從需求到微服務:以領域驅動設計與機器學習為方法

Abstract

微服務架構已成為開發應用程式的熱門選擇，因其具有簡潔、可獨立部署和可溝通的服務特性。然而,儘管有指導設計過程的實際範例與原則，但並沒有明確的規則來說明如何根據需求來設計微服務。針對此議題，領域驅動設計(DDD)因提供將領域分解的方法而變得相關，但在實際規劃微服務架構時仍會面臨許多挑戰。這些困難包括直接從領域模型生成微服務或決定微服務的適當粒度。為了應對這些挑戰，在此篇研究中，我們提出了一個新穎的兩階段流程。該流程不考慮領域模型(domain model)，而是利用 DDD 和 event storming 的原則來定義子領域(subdomain)邊界。在子領域的邊界決定之後，我們利用配對機制將這些子領域與潛在的微服務候選者進行映射，以此確保配對是多元並具有彈性的。最後我們使用機器學習來自動化此流程，進而提升系統的效率。

The microservices architecture has emerged as a popular choice for developing applications as compact, independently deployable, and conversational services. While there exist proven practices and principles that can guide the design process, there are no definitive rules that dictate how to design microservices based on requirements. Domain-driven design (DDD) has gained relevance as it provides a means for the decomposition of domains into contexts, but challenges arise when applying DDD to plan microservices architecture. Difficulties include deriving microservices directly from domain models or determining the appropriate granularity of microservices. To cope with these challenges, in this work, we propose a novel two-stage process that disregards domain models in the process. It leverages DDD and event-storming principles to define subdomain boundaries. Subsequently, a sophisticated matchmaking mechanism connects these subdomains with potential microservice candidates, offering versatile and flexible mapping. The process is automated using machine learning, enhancing its effectiveness.