【RSocket】使用 RSocket（二）——四种通信模式实践

Source Code: https://github.com/joexu01/rsocket-demo

0. 四种通信模式

让我们来简单复习一下 RSocket 的四种通信模式：

• 即发即忘 - FireAndForget：立即发送一个请求，无需为这个请求发送响应报文。适用于监控埋点，日志上报等，这种场景下无需回执，丢失几个请求无伤大雅

• 请求响应 - RequestResponse：请求方发送一条请求消息，响应方收到请求后并返回一条响应消息。传统的HTTP是典型的Request-Response

• 流式响应 - RequestStream：请求方发送一个请求报文，响应方发回N个响应报文。传统的MQ是典型的RequestStream

• 双向通道 - Channel：创建一个通道上下文，双方可以互相发送消息。IM是个典型的RequestChannel通讯场景

1. 客户端生成和解析路由信息

*本篇文章的客户端示例文件在 rsocket-client-raw/src/main/java/org/example/FourCommunicationScheme.java

我们使用 decodeRoute 和 encodeRoute 函数来解码和编码路由信息。

static String decodeRoute(ByteBuf metadata) {
        final RoutingMetadata routingMetadata = new RoutingMetadata(metadata);

        return routingMetadata.iterator().next();
}

static ByteBuf encodeRoute(String route) {
        return TaggingMetadataCodec.createTaggingContent(
                ByteBufAllocator.DEFAULT,
                Collections.singletonList(route));
}

2. RequestResponse

服务端处理函数

在这里我们编写一个简单的 Handler，它的 Route 是 test.echo，它接收一个请求并返回请求 Payload 的 data 中的字符串。

@MessageMapping("test.echo")
public Mono<String> simplyEcho(String data) throws InterruptedException {
    Thread.sleep(1500);
    logger.info("[test.echo]Received echo string from client: {}", data);
    return Mono.just(String.format("[test.echo]I received your string: %s. Thank you.", data));
}

注意，这里的参数也可以是 Mono<String> ，然后对 Mono 进行操作并返回。事实上，如果严格按照响应式编程的策略，这里应该直接对 Mono 进行操作。

客户端发送请求

1. 生成 metadata 的 route 信息，然后将字符串和 metadata 放入 Payload

ByteBuf routeMetadata = encodeRoute("test.echo");
Payload echoPayload = ByteBufPayload.create(
        ByteBufUtil.writeUtf8(ByteBufAllocator.DEFAULT, "This is a message from client using rsocket-java library."),
        routeMetadata);

2. 新建一个 RequestResponse，并对这次请求做一些设置；可以看到 RequestResponse 方法返回的数据类型是 Mono<Payload>。然后我们对这个 Mono 设定一些操作（具体操作请看代码注释）：

Mono<Payload> requestResponse = socket.requestResponse(echoPayload);
requestResponse
        // 当 subscribe() 操作开始执行时打印一下日志
        .doOnSubscribe(subscription -> logger.info("Test1 subscribed to {}", subscription.toString()))
        // 当携带的请求成功后要做的事情
        .doOnSuccess(payload -> {
            logger.info("Test1 - Successfully returned: {}", payload.getDataUtf8());
            payload.release();
        })
        .doOnError(throwable -> logger.info("Test1 doOnError: {}", throwable.toString()))
        // 可以使用 timeout 丢弃等待超时的 Mono
        //.timeout(Duration.ofSeconds(1))
        // 可以使用 doOnTerminate 在请求结束后做一些工作
        // .doOnTerminate(() -> {})
        // 但是一定要设置 doOnError
        //.doOnError(TimeoutException.class, e -> logger.info("Test1 doOnError: {}", e.toString()))
        // .onErrorReturn(TimeoutException.class, DefaultPayload.create("Payload: Test1 - timeout"))
        // 可以使用 log() 来观察数据的状态
        //.log()
        // 客户端在执行 subscribe() 操作时才会开始从服务端接收数据流
        // 在响应式编程中使用 subscribe 操作符是订阅一个数据流并处理发布的数据、错误和完成信号的核心方式之一
        .subscribe();

请求发出后主线程不会阻塞，所以我们需要使用 socket.onClose().block(); 保持连接。

然后我们尝试运行服务端和客户端，看看一看客户端的输出：

[main] INFO org.example.RSocketClientRaw - My UUID is 0718ef3b-9ee0-42f1-9003-700a8aa9a98d
[main] INFO org.example.RSocketClientRaw - Test1 subscribed to RequestResponseRequesterMono
[reactor-tcp-epoll-2] INFO org.example.RSocketClientRaw - Test1 - Successfully returned: [test.echo]I received your string: This is a message from client using rsocket-java library.. Thank you.

