Exploring Java Microservices: Current Situation, Choices, and Future Directions with Service Mesh

Microservices are an architectural style in which a large, complex application is composed of one or more loosely coupled services that can be deployed independently; the concept was popularized by Martin Fowler in March 2014.

Huajiao currently runs many services (H5 website, app back‑ends, live streaming, economy system, cloud control, etc.) using PHP. While PHP offers rapid development and leverages existing team expertise and LNMP optimization, the ecosystem suffers from high coupling between governance and business code, duplicated error codes, costly upgrades, slow scaling, and performance limits imposed by the synchronous LNMP stack.

Several Java‑based microservice solutions are surveyed: Alibaba’s Dubbo (now Apache) and Spring Cloud Alibaba, Tencent’s Tars, Huawei’s ServiceComb, and the ubiquitous Spring ecosystem (Spring Boot, Spring Cloud). These frameworks provide service registration, discovery, configuration, and governance capabilities.

The team selects Spring Boot 2 as the foundational framework because of its maturity, large community, annotation‑driven configuration, ability to package applications as standalone JARs, seamless integration with Spring Cloud components, built‑in Actuator health monitoring, and easy inclusion of third‑party HTTP/gRPC clients.

For data access, MyBatis is chosen over Hibernate for its lightweight flexibility and the ability to generate models via MyBatis‑Generator. Undertow is preferred as the web container for its superior performance and lighter footprint compared with Tomcat.

Development standards are enforced using Alibaba’s Java Coding Guidelines plugin to ensure consistent code style and early detection of potential issues. springboot2 + mybatis + undertow packaged as a jar Incremental improvements include migrating CI/CD from Jenkins to GitLab (enabling code review, automatic build‑test‑deploy pipelines, and Kubernetes integration), adopting Eureka or Nacos for service registration (considering CAP trade‑offs: Zookeeper is CP, Eureka is AP, Nacos offers a balanced approach), and adding OpenFeign with Hystrix for lightweight REST clients, MDC‑based trace IDs for observability, and annotation‑driven Redis mechanisms for concurrency control and idempotency.

Future plans outline the deployment of core microservice components: an external gateway, a service registry, a configuration center, logging, governance, and containerization with Kubernetes (Wayne+k8s). Architectural diagrams are provided via the following images:

Service Mesh is introduced as the next‑generation infrastructure layer that abstracts inter‑service communication through sidecar proxies, allowing applications to remain unaware of networking concerns. Its key characteristics include being a communication middle‑layer, lightweight proxy, application‑agnostic, and decoupling retries, timeouts, monitoring, tracing, and discovery. First‑generation meshes (Linkerd, Envoy) and second‑generation meshes (Istio, Conduit) are mentioned, along with industry adoption by Ant Financial, Huawei, and Tencent.

Reference links for further reading are listed, covering Spring Cloud, Service Mesh concepts, Istio documentation, and curated Service Mesh resources.