Stretched cluster hypervisor stack?

Most cluster solutions are built for a single network switch (well, single datacentre) latencies, but there's little benefit in having e.g. 5 nodes in a single location which can experience power event or simply just get cut off the rest of the Internet - those services are not highly available worldwide.

Save for building everything from scratch on top of cloud providers' bespoke APIs, what stretched-cluster-like deployments do you do to have multi-site redundancy with as limited downtime as possible?

https://redd.it/1x2mfba
@r_SelfHosted