I hope there is a public postmortem that explains this. I have Cloud SQL instances with high availability enabled that have been down for over 5 hours with no workaround, when (according to docs) they should have failed over to the unaffected zones within minutes.