Spring Cloud升级之路 - Hoxton - 10. 网关重试带Body的请求Body丢失的问题

带 Body 的重试 Body 丢失

之前我们的配置里面，只对 Get 请求针对 500 响应码重试，但是针对像 Post 这样的请求，只对那种根本还没到发送 Body 的阶段的异常（例如连接异常）这种重试，那么如果我们要对带 Body 的 Post 进行重试呢？或者就是用 Postman 构建一个带 Body 的 Get 请求，重试是否正常呢？

我们启动之前第6节的 EurekaServer，修改/test-exception-thrown接口，增加 RequestBody 参数：

@RequestMapping(value = "/test-exception-thrown", method = {RequestMethod.GET, RequestMethod.POST, RequestMethod.PUT, RequestMethod.DELETE})
public String testExceptionThrown(HttpServletRequest httpServletRequest, @RequestBody Map body) {
    log.info("testExceptionThrow called {}, {}", httpServletRequest.getMethod(), body);
    if (shouldThrowException) {
        throw new IllegalStateException();
    }
    return zone;
}

启动zone1-service-provider-instance1，zone1-service-provider-instance2，其中，zone1-service-provider-instance1是接口访问会抛出异常的那个实例。启动网关，使用 Postman 调用接口，发现出现重试，请求先发送到了zone1-service-provider-instance1，之后重试到了zone1-service-provider-instance2，但是zone1-service-provider-instance2返回 400 错误，也就是没有收到 RequestBody，这是怎么回事呢？

Api网关调用日志

2020-07-28 01:55:29.781  INFO [service-api-gateway,fc71e34f22e1bd17,fc71e34f22e1bd17]
            [7860] [boundedElastic-1][com.github.hashjang.hoxton.api.gateway.filter.InstanceCircuitBreakerFilter:54]: try to send request to: http://192.168.0.142:8001/test-exception-thrown: stats: {"numberOfNotPermittedCalls":0,"numberOfSlowCalls":0,"numberOfBufferedCalls":0,"slowCallRate":-1.0,"failureRate":-1.0,"numberOfSuccessfulCalls":0,"numberOfFailedCalls":0,"numberOfSlowSuccessfulCalls":0,"numberOfSlowFailedCalls":0}
2020-07-28 01:55:30.115  INFO [service-api-gateway,fc71e34f22e1bd17,fc71e34f22e1bd17]
            [7860] [boundedElastic-1][com.github.hashjang.hoxton.api.gateway.filter.InstanceCircuitBreakerFilter:54]: try to send request to: http://192.168.0.142:8001/test-exception-thrown: stats: {"numberOfNotPermittedCalls":0,"numberOfSlowCalls":0,"numberOfBufferedCalls":1,"slowCallRate":-1.0,"failureRate":-1.0,"numberOfSuccessfulCalls":1,"numberOfFailedCalls":0,"numberOfSlowSuccessfulCalls":0,"numberOfSlowFailedCalls":0}

zone1-service-provider-instance1日志：

2020-07-28 01:55:29.789 ERROR
            [service-provider,,] [24956]
            [XNIO-2 task-4][io.undertow.servlet.api.LoggingExceptionHandler:80]:UT005023: Exception handling request to /test-exception-thrown

zone1-service-provider-instance2日志：

2020-07-28 01:55:30.133  WARN
            [service-provider,fc71e34f22e1bd17,da6d3f91fcfc053f] [24956]
            [XNIO-2 task-5][org.springframework.web.servlet.handler.AbstractHandlerExceptionResolver:199]:Resolved [org.springframework.http.converter.HttpMessageNotReadableException: Required request body is missing: public java.lang.String com.github.hashjang.hoxton.service.provider.controller.TestServiceController.testExceptionThrown(javax.servlet.http.HttpServletRequest,java.util.Map)]

为了定位问题，我们添加一个放在最开头的 LogFilter，开启 Body 的追踪：

@Component
public class LogFilter implements GlobalFilter, Ordered {
    @Override
    public Mono filter(ServerWebExchange exchange, GatewayFilterChain chain) {
        return chain.filter(exchange.mutate().request(new ServerHttpRequestDecorator(exchange.getRequest()) {
            @Override
            public Flux getBody() {
                //body开启日志，记录操作body的filter
                return exchange.getRequest().getBody().log();
            }
        }).build());
    }

    @Override
    public int getOrder() {
        //放在最开头
        return Ordered.HIGHEST_PRECEDENCE;
    }
}

