Multi-region consistency: cache, reads, and writes

Not every request needs global writes

Most products can keep a single source of truth while distributing reads, cache, and stateless computation. Start by identifying which operations truly require synchronous global coordination.

Three practical patterns

• Single-writer: writes go to one primary data region; read replicas and edge caches serve global traffic.
• Read-local, write-central: applications read from a nearby replica but route mutations to the primary.
• Conflict-aware multi-writer: several regions accept writes and the application defines conflict resolution explicitly.

Cache is not a database

Treat edge cache as derived state. A cache can speed a read path but must be safe to discard at any time. Persistent business state belongs in a durable data system with a documented consistency model.

Operational checklist

• Document the authoritative source for each entity.
• Test stale-read scenarios, not only the happy path.
• Measure replication lag as a first-class metric.
• Define the behavior when a region cannot reach the primary.