Here's an example of a database abstraction choice: the data model. Do you treat data as a series of normalized row-based data tables with joins optimized for

Here's an example of a database abstraction choice: the data model. Do you treat data as a series of normalized row-based data tables with joins optimized for strongly consistent OLTP? [PostgreSQL] Or as a denormalized row-based key-key-value single table optimized for eventually consistent OLTP? [ScyllaDB] Or as a denormalized column-based single table optimized for OLAP? [Pinot] Or as a denormalized key-value optimized for in-memory caching? [Redis] And so on for document stores, or property graphs, etc. Each of these design choices then gives way to the abstractions of implementation.