Storage is already abstracted via Kubernetes with persistent volume claims and CSI (container storage interface)

You can specify the default storage class to use the provider's CSI driver. The PVC is then specified and consumed the same way in a portable manner across providers.

With "storage" I'm thinking something more powerful beyond plain files, e.g. relational databases.

Block storage? When you create your persistent volume claim, you can specify an accessMode. "readWriteOnce" will target block storage of the storageClass, "readWriteMany" will use file.

You can also choose something specific like a mounted hostVolume which is not provider specifc.

Cutting edge stuff like NVMe-oF is even possible, the CSI ecosystem is very active. https://github.com/spdk/spdk-csi

We run almost all of our production databases on k8s - postgres, clickhouse, cockroachdb and more. Many others do too. K8s itself already comes with a db of sorts - etcd so it’s not a non-standard usecase.

planetscale (if I need rdbms) or firestore or dynamodb (if I need just document). I checked latency of planetscale and dynamodb from frankfurt,paris,singapore from DO/linode/scaleway to db in same region; these are low latency (mostly single digit).

For single server, I'd just use sqlite.