如何在 Spring WebFlux 中记录请求和响应主体

Question

我想在 Spring WebFlux with Kotlin 上的 REST API 中集中记录请求和响应。到目前为止，我已经尝试过这种方法

@Bean
fun apiRouter() = router {
    (accept(MediaType.APPLICATION_JSON) and "/api").nest {
        "/user".nest {
            GET("/", userHandler::listUsers)
            POST("/{userId}", userHandler::updateUser)
        }
    }
}.filter { request, next ->
    logger.info { "Processing request $request with body ${request.bodyToMono<String>()}" }
    next.handle(request).doOnSuccess { logger.info { "Handling with response $it" } }
}

这里request method和path log成功但是body是Mono，怎么log呢？是否应该反过来，我必须订阅请求正文 Mono 并将其记录在回调中？另一个问题是此处的 ServerResponse 接口无法访问响应主体。我怎样才能在这里得到它？

我尝试过的另一种方法是使用 WebFilter

@Bean
fun loggingFilter(): WebFilter =
        WebFilter { exchange, chain ->
            val request = exchange.request
            logger.info { "Processing request method=${request.method} path=${request.path.pathWithinApplication()} params=[${request.queryParams}] body=[${request.body}]"  }

            val result = chain.filter(exchange)

            logger.info { "Handling with response ${exchange.response}" }

            return@WebFilter result
        }

同样的问题：请求正文是 Flux 而没有响应正文。

有没有办法访问完整的请求和响应以从某些过滤器进行日志记录？我不明白什么？

Answer 1

这与Spring MVC 中的情况或多或少相似。

在 Spring MVC 中，您可以使用 AbstractRequestLoggingFilter 过滤器和 ContentCachingRequestWrapper and/or ContentCachingResponseWrapper。这里有很多权衡：

如果您想访问 servlet 请求属性，您需要实际读取和解析请求正文
记录请求体意味着缓冲请求体，这会占用大量内存
如果您想访问响应主体，您需要包装响应并在写入时缓冲响应主体，以供以后检索

ContentCaching*Wrapper 类在 WebFlux 中不存在，但您可以创建类似的。但请记住这里的其他要点：

在内存中缓冲数据以某种方式违背了反应式堆栈，因为我们正在尝试在那里非常有效地利用可用资源
您不应篡改实际数据流并刷新 more/less 频率超过预期，否则您可能会破坏流式处理用例
在该级别，您只能访问 DataBuffer 个实例，这些实例（大致）是内存高效的字节数组。那些属于缓冲池并被回收用于其他交易所。如果这些不正确 retained/released，就会产生内存泄漏（并且缓冲数据以供以后使用当然适合这种情况）
同样在该级别，它只是字节，您无权访问任何编解码器来解析 HTTP 正文。如果内容不是人类可读的，我会忘记缓冲内容

您问题的其他答案：

是的，WebFilter 可能是最好的方法
不，您不应该订阅请求正文，否则您将使用处理程序无法读取的数据；您可以 flatMap 在 doOn operators
包装响应应该可以让您在写入时访问响应主体；不过不要忘记内存泄漏

Answer 2

我没有找到记录 request/response 正文的好方法，但是如果您只对元数据感兴趣，那么您可以像下面那样做。

import org.springframework.http.HttpHeaders
import org.springframework.http.HttpStatus
import org.springframework.http.server.reactive.ServerHttpResponse
import org.springframework.stereotype.Component
import org.springframework.web.server.ServerWebExchange
import org.springframework.web.server.WebFilter
import org.springframework.web.server.WebFilterChain
import reactor.core.publisher.Mono

@Component
class LoggingFilter(val requestLogger: RequestLogger, val requestIdFactory: RequestIdFactory) : WebFilter {
    val logger = logger()

    override fun filter(exchange: ServerWebExchange, chain: WebFilterChain): Mono<Void> {
        logger.info(requestLogger.getRequestMessage(exchange))
        val filter = chain.filter(exchange)
        exchange.response.beforeCommit {
            logger.info(requestLogger.getResponseMessage(exchange))
            Mono.empty()
        }
        return filter
    }
}

