The article explains how to apply hexagonal architecture to Apache Flink microservices, specifically through a dynamic pricing microservice example. It highlights the pitfalls of embedding business logic directly into Flink's ProcessFunctions, which can lead to messy code. Instead, it advocates for a clean separation of concerns by first identifying the core domain questions without getting bogged down by implementation details like Flink's specific features. Key steps involve distilling the domain questions, designing a domain layer without any Flink dependencies, and defining interfaces (or "ports") that reflect the domain's needs. The author presents several core records, such as `Product`, `Money`, and `DemandMetrics`, that encapsulate essential data structures. The `PricingEngineService` class then implements the pricing logic, fetching necessary inputs via repositories and applying business rules to compute a final price, all while maintaining a clear separation from Flink. Unit testing is emphasized, with a focus on test-driven development (TDD). Since the `PricingEngineService` interacts solely with interfaces, it can be effectively tested with tools like Mockito. This approach not only ensures that the pricing logic is robust but also keeps the code modular and easier to manage. By the end, the article illustrates how a well-structured domain service can simplify the integration of complex business logic into a Flink application.