UK Microservices and Data Architecture Guide (2026)

CQRS (Command Query Responsibility Segregation): separate read and write models. For UK regulated software: (1) NHS patient record (writes: every clinical action -- PostgreSQL ACID, reads: patient dashboard -- Redis cache + read replica), (2) FCA payment ledger (writes: every payment transaction -- ACID PostgreSQL, reads: account balance dashboard -- read replica), (3) GDS benefit claims (writes: claim submission -- ACID, reads: case worker dashboard -- eventually consistent read model). CQRS is not needed for most UK regulated products. It adds significant complexity (two data models, eventual consistency handling, read model synchronisation). ClickMasters uses CQRS only when: (a) read patterns are significantly different from write patterns (high-read GDS dashboards with high-write NHS telemetry), (b) write model requires ACID guarantees but read model can tolerate eventual consistency, (c) read scaling needs differ from write scaling. ClickMasters CQRS projects: 8 of 120+ projects use CQRS -- mostly NHS population health platforms with high-read analytics.

Data mesh for UK HealthTech: data mesh (Zhamak Dehghani -- decentralised data ownership, domain-oriented data products) is a large-organisation pattern. For UK HealthTech: (1) applicable for NHS organisations with 20+ data domains (NHS Digital, NHS England -- at scale they are adopting data mesh principles), (2) not applicable for UK HealthTech startups or scaleups (< GBP50M ARR -- premature architectural complexity), (3) data mesh for NHS: NHS FHIR R4 UK Core is the semantic layer for NHS data products -- FHIR is NHS data mesh in practice. ClickMasters: for UK HealthTech startups, a well-structured PostgreSQL schema with domain-separated schemas (patients, appointments, prescriptions, billing as separate schemas) achieves 80% of data mesh benefits with 5% of the complexity. When to consider data mesh: NHS organisation with multiple product teams each owning NHS data domains and needing to share data between teams without creating data silos.