@Component
class RequestLogger {

    fun getRequestMessage(exchange: ServerWebExchange): String {
        val request = exchange.request
        val method = request.method
        val path = request.uri.path
        val acceptableMediaTypes = request.headers.accept
        val contentType = request.headers.contentType
        return ">>> $method $path ${HttpHeaders.ACCEPT}: $acceptableMediaTypes ${HttpHeaders.CONTENT_TYPE}: $contentType"
    }

    fun getResponseMessage(exchange: ServerWebExchange): String {
        val request = exchange.request
        val response = exchange.response
        val method = request.method
        val path = request.uri.path
        val statusCode = getStatus(response)
        val contentType = response.headers.contentType
        return "<<< $method $path HTTP${statusCode.value()} ${statusCode.reasonPhrase} ${HttpHeaders.CONTENT_TYPE}: $contentType"
    }

    private fun getStatus(response: ServerHttpResponse): HttpStatus =
        try {
            response.statusCode
        } catch (ex: Exception) {
            HttpStatus.CONTINUE
        }
}

Answer 3

我是 Spring WebFlux 的新手，我不知道如何在 Kotlin 中做到这一点，但应该与 Java 使用 WebFilter 时相同：

public class PayloadLoggingWebFilter implements WebFilter {

    public static final ByteArrayOutputStream EMPTY_BYTE_ARRAY_OUTPUT_STREAM = new ByteArrayOutputStream(0);

    private final Logger logger;
    private final boolean encodeBytes;

    public PayloadLoggingWebFilter(Logger logger) {
        this(logger, false);
    }

    public PayloadLoggingWebFilter(Logger logger, boolean encodeBytes) {
        this.logger = logger;
        this.encodeBytes = encodeBytes;
    }

    @Override
    public Mono<Void> filter(ServerWebExchange exchange, WebFilterChain chain) {
        if (logger.isInfoEnabled()) {
            return chain.filter(decorate(exchange));
        } else {
            return chain.filter(exchange);
        }
    }

    private ServerWebExchange decorate(ServerWebExchange exchange) {
        final ServerHttpRequest decorated = new ServerHttpRequestDecorator(exchange.getRequest()) {

            @Override
            public Flux<DataBuffer> getBody() {

                if (logger.isDebugEnabled()) {
                    final ByteArrayOutputStream baos = new ByteArrayOutputStream();
                    return super.getBody().map(dataBuffer -> {
                        try {
                            Channels.newChannel(baos).write(dataBuffer.asByteBuffer().asReadOnlyBuffer());
                        } catch (IOException e) {
                            logger.error("Unable to log input request due to an error", e);
                        }
                        return dataBuffer;
                    }).doOnComplete(() -> flushLog(baos));

                } else {
                    return super.getBody().doOnComplete(() -> flushLog(EMPTY_BYTE_ARRAY_OUTPUT_STREAM));
                }
            }

        };

        return new ServerWebExchangeDecorator(exchange) {

            @Override
            public ServerHttpRequest getRequest() {
                return decorated;
            }

            private void flushLog(ByteArrayOutputStream baos) {
                ServerHttpRequest request = super.getRequest();
                if (logger.isInfoEnabled()) {
                    StringBuffer data = new StringBuffer();
                    data.append('[').append(request.getMethodValue())
                        .append("] '").append(String.valueOf(request.getURI()))
                        .append("' from ")
                            .append(
                                Optional.ofNullable(request.getRemoteAddress())
                                            .map(addr -> addr.getHostString())
                                        .orElse("null")
                            );
                    if (logger.isDebugEnabled()) {
                        data.append(" with payload [\n");
                        if (encodeBytes) {
                            data.append(new HexBinaryAdapter().marshal(baos.toByteArray()));
                        } else {
                            data.append(baos.toString());
                        }
                        data.append("\n]");
                        logger.debug(data.toString());
                    } else {
                        logger.info(data.toString());
                    }

