引入 Gateway 網關，這些坑必定要學會避開！！！

Spring cloud gateway是替代zuul的網關產品，基於Spring 五、Spring boot 2.0以上、Reactor, 提供任意的路由匹配和斷言、過濾功能。上一篇文章談了一下Gateway網關使用不規範，同事加班淚兩行~，這篇文章將會側重於其餘的幾個須要注意的地方。

網關實現

這裏介紹編碼方式實現

HystrixObservableCommand.Setter getSetter() {
        HystrixCommandGroupKey groupKey = HystrixCommandGroupKey.Factory.asKey("group-accept");
        HystrixObservableCommand.Setter setter = HystrixObservableCommand.Setter.withGroupKey(groupKey);
        HystrixCommandKey commandKey = HystrixCommandKey.Factory.asKey("command-accept");

        setter.andCommandKey(commandKey);

        HystrixCommandProperties.Setter proertiesSetter = HystrixCommandProperties.Setter();
        proertiesSetter
                /* *
                 * 線程策略配置
                 */
                //設置線程模式 缺省 1000ms
                .withExecutionIsolationStrategy(HystrixCommandProperties.ExecutionIsolationStrategy.THREAD)
                //執行是否啓用超時時間 缺省 true
                .withExecutionTimeoutEnabled(true)
                //使用線程隔離時，是否對命令執行超時的線程調用中斷 缺省false
                .withExecutionIsolationThreadInterruptOnFutureCancel(false)
                //執行超時的時候是否要它中斷 缺省 true
                .withExecutionIsolationThreadInterruptOnTimeout(true)
                //執行的超時時間 缺省 1000ms
                .withExecutionTimeoutInMilliseconds(2000)
                /* *
                 * 熔斷策略
                 */
                //是否開啓溶斷 缺省 true
                .withCircuitBreakerEnabled(true)
                // 是否容許熔斷器忽略錯誤,默認false, 不開啓 ；
                // true，斷路器強制進入「關閉」狀態，它會接收全部請求。
                // 若是forceOpen屬性爲true，該屬性不生效
                .withCircuitBreakerForceClosed(false)
                // 是否強制開啓熔斷器阻斷全部請求, 默認爲false
                // 爲true時，全部請求都將被拒絕，直接到fallback.
                // 若是該屬性設置爲true，斷路器將強制進入「打開」狀態，
                // 它會拒絕全部請求。該屬性優於forceClosed屬性
                .withCircuitBreakerForceOpen(false)
                // 用來設置當斷路器打開以後的休眠時間窗。
                // 休眠時間窗結束以後，會將斷路器設置爲「半開」狀態，嘗試熔斷的請求命令，
                // 若是依然請求錯誤就將斷路器繼續設置爲「打開」狀態，若是成功，就設置爲「關閉」狀態
                // 熔斷器默認工做時間,默認:5000豪秒.
                // 熔斷器中斷請求10秒後會進入半打開狀態,放部分流量過去重試.
                .withCircuitBreakerSleepWindowInMilliseconds(5000)
                // 熔斷器在整個統計時間內是否開啓的閥值.
                // 在metricsRollingStatisticalWindowInMilliseconds（默認10s）內默認至少請求10次，
                // 熔斷器才發揮起做用,9次熔斷器都不起做用。
                .withCircuitBreakerRequestVolumeThreshold(100)
                // 該屬性用來設置斷路器打開的錯誤百分比條件。默認值爲50.
                // 表示在滾動時間窗中，在請求值超過requestVolumeThreshold閾值的前提下，
                // 若是錯誤請求數百分比超過50，就把斷路器設置爲「打開」狀態，不然就設置爲「關閉」狀態
                .withCircuitBreakerErrorThresholdPercentage(50);

        setter.andCommandPropertiesDefaults(proertiesSetter);

        return setter;
    }
@Bean
public RouteLocator customRouteLocator(RouteLocatorBuilder builder) {
 RouteLocatorBuilder.Builder routes = builder.routes();
 RouteLocatorBuilder.Builder serviceProvider = routes
  .route("accept",
   r -> r.method(HttpMethod.GET)
    .and()
    .path("/gateway-accept/**")
    .and()
    .header(HttpHeaders.CONTENT_TYPE, "application/json;charset=UTF-8")
    .filters(f -> {
     f.rewritePath("/gateway-accept/(?<path>.*)", "/${path}");
     f.requestRateLimiter(
      config -> config.setKeyResolver(new GenericAccessResolver())
       .setRateLimiter(redisRateLimiter()));
     f.hystrix(config -> config.setName("accept")
      .setFallbackUri("forward:/gateway-fallback")
      .setSetter(getSetter()));
     return f;
    })
    .uri("http://localhost:8888")
     );

 return serviceProvider.build();
}

 