重启网关，发送请求：

2020-07-28 02:22:16.026  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [boundedElastic-1][com.github.hashjang.hoxton.api.gateway.filter.InstanceCircuitBreakerFilter:54]: try to send request to: http://192.168.0.142:8001/test-exception-thrown: stats: {"numberOfNotPermittedCalls":0,"numberOfSlowSuccessfulCalls":0,"numberOfSuccessfulCalls":0,"numberOfSlowFailedCalls":0,"numberOfFailedCalls":0,"failureRate":-1.0,"slowCallRate":-1.0,"numberOfBufferedCalls":0,"numberOfSlowCalls":0}
2020-07-28 02:22:16.034  INFO [service-api-gateway,,]
            [4408] [reactor-http-nio-4][reactor.util.Loggers$Slf4JLogger:274]: onContextUpdate(Context3{class brave.propagation.TraceContext=1ae80e0b643da3c7/1ae80e0b643da3c7, class org.springframework.cloud.sleuth.instrument.web.client.HttpClientBeanPostProcessor$CurrentClientSpan=NoopSpan(1ae80e0b643da3c7/38ecc8fd2b789c2e), reactor.onDiscard.local=reactor.core.publisher.Operators$$Lambda$1179/0x00000008019d1840@469db4af})
2020-07-28 02:22:16.034  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onSubscribe([Fuseable] ScopePassingSpanSubscriber)
2020-07-28 02:22:16.034  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: request(unbounded)
2020-07-28 02:22:16.035  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onNext(PooledSlicedByteBuf(ridx: 0, widx: 10, cap: 10/10, unwrapped: PooledUnsafeDirectByteBuf(ridx: 326, widx: 326, cap: 1024)))
2020-07-28 02:22:16.035  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onComplete()
2020-07-28 02:22:16.165  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [boundedElastic-1][com.github.hashjang.hoxton.api.gateway.filter.InstanceCircuitBreakerFilter:54]: try to send request to: http://192.168.0.142:8002/test-exception-thrown: stats: {"numberOfNotPermittedCalls":0,"numberOfSlowSuccessfulCalls":0,"numberOfSuccessfulCalls":0,"numberOfSlowFailedCalls":0,"numberOfFailedCalls":0,"failureRate":-1.0,"slowCallRate":-1.0,"numberOfBufferedCalls":0,"numberOfSlowCalls":0}
2020-07-28 02:22:16.169  INFO [service-api-gateway,,]
            [4408] [reactor-http-nio-3][reactor.util.Loggers$Slf4JLogger:274]: onContextUpdate(Context3{class brave.propagation.TraceContext=1ae80e0b643da3c7/1ae80e0b643da3c7, class org.springframework.cloud.sleuth.instrument.web.client.HttpClientBeanPostProcessor$CurrentClientSpan=NoopSpan(1ae80e0b643da3c7/d05978ee5d1cb64c), reactor.onDiscard.local=reactor.core.publisher.Operators$$Lambda$1179/0x00000008019d1840@3a55aa8})
2020-07-28 02:22:16.170  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onSubscribe([Fuseable] ScopePassingSpanSubscriber)
2020-07-28 02:22:16.170  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: request(unbounded)
2020-07-28 02:22:16.170  INFO [service-api-gateway,1ae80e0b643da3c7,1ae80e0b643da3c7]
            [4408] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onComplete()

我们发现，这个 Body 的 Flux 在重试的时候，使用的还是原来同样的 Flux，但是这个 Flux 已经被第一次调用消费过了，所以重试的时候，再去消费，直接返回消费完成，不会有：onNext(PooledSlicedByteBuf(ridx: 0, widx: 10, cap: 10/10, unwrapped: PooledUnsafeDirectByteBuf(ridx: 326, widx: 326, cap: 1024)))

那么如何解决呢？有两种方式，一种是自己实现 Body 缓存，参考我提的 Issue + PR（https://github.com/spring-cloud/spring-cloud-gateway/pull/1863），但是这实际上是我的乌龙，我没注意到 Spring Cloud Gateway实际上已经实现了：

Publisher publisher = chain.filter(exchange.mutate().request(
	new ServerHttpRequestDecorator(request) {
		@Override
		public Flux getBody() {
			int currentIteration = exchange
					.getAttributeOrDefault(RETRY_ITERATION_KEY, -1);
			//根据currentIteration判断是否是重试，如果不是，就返回原始Request 的 body Flux
			//如果是，则返回缓存的String重新生成的Flux，保证重试也有正确的body
			return currentIteration > -1 ? Flux.from(Mono.just(dataBufferFactory.wrap(((String) exchange.getAttributes().get(BODY)).getBytes()))) :
					request.getBody().map(dataBuffer -> {
						if (LEGAL_LOG_MEDIA_TYPES.contains(contentType)) {
							try {
								String body = (String) exchange.getAttributes().get(BODY);
								if (body == null) {
									byte[] content = new byte[dataBuffer.readableByteCount()];
									try {
										dataBuffer.read(content);
									} finally {
										DataBufferUtils.release(dataBuffer);
									}
									String s = new String(content, Charset.defaultCharset());
									exchange.getAttributes().put(BODY, s);
									dataBuffer = dataBufferFactory.wrap(s.getBytes());
								} else {
									dataBuffer = dataBufferFactory.wrap(body.getBytes());
								}