                }
            }
        };
    }

}

这里有一些测试：github

我想这就是 Brian Clozel (@brian-clozel) 的意思。

Answer 4

布赖恩说的。此外，记录 request/response 主体对于反应式流没有意义。如果您将流过管道的数据想象成一个流，那么您在任何时候都不会拥有完整的内容除非您对其进行缓冲，这违背了整点。对于小 request/response，您可以不用缓冲，但为什么要使用反应模型（除了给您的同事留下深刻印象 :-)）？

我能想到的记录 request/response 的唯一原因是调试，但是对于反应式编程模型，调试方法也必须修改。 Project Reactor 文档有一个关于调试的优秀部分，您可以参考：http://projectreactor.io/docs/core/snapshot/reference/#debugging

Answer 5

您实际上可以为 Netty 和 Reactor-Netty 启用 DEBUG 日志记录，以查看正在发生的事情的全貌。你可以玩下面的东西，看看你想要什么，不想要什么。我已经尽力了。

reactor.ipc.netty.channel.ChannelOperationsHandler: DEBUG
reactor.ipc.netty.http.server.HttpServer: DEBUG
reactor.ipc.netty.http.client: DEBUG
io.reactivex.netty.protocol.http.client: DEBUG
io.netty.handler: DEBUG
io.netty.handler.proxy.HttpProxyHandler: DEBUG
io.netty.handler.proxy.ProxyHandler: DEBUG
org.springframework.web.reactive.function.client: DEBUG
reactor.ipc.netty.channel: DEBUG

Answer 6

假设我们正在处理一个简单的 JSON 或 XML 响应，如果相应记录器的 debug 级别由于某种原因不够，可以在转换之前使用字符串表示反对：

Mono<Response> mono = WebClient.create()
                               .post()
                               .body(Mono.just(request), Request.class)
                               .retrieve()
                               .bodyToMono(String.class)
                               .doOnNext(this::sideEffectWithResponseAsString)
                               .map(this::transformToResponse);

以下是副作用和转化方法：

private void sideEffectWithResponseAsString(String response) { ... }
private Response transformToResponse(String response) { /*use Jackson or JAXB*/ }

Answer 7

这是我为 java 想出的。

public class RequestResponseLoggingFilter implements WebFilter {

    @Override
    public Mono<Void> filter(ServerWebExchange exchange, WebFilterChain chain) {
        ServerHttpRequest httpRequest = exchange.getRequest();
        final String httpUrl = httpRequest.getURI().toString();

        ServerHttpRequestDecorator loggingServerHttpRequestDecorator = new ServerHttpRequestDecorator(exchange.getRequest()) {
            String requestBody = "";

            @Override
            public Flux<DataBuffer> getBody() {
                return super.getBody().doOnNext(dataBuffer -> {
                    try (ByteArrayOutputStream byteArrayOutputStream = new ByteArrayOutputStream()) {
                        Channels.newChannel(byteArrayOutputStream).write(dataBuffer.asByteBuffer().asReadOnlyBuffer());
                        requestBody = IOUtils.toString(byteArrayOutputStream.toByteArray(), "UTF-8");
                        commonLogger.info(LogMessage.builder()
                                .step(httpUrl)
                                .message("log incoming http request")
                                .stringPayload(requestBody)
                                .build());
                    } catch (IOException e) {
                        commonLogger.error(LogMessage.builder()
                                .step("log incoming request for " + httpUrl)
                                .message("fail to log incoming http request")
                                .errorType("IO exception")
                                .stringPayload(requestBody)
                                .build(), e);
                    }
                });
            }
        };

        ServerHttpResponseDecorator loggingServerHttpResponseDecorator = new ServerHttpResponseDecorator(exchange.getResponse()) {
            String responseBody = "";
            @Override
            public Mono<Void> writeWith(Publisher<? extends DataBuffer> body) {
                Mono<DataBuffer> buffer = Mono.from(body);
                return super.writeWith(buffer.doOnNext(dataBuffer -> {
                    try (ByteArrayOutputStream byteArrayOutputStream = new ByteArrayOutputStream()) {
                        Channels.newChannel(byteArrayOutputStream).write(dataBuffer.asByteBuffer().asReadOnlyBuffer());
                        responseBody = IOUtils.toString(byteArrayOutputStream.toByteArray(), "UTF-8");
                        commonLogger.info(LogMessage.builder()
                                .step("log outgoing response for " + httpUrl)
                                .message("incoming http request")
                                .stringPayload(responseBody)
                                .build());
                    } catch (Exception e) {
                        commonLogger.error(LogMessage.builder()
                                .step("log outgoing response for " + httpUrl)
                                .message("fail to log http response")
                                .errorType("IO exception")
                                .stringPayload(responseBody)
                                .build(), e);
                    }
                }));
            }
        };
        return chain.filter(exchange.mutate().request(loggingServerHttpRequestDecorator).response(loggingServerHttpResponseDecorator).build());
    }

