I am a big SQL fan but not all data has to be relational.

Let’s say I want the GPS coordinates of ten million vehicles every 5 seconds. I have a vehicle id, a timestamp, and coordinates. I do not care if a few writes get lost. Why does this have to be relational?

And perhaps I also record other info that may change from vehicle to vehicle. Perhaps just values that are true if present. DoorOpen, BrakeApplied, LightsOn, LightBarOn, EngineOn, etc. I may only be displaying this data in a UI. I may get different values from every vehicle or even every write. There is no “schema”. I mean, I can have a JSON field or something in my relational table? But this is not exactly relational anymore.

can't cache my notes because I forgot where I put my notebook

Also blockchain is append only. So no deletions.

Confidentiality, regulatory compliance, schema modeling (IDLs?), query load balancing (rpc infra) and operational tooling pretty good in my world now (Solana). Throughout and latency decent now (400ms conf, couple second finality, 4k tps) but obviously the type of system that needs a trustless immutable database has tradeoffs.

How easily can you write a crud system? No clue that’s on you. A line of business service? Why the hell would that need a trustless immutable ledger. You’re barking up the wrong tree. Low level financial systems are pretty much the only thing that benefits. Small niche for proof of existence perhaps. And maybe other immutable use case things.

Unfortunately deleted means deleted. Data gone.

My recommendation is to use it as a reverse proxy to expose services but ultimately host from another device.

It's 1 gb of ram anyway (the free 24 gb of ram arm vps' are scarce and hard to get).

Another tip is that they monitor cpu and ram and are more likely to delete it if is idling. There exist programs on github that do nothing but waste cpu in order to try to keep the oracle vps up.