So that depends on what's happening.

If it's host maintenance (the machine your database is on needs to be brought down) then you should "see" the same results as Google Compute Engine live migration, which is that you hopefully see nothing :) This was one of the most common reasons a first gen instance would require restarting, alongside us pushing updates. So nixing that one, alongside the other non-restarting updates we can do now, is good stuff.

If a whole zone goes down, we start the failover process automatically for you. This is order minutes. It won't be seamless, but zones going down is hopefully a very rare occurrence.