BookTribe

The problem

A book-lending social network that started as a monolithic Spring Boot and Angular application. I used it to learn how to split a system into microservices with Spring Cloud, one domain at a time, without rewriting it from scratch.

• A paginated catalog of the books shared by the community
• Publishing your own books with a cover, shareable or archived
• Borrowing and returning, with the return approved by the owner
• Ratings and reviews on the books you borrowed

Architecture

Spring Boot 3 microservices on Spring Cloud: every service reads its configuration from the Config Server and registers with Eureka, and the Angular frontend talks only to the gateway.

• Spring Cloud Gateway as the single entry point, with CORS configured there instead of in each service
• Services: books and loans, feedback with its own database, email notifications with Thymeleaf templates
• Authentication delegated to Keycloak: services validate the JWT as resource servers, the frontend uses keycloak-js with PKCE
• An Angular client generated with ng-openapi-gen from the merged specs of the two services
• Deployment prepared: multi-stage Dockerfiles, a production Docker Compose with Postgres and Keycloak, the frontend meant for Vercel and the backend for my server via Tailscale Funnel

Technical choices

• A three-phase migration, each step checked against a real stack: throwaway Postgres, Keycloak and MailDev, and real JWTs
• Feedback stores only the book id instead of a JPA foreign key, so the two domains have separate databases
• Service-to-service calls with OpenFeign and Resilience4j circuit breakers with fallbacks: if feedback is down the book list shows a rating of 0, if books are down feedback answers 503 with a clear message
• Configuration from environment variables with local defaults, plus an optional seeder with sample books, loans and reviews

What I learned

• Extracting a service surfaces bugs the monolith was hiding: an email template looked up by the wrong name in code that was never reached, and an ownership check that compared a string with an integer
• Identity does not travel between microservices on its own: the JWT was not forwarded on Feign calls until I added an interceptor
• Resilience has to be tested in the failure case: with a misnamed property the fallbacks never fired, and I only noticed by shutting a service down
• Microservices have a concrete cost: turning a join into a network call forces you to decide what to show when the other service does not answer, and five JVMs with Keycloak and Postgres take about 3 GB of RAM