在上面的代碼中，主要作了3件事情：限流、熔斷策略及降級方法配置

限流

• 配置redis

spring:
  redis:
    database: 0
    host: 127.0.0.1
    port: 6379
    password:
    timeout: 1500
    lettuce:
      pool:
        max-active: 300 #鏈接池最大鏈接數（使用負值表示沒有限制）
        max-idle: 10    #鏈接池中的最大空閒鏈接
        min-idle: 5     #鏈接池中的最小空閒鏈接
        max-wait: -1    #鏈接池最大阻塞等待時間（使用負值表示沒有限制）

 

• 自定義解析

/**
 * @description: 按照訪問地址進行限流(也能夠安裝其餘條件進行限流)，具體能夠看exchange.getRequest()的方法和屬性
 **/
public class GenericAccessResolver implements KeyResolver {
    @Override
    public Mono<String> resolve(ServerWebExchange exchange) {
        return Mono.just(exchange.getRequest().getPath().value());
    }
}

 

• 自定義限流配置

RedisRateLimiter redisRateLimiter() {
 //1000,1500對應replenishRate、burstCapacity
 return new RedisRateLimiter(1000, 1500);
}

 

• 網關使用自定義限流器（網關使用代碼實現）

@Bean
public RouteLocator customRouteLocator(RouteLocatorBuilder builder) {
    RouteLocatorBuilder.Builder routes = builder.routes();
    RouteLocatorBuilder.Builder serviceProvider = routes
        .route("accept",
               r -> r.method(HttpMethod.GET)
               .and()
               .path("/gateway-accept/**")
               .and()
               .header(HttpHeaders.CONTENT_TYPE, "application/json;charset=UTF-8")
               //.and()
               //.readBody(String.class, readBody -> true)
               .filters(f -> {
                   f.rewritePath("/gateway-accept/(?<path>.*)", "/${path}");
                   f.requestRateLimiter(config -> config.setKeyResolver(new GenericAccessResolver()).setRateLimiter(redisRateLimiter()));                                   
                   return f;
               })
               .uri("http://localhost:8888")
              );

    return serviceProvider.build();
}

• 測試

• jmeter配置

• 結果

• 其餘

  若是有多個路由，使用不一樣的限流策略，能夠自定義KeyResolver和RedisRateLimiter， 在路由定義時加入

//基於ip限流
public class OtherAccessResolver implements KeyResolver {
    @Override
    public Mono<String> resolve(ServerWebExchange exchange) {
        return Mono.just(exchange.getRequest().getRemoteAddress().getHostName());
    }
}
RedisRateLimiter otherRedisRateLimiter() {
 //1000,1500對應replenishRate、burstCapacity
 return new RedisRateLimiter(100, 500);
}
@Bean
public RouteLocator customRouteLocator(RouteLocatorBuilder builder) {
 RouteLocatorBuilder.Builder routes = builder.routes();
 RouteLocatorBuilder.Builder serviceProvider = routes
  .route("accept",
   r -> r.method(HttpMethod.GET)
    .and()
    .path("/gateway-accept/**")
    .and()
    .header(HttpHeaders.CONTENT_TYPE, "application/json;charset=UTF-8")
    .filters(f -> {
     f.rewritePath("/gateway-accept/(?<path>.*)", "/${path}");
     f.requestRateLimiter(
      config -> config.setKeyResolver(new GenericAccessResolver())
       .setRateLimiter(redisRateLimiter()));
     f.hystrix(config -> config.setName("accept")
      .setFallbackUri("forward:/gateway-fallback")
      .setSetter(getSetter()));
     return f;
    })
    .uri("http://localhost:8888"))
        .route("sign",
   r -> r.method(HttpMethod.POST)
    .and()
    .path("/gateway-sign/**")
    .and()
    .header(HttpHeaders.CONTENT_TYPE, "application/json;charset=UTF-8")
    .filters(f -> {
     f.rewritePath("/gateway-sign/(?<path>.*)", "/${path}");
     f.requestRateLimiter(
      config -> config.setKeyResolver(new OtherAccessResolver())
       .setRateLimiter(otherRedisRateLimiter()));
     f.hystrix(config -> config.setName("sign")
      .setFallbackUri("forward:/gateway-fallback")
      .setSetter(getSetter()));
     return f;
    })
    .uri("http://localhost:7777")
     );

 return serviceProvider.build();
}

 

熔斷策略

熔斷策略主要是線程配置和熔斷配置，上面已經說明很清楚了。在上篇文章中，爲了解決網關調用後臺服務Connection prematurely closed BEFORE response的問題，要設置後臺服務線程的空閒時間和網關線程池線程的空閒時間，並讓網關線程池線程的空閒時間小於後臺服務的空閒時間

