How should Database per Service be used in production for Microservices Architecture?

Keep data ownership explicit, expose APIs or events for sharing data, use read models when needed, and avoid distributed joins across service databases.