This part of the documentation covers support for reactive-stack web applications built on a
Reactive Streams API to run on non-blocking
servers, such as Netty, Undertow, and Servlet 3.1+ containers. Individual chapters cover
the Spring WebFlux framework,
the reactive WebClient, support for testing,
and reactive libraries. For Servlet-stack web applications, see
Web on Servlet Stack.
The spring-test module provides mock implementations of ServerHttpRequest,
ServerHttpResponse, and ServerWebExchange.
See Spring Web Reactive for a discussion of mock objects.
WebTestClient builds on these mock request and
response objects to provide support for testing WebFlux applications without an HTTP
server. You can use the WebTestClient for end-to-end integration tests, too.
spring-webflux depends on reactor-core and uses it internally to compose asynchronous
logic and to provide Reactive Streams support. Generally, WebFlux APIs return Flux or
Mono (since those are used internally) and leniently accept any Reactive Streams
Publisher implementation as input. The use of Flux versus Mono is important, because it
helps to express cardinality — for example, whether a single or multiple asynchronous values are
expected, and that can be essential for making decisions (for example, when encoding or
decoding HTTP messages).
For annotated controllers, WebFlux transparently adapts to the reactive library chosen by
the application. This is done with the help of the
ReactiveAdapterRegistry, which
provides pluggable support for reactive library and other asynchronous types. The registry
has built-in support for RxJava and CompletableFuture, but you can register others, too.
For functional APIs (such as [webflux-fn], the WebClient, and others), the general rules
for WebFlux APIs apply — Flux and Mono as return values and a Reactive Streams
Publisher as input. When a Publisher, whether custom or from another reactive library,
is provided, it can be treated only as a stream with unknown semantics (0..N). If, however,
the semantics are known, you can wrap it with Flux or Mono.from(Publisher) instead
of passing the raw Publisher.
For example, given a Publisher that is not a Mono, the Jackson JSON message writer
expects multiple values. If the media type implies an infinite stream (for example,
application/json+stream), values are written and flushed individually. Otherwise,
values are buffered into a list and rendered as a JSON array.