}

Answer 8

这是 GitHub Repo 完整实施 记录请求和响应 body 以及 http headers 基于 webflux/java 的应用程序...

Answer 9

如果你使用控制器而不是处理程序，最好的方法是使用 @Log annotation.And 注释你的控制器 class 仅供参考，这将普通的 json 对象作为请求而不是单声道。

@Target(AnnotationTarget.FUNCTION)
@Retention(AnnotationRetention.RUNTIME)
annotation class Log

@Aspect
@Component
class LogAspect {
    companion object {
        val log = KLogging().logger
    }

    @Around("@annotation(Log)")
    @Throws(Throwable::class)
    fun logAround(joinPoint: ProceedingJoinPoint): Any? {
        val start = System.currentTimeMillis()
        val result = joinPoint.proceed()
        return if (result is Mono<*>) result.doOnSuccess(getConsumer(joinPoint, start)) else result
    }

    fun getConsumer(joinPoint: ProceedingJoinPoint, start: Long): Consumer<Any>? {
        return Consumer {
            var response = ""
            if (Objects.nonNull(it)) response = it.toString()
            log.info(
                "Enter: {}.{}() with argument[s] = {}",
                joinPoint.signature.declaringTypeName, joinPoint.signature.name,
                joinPoint.args
            )
            log.info(
                "Exit: {}.{}() had arguments = {}, with result = {}, Execution time = {} ms",
                joinPoint.signature.declaringTypeName, joinPoint.signature.name,
                joinPoint.args[0],
                response, System.currentTimeMillis() - start
            )
        }
    }
}

Answer 10

自Spring Boot 2.2.x，Spring Webflux 支持Kotlin coroutines. With coroutines, you can have the advantages of non-blocking calls without having to handle Mono and Flux wrapped objects. It adds extensions to ServerRequest and ServerResponse，添加ServerRequest#awaitBody() 和ServerResponse.BodyBuilder.bodyValueAndAwait(body: Any) 等方法。所以你可以像这样重写你的代码：

@Bean
fun apiRouter() = coRouter {
    (accept(MediaType.APPLICATION_JSON) and "/api").nest {
        "/user".nest {
            /* the handler methods now use ServerRequest and ServerResponse directly
             you just need to add suspend before your function declaration:
             suspend fun listUsers(ServerRequest req, ServerResponse res) */ 
            GET("/", userHandler::listUsers)
            POST("/{userId}", userHandler::updateUser)
        }
    }

    // this filter will be applied to all routes built by this coRouter
    filter { request, next ->
      // using non-blocking request.awayBody<T>()
      logger.info("Processing $request with body ${request.awaitBody<String>()}")
        val res = next(request)
        logger.info("Handling with Content-Type ${res.headers().contentType} and status code ${res.rawStatusCode()}")
        res 
    }
}

为了创建一个带有coRoutines的WebFilter Bean，我想你可以使用这个CoroutineWebFilter接口（我没测试过，不知道行不行）。

Answer 11

我认为这里做的合适的事情是将每个请求的内容以异步方式写入文件（java.nio）并设置一个间隔来异步读取这些请求主体文件并写入它们以内存使用感知方式记录日志（一次至少一个文件，但一次最多 100 MB）并在记录它们后从磁盘中删除文件。

如何在 Spring WebFlux 中记录请求和响应主体

How to log request and response bodies in Spring WebFlux

spring-boot

kotlin

project-reactor

spring-webflux