Advanced: Replication, Consistency, CAP

Partitions are inevitable, so the real CAP choice is CP (refuse rather than be wrong) versus AP (stay up and reconcile later).

Single-primary for most apps (simple, read scaling); multi-primary only when you truly need multi-region writes (and accept conflict resolution); cascading to offload fan-out from the primary.

Physical replication is an exact binary copy of the whole cluster on the same engine; logical replication is selective row-level change events that can cross versions and engines.

Synchronous trades latency for zero data loss; asynchronous trades a small data-loss window for speed.

Start from "what must never happen": hard invariants demand strong consistency, cosmetic data permits eventual consistency; let requirements choose the technology.