Comment by crazygringo

Sure, but replication and automatic failover is a huge pain to configure. It's a gigantic step in architecture complexity, requiring multiple database servers.

I'm not saying it can't be done. But block storage is built for reliability in a way that ephemeral instances are not. There's a good reason why every guide will tell you to set your database up on block storage rather than an instance's local disk. If your instance fails, just spin up another instantly and reconnect to the same block storage.

Pre-cloud, the equivalent would have been using redundant RAID storage to handle disk failures (easy), before upgrading to replication with an always-running replica (harder).