							} catch (Exception e) {
								log.error("error read body in retry", e);
							}
						}
						return dataBuffer;
					});
		}
	}
).build())

另一种是使用 Spring Cloud Gateway 已有的缓存机制AdaptCachedBodyGlobalFilter:
AdaptCachedBodyGlobalFilter源码：

public class AdaptCachedBodyGlobalFilter
		implements GlobalFilter, Ordered, ApplicationListener {

    /**
	 * 缓存RequestBody的Route
	 */
	private ConcurrentMap routesToCache = new ConcurrentHashMap<>();

	/**
	 * 缓存RequestBody的Attribute Key
	 */
	@Deprecated
	public static final String CACHED_REQUEST_BODY_KEY = CACHED_REQUEST_BODY_ATTR;

    /**
	 * 收到EnableBodyCachingEvent，则将EnableBodyCachingEvent中的RouteId加入到要缓存的Route的Map
	 */
	@Override
	public void onApplicationEvent(EnableBodyCachingEvent event) {
		this.routesToCache.putIfAbsent(event.getRouteId(), true);
	}
    //。。。。略
}

由于我们是全局的重试，所以可以对每一个Route都加上缓存 Body 的机制，所以可以这么实现：

ApiGatewayConfig

@Configuration
@EnableConfigurationProperties(ApiGatewayRetryConfig.class)
@LoadBalancerClients(defaultConfiguration = CommonLoadBalancerConfig.class)
public class ApiGatewayConfig {
    @Autowired
    private AdaptCachedBodyGlobalFilter adaptCachedBodyGlobalFilter;
    @Autowired
    private GatewayProperties gatewayProperties;

    @PostConstruct
    public void init() {
        //让每一个路径都做body Cache，这样重试有Body的请求的时候，重试的请求不会没有body，因为原始body是一次性的基于netty的FluxReceive
        gatewayProperties.getRoutes().forEach(routeDefinition -> {
            EnableBodyCachingEvent enableBodyCachingEvent = new EnableBodyCachingEvent(new Object(), routeDefinition.getId());
            adaptCachedBodyGlobalFilter.onApplicationEvent(enableBodyCachingEvent);
        });
    }
}

这样修改后，重启网关，我们再调用触发重试：

2020-07-28 02:48:18.972  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onContextUpdate(Context2{class brave.propagation.TraceContext=72eba79a3afc324f/72eba79a3afc324f, reactor.onDiscard.local=reactor.core.publisher.Operators$$Lambda$1041/0x0000000801979440@119927bc})
2020-07-28 02:48:18.972  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onSubscribe([Fuseable] ScopePassingSpanSubscriber)
2020-07-28 02:48:18.973  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: request(unbounded)
2020-07-28 02:48:18.973  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onNext(PooledSlicedByteBuf(ridx: 0, widx: 10, cap: 10/10, unwrapped: PooledUnsafeDirectByteBuf(ridx: 326, widx: 326, cap: 512)))
2020-07-28 02:48:18.974  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [reactor-http-nio-2][reactor.util.Loggers$Slf4JLogger:274]: onComplete()
2020-07-28 02:48:18.986  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [boundedElastic-1][com.github.hashjang.hoxton.api.gateway.filter.InstanceCircuitBreakerFilter:54]: try to send request to: http://192.168.0.142:8001/test-exception-thrown: stats: {"numberOfNotPermittedCalls":0,"failureRate":-1.0,"slowCallRate":-1.0,"numberOfSlowFailedCalls":0,"numberOfFailedCalls":0,"numberOfSlowSuccessfulCalls":0,"numberOfSuccessfulCalls":0,"numberOfBufferedCalls":0,"numberOfSlowCalls":0}
2020-07-28 02:48:19.138  INFO [service-api-gateway,72eba79a3afc324f,72eba79a3afc324f]
            [6784] [boundedElastic-1][com.github.hashjang.hoxton.api.gateway.filter.InstanceCircuitBreakerFilter:54]: try to send request to: http://192.168.0.142:8002/test-exception-thrown: stats: {"numberOfNotPermittedCalls":0,"failureRate":-1.0,"slowCallRate":-1.0,"numberOfSlowFailedCalls":0,"numberOfFailedCalls":0,"numberOfSlowSuccessfulCalls":0,"numberOfSuccessfulCalls":1,"numberOfBufferedCalls":1,"numberOfSlowCalls":0}

发现重试调用，Body没有丢，重试成功了