界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
开发与工程 · SKILL.md

webflux-reactive-patterns

webflux-reactive-patterns

【开发与工程】Use when the complete Spring Boot 3 request path is reactive: WebFlux controllers, Reactor pipelines, R2DBC persistence, streaming, backpressure, or cancellation. For defining and registering declarative outbound API interfaces, use http-interface-clients.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: webflux-reactive-patterns description: > Use when the complete Spring Boot 3 request path is reactive: WebFlux controllers, Reactor pipelines, R2DBC persistence, streaming, backpressure, or cancellation. For defining and registering declarative outbound API interfaces, use http-interface-clients.

WebFlux Reactive Patterns

Use WebFlux when the complete request path can benefit from non-blocking I/O.

Pipeline rules

  • Return Mono or Flux; do not call subscribe() in request-handling code.
  • Compose work with map, flatMap, concatMap, and zip according to ordering requirements.
  • Keep blocking JDBC, filesystem, and legacy SDK calls out of event-loop threads.
  • Isolate unavoidable blocking calls on boundedElastic at the narrow adapter boundary.
  • Apply timeouts at remote-call boundaries and preserve cancellation.

Context and errors

  • Put request-scoped metadata in Reactor Context, not ThreadLocal.
  • Translate domain failures centrally without swallowing cancellation or infrastructure errors.
  • Use onErrorResume only when a defined fallback exists.
  • Avoid onErrorContinue; it makes partial processing hard to reason about.

Persistence and streaming

  • Use R2DBC for reactive database access; JDBC makes the pipeline blocking.
  • Bound concurrency for fan-out operations.
  • Define backpressure and buffering limits for streaming endpoints.
  • Avoid collecting an unbounded Flux into memory.

Testing

  • Use StepVerifier for publisher behavior and WebTestClient for HTTP contracts.
  • Test cancellation, timeout, empty results, errors, ordering, and backpressure.
  • Enable blocking-call detection in tests when the project supports it.

Examples

  • See examples/good-reactive-service.java and examples/bad-reactive-service.java.

Gotchas

  • Agent calls block() in a controller or service - keep the full request path non-blocking.
  • Agent calls subscribe() manually - the web runtime owns subscription.
  • Agent stores tenant or security data in ThreadLocal - use Reactor Context.
  • Agent uses unbounded flatMap - set concurrency according to downstream capacity.
  • Agent wraps JDBC in Mono.just - that still blocks the caller thread.
我的收藏