You get multiple ones and failover to them or you schedule a maintenance window. Yes, it'll be more expensive, but you probably already have a good reason & the associated budget if you colocate as opposed to renting from Hetzner/OVH/etc.
Is this so different from the cloud? Even an EC2 instance should be assumed to be a single point of failure and you also need multiple (preferably in different AZs or regions) if uptime is important.
Not sure what you are talking about here but for example an RDS instance is simply an EC2 instance with a bunch of automation.
If you want to save some money write the scripts yourself and run them in a cron job. Took me a couple hours and I'm no sysadmin.
To me the best thing to do is move all your services into k8s and automate you DB (use open source). Then you can move to any provider when you realise the cloud pricing is shit.