服务端日志：

2023-03-12 21:47:29.291  INFO 32099 --- [or-http-epoll-2] o.example.controller.RSocketController   : [connect.setup]Client connection: 0718ef3b-9ee0-42f1-9003-700a8aa9a98d
2023-03-12 21:47:32.304  INFO 32099 --- [or-http-epoll-2] o.example.controller.RSocketController   : [test.echo]Received echo string from client: This is a message from client using rsocket-java library.

客户端成功地发出请求并收到来自服务端的回复。

3. FireAndForget

服务端

@MessageMapping("upload.log")
public void fireAndForgetHandler(@Headers Map<String, Object> header, RSocketRequester requester, String data) {
    header.forEach((k, v) -> System.out.printf("[upload.log]header key: %s, val: %s\n", k, v));
    System.out.printf("[upload.log]UploadEventLogs: Received log string from client: %s\n", data);
}

服务端接受一个请求，不返回任何结果（Fire'n'Forget），只在服务端打印 Header 的内容。

客户端

// 测试 FnF
routeMetadata = TaggingMetadataCodec.createTaggingContent(ByteBufAllocator.DEFAULT, Collections.singletonList("upload.log"));
socket.fireAndForget(
    ByteBufPayload.create(
    ByteBufUtil.writeUtf8(ByteBufAllocator.DEFAULT, "This is a log from client using rsocket-java library."),
        routeMetadata))
    .doOnSubscribe(subscription -> logger.info("Test2 - Fire And Forget onSubscribe: {}", subscription.toString()))
    .subscribe();

客户端输出：

[main] INFO org.example.RSocketClientRaw - Test2 - Fire And Forget onSubscribe: FireAndForgetRequesterMono

服务端输出：

2023-03-10 15:10:25.675  INFO 5318 --- [or-http-epoll-4] o.example.controller.RSocketController   : [test.echo]Received echo string from client: This is a message from client using rsocket-java library.
[upload.log]header key: dataBufferFactory, val: NettyDataBufferFactory (PooledByteBufAllocator(directByDefault: true))
[upload.log]header key: rsocketRequester, val: org.springframework.messaging.rsocket.DefaultRSocketRequester@607cc59
[upload.log]header key: lookupDestination, val: upload.log
[upload.log]header key: contentType, val: application/binary
[upload.log]header key: rsocketFrameType, val: REQUEST_FNF
[upload.log]UploadEventLogs: Received log string from client: This is a log from client using rsocket-java library.

4. RequestStream

服务端

服务端接收一个 Mono<String> 然后返回给客户端包含 10 个 String 的 Flux。

事实上，严格按照响应式编程的策略，这里应该直接对 Mono 进行操作，可以使用 flatMapMany() 把生成的数据流通过异步方式处理，扩展出新的数据流。下面是扩展新数据流的简单示例：

Mono.just(3)
    .flatMapMany(i -> Flux.range(0, i))
    .subscribe(System.out::println);

在这里为了演示方便就先打印 Mono 然后新生成一个 Flux。

@MessageMapping("handler.request.stream")
public Flux<String> responseStreaming(Mono<String> request) {
    request
            .doOnNext(s -> logger.info("[handler.request.stream]: {}", s))
            // 可以使用 then() 结束操作链
            .then()
            .subscribe();

    return Flux
            .range(1, 10)
            .map(idx -> String.format("Resp from Server: %s, Thank you!", idx));
}

