Absorbing Screen-Level Complexity: The Record of Yeogiyo's BFF

Yeogiyo operates an MSA structure separated by domain—orders, search, coupons, reviews, flights, payments, and more—but from a screen perspective, the problem of having to combine multiple APIs for a single page grew larger. As of December 2025, the number of services registered in the B2C Gateway reached about 50-plus, and clients had to directly gather data such as search results, reviews, coupons, points, and order status.

The solution that emerged here was BFF (Backend For Frontend). BFF is defined not as a simple routing layer, but as a UI-Driven server layer that Aggregates multiple domain APIs to fit the data structure required by a specific screen, converts domain models into ViewModels, and absorbs screen-level business branching and parallel calls.

In the existing structure, clients called each domain API directly through the Gateway, taking on all the mashup, data combination, and branching logic themselves. After introducing BFF, the client only calls a single BFF API, and internally within the Private Zone, the BFF quickly gathers the necessary domain data and returns a refined response.

As real-world application cases, the Partner Center and Ad Center are introduced. Constructing a single order detail screen required 7~8 APIs—order information, payment information, affiliate store information, remaining balance history, card/virtual account history, additional order information, and more—and whether to display the CTA also required comprehensively judging conditions such as payment status, contract agreement, payment method, and suspension status. For automatic transfers, withdrawal request information had to be additionally queried, so the response model itself was separated out.

As a result, BFF moved screen-level complexity to the server, leading to elimination of multiple API calls, simplification of client logic, centralization of state branching logic, and improved response speed to changes. However, as the number of intermediate layers increases, issues such as cascading failures and memory/heap overload also arise, which required operational mechanisms such as exponential backoff based on Spring Retry, fast failure/fallback responses, review of Circuit Breaker, forced paging, and Caffeine cache.

The key point is that while MSA is a structure that finely divides domains, BFF is a layer that re-absorbs that complexity in units of 'a single screen' that the user actually faces.