want to get hundreds of emails reporting the same error (especially when you can't do anything about that).

Even better way of handling this would be some kind of custom logging system that will allow us to track special cases, analyze them and make improvements based on that.