配置方法

spring:
  cloud:
    gateway:
      httpclient:
        pool:
            max-connections: 500
            max-idle-time: 10000

 

編碼實現

翻閱Spring Cloud Gateway英文資料，知道路由提供一個metadata方法，能夠設置路由的元數據(https://docs.spring.io/spring-cloud-gateway/docs/2.2.6.RELEASE/reference/html/#route-metadata-configuration)，這些元數據在RouteMetadataUtils中定義：

package org.springframework.cloud.gateway.support;

public final class RouteMetadataUtils {
    public static final String RESPONSE_TIMEOUT_ATTR = "response-timeout";
    public static final String CONNECT_TIMEOUT_ATTR = "connect-timeout";

    private RouteMetadataUtils() {
        throw new AssertionError("Must not instantiate utility class.");
    }
}

 

其中沒有我要的線程數量（max-connection）和空閒時間（max-idle-time）的設置，沒有關係，本身加上去：

@Bean
public RouteLocator customRouteLocator(RouteLocatorBuilder builder) {
        RouteLocatorBuilder.Builder routes = builder.routes();
        RouteLocatorBuilder.Builder serviceProvider = routes
           .route("accept",
               r -> r.method(HttpMethod.GET)
                     .and()
                     .path("/gateway-accept/**")
                     .and()
                     .header(HttpHeaders.CONTENT_TYPE, "application/json;charset=UTF-8")
                     .filters(f -> {
                          f.rewritePath("/gateway-accept/(?<path>.*)", "/${path}");
                          f.requestRateLimiter(
                             config -> config.setKeyResolver(new GenericAccessResolver())
                                           .setRateLimiter(redisRateLimiter()));
                          f.hystrix(config -> config.setName("accept")
                                      .setFallbackUri("forward:/gateway-fallback")
                                      .setSetter(getSetter()));
                                return f;
                          })
                     .uri("http://localhost:8888")
                     .metadata("max-idle-time", 10000)  //網關調用後臺線程空閒時間設置
                     .metadata("max-connections", 200)  //網關調用後臺服務線程數量設置
          );

     return serviceProvider.build();
}

 

測試果真和yml配置同樣有效果。

降級方法

降級方法自己沒有什麼特別，有一個問題須要注意，調用降級方法也是使用線程池的，缺省在HystrixThreadPoolProperties中定義：

public abstract class HystrixThreadPoolProperties {

    /* defaults */
    static int default_coreSize = 10;            // core size of thread pool
    static int default_maximumSize = 10;         // maximum size of thread pool
    static int default_keepAliveTimeMinutes = 1; // minutes to keep a thread alive
    static int default_maxQueueSize = -1;        // size of queue (this can't be dynamically changed so we use 'queueSizeRejectionThreshold' to artificially limit and reject)
                                                 // -1 turns it off and makes us use SynchronousQueue

 

錯誤

若是上面的限流設置比較大，好比1000，最大突發2000，網關調用後臺服務發生熔斷降級， 熔斷後降級的方法調用太頻繁，10個線程不夠用，會致使如下500錯誤：

2021-02-01 14:29:45.076 ERROR 64868 --- [ioEventLoop-5-1] a.w.r.e.AbstractErrorWebExceptionHandler : [a0ed6911-18982]  500 Server Error for HTTP GET "/gateway-accept/test"

com.netflix.hystrix.exception.HystrixRuntimeException: command-accept fallback execution rejected.
 at com.netflix.hystrix.AbstractCommand.handleFallbackRejectionByEmittingError(AbstractCommand.java:1043) ~[hystrix-core-1.5.18.jar:1.5.18]
 Suppressed: reactor.core.publisher.FluxOnAssembly$OnAssemblyException: 
Error has been observed at the following site(s):
 |_ checkpoint ⇢ org.springframework.cloud.gateway.filter.WeightCalculatorWebFilter [DefaultWebFilterChain]
 |_ checkpoint ⇢ HTTP GET "/gateway-accept/test" [ExceptionHandlingWebHandler]
com.netflix.hystrix.exception.HystrixRuntimeException: command-accept fallback execution rejected.
 at com.netflix.hystrix.AbstractCommand.handleFallbackRejectionByEmittingError(AbstractCommand.java:1043) ~[hystrix-core-1.5.18.jar:1.5.18]
 Suppressed: reactor.core.publisher.FluxOnAssembly$OnAssemblyException: 
Error has been observed at the following site(s):
 |_ checkpoint ⇢ org.springframework.cloud.gateway.filter.WeightCalculatorWebFilter [DefaultWebFilterChain]
 |_ checkpoint ⇢ HTTP GET "/gateway-accept/test" [ExceptionHandlