CQRS Is Not Two Databases
CQRS is often introduced as a picture with two databases and a message bus. That picture is one possible implementation, not the definition.
The useful idea is that writes and reads have different concerns. A write enforces invariants and changes state. A read answers a question efficiently. You can separate those models in code long before adding infrastructure.
For instance, a checkout command may require strict validation around available credit. A reporting query may join several read-only tables and return a flat projection. Making both travel through the same generic repository because "everything is an entity" can be needlessly restrictive.
HTTP request
|-- command --> invariants --> persistence
|-- query --> projection --> response
Separate databases become interesting when read scale, data shapes, or availability needs justify eventual consistency. They also introduce lag, replay, operational tooling, and new failure states. A dashboard showing stale information may be fine. A payment authorization using stale limits probably is not.
Start by separating intent. Add physical separation only after measuring a real need. Architecture diagrams are cheaper than operating distributed state.
Top comments (0)