客户端

请看代码注释来理解对数据流 Flux 的各种操作：

// 测试 RequestStream
routeMetadata = encodeRoute("handler.request.stream");
Flux<Payload> requestStream = socket.requestStream(
        ByteBufPayload.create(
                ByteBufUtil.writeUtf8(ByteBufAllocator.DEFAULT, "TEST3 - Request&Stream"),
                routeMetadata));

requestStream
        // 当然可以使用 map 对每个 Payload 进行操作，这会改变数据
        // .map(payload -> System.out.printf("%s\n", payload.getDataUtf8()))
        .doOnSubscribe(subscription -> logger.info("Test3 subscribed to {}", subscription.toString()))
        // 使用 doOnNext 不会对流的数据进行改变
        // doOnNext()是一个 Reactor 式流操作符，它允许编写者注册一个在每次出现新元素时执行的回调函数
        .doOnNext(nextPayload -> System.out.println("Test3 Received payload: " + nextPayload.getDataUtf8()))
        // 当需要从流中选择一些特定的元素时，可以使用 Flux.take(long n) 操作符
        // 该操作符将创建一个新的 Flux，该 Flux 包含原始 Flux 的前 n 个元素
        // take 操作符发出了指定数量的元素之后，就不再接收任何元素，并且将取消其上游发布者的订阅
        // 在这里服务端使用 Flux.range 来限定 Flux 流中的元素个数
        // 如果服务端使用 Flux.interval 生成一个无限长度的流，客户端使用 take 接收限定个数的元素
        // 便会取消发布者的订阅
        .take(5)
        .subscribe();

客户端输出结果：

[main] INFO org.example.RSocketClientRaw - My UUID is 28afc749-75e1-4289-8607-14810103de6c
[main] INFO org.example.RSocketClientRaw - Test3 subscribed to RequestStreamRequesterFlux
Test3 Received payload: Resp from Server: 1, Thank you!
Test3 Received payload: Resp from Server: 2, Thank you!
Test3 Received payload: Resp from Server: 3, Thank you!
Test3 Received payload: Resp from Server: 4, Thank you!
Test3 Received payload: Resp from Server: 5, Thank you!

服务端接收到了请求：

2023-03-12 22:01:33.520  INFO 32099 --- [or-http-epoll-3] o.example.controller.RSocketController   : [handler.request.stream]: TEST3 - Request&Stream

5. Channel

服务端

服务端接收来自客户端的整数字符串，将它们乘以2以后发送回去。我们不妨把处理客户端请求流的函数封装为一个 Spring Service：

@Service
public class MathService {
    public Flux<String> doubleInteger(Flux<String> request) {
        return request
                .map(s -> {
                    System.out.println("received " + s);
                    int i = Integer.parseInt(s);
                    return String.valueOf(i * 2);
                });
    }

}

编写处理函数：

@Autowired
private MathService mathService;

@MessageMapping("handler.request.channel")
public Flux<String> responseChannel(Flux<String> payloads) {
    return this.mathService.doubleInteger(payloads);
}

客户端

Flux<Payload> payloadFlux = Flux.range(-5, 10)
        .delayElements(Duration.ofMillis(500))
        .map(obj ->
        {
            ByteBuf metadata = encodeRoute("handler.request.channel");
            return ByteBufPayload.create(
                    ByteBufUtil.writeUtf8(ByteBufAllocator.DEFAULT, obj.toString()), metadata);
        });

Flux<Payload> channelResp = socket.requestChannel(payloadFlux);
channelResp
        .doOnSubscribe(subscription -> logger.info("Test4 subscribed to {}", subscription.toString()))
        .doOnError(throwable -> logger.info(throwable.toString()))
        .doOnNext(nextPayload -> System.out.println("Test4 Received payload: " + nextPayload.getDataUtf8()))
        .subscribe();

客户端输出：

[main] INFO org.example.RSocketClientRaw - My UUID is 96ff8fe7-416c-4607-9518-463114725a7a
[main] INFO org.example.RSocketClientRaw - Test4 subscribed to RequestChannelRequesterFlux
Test4 Received payload: -10
Test4 Received payload: -8
Test4 Received payload: -6
Test4 Received payload: -4
Test4 Received payload: -2
Test4 Received payload: 0
Test4 Received payload: 2
Test4 Received payload: 4
Test4 Received payload: 6
Test4 Received payload: 8

服务端输出：

2023-03-12 22:07:05.542  INFO 33083 --- [or-http-epoll-2] o.example.controller.RSocketController   : [connect.setup]Client connection: 96ff8fe7-416c-4607-9518-463114725a7a

received -5
received -4
received -3
received -2
received -1
received 0
received 1
received 2
received 3
received 4

---

下一篇文章会展示服务端如何主动调用客户端的函数。如有错误欢迎在评论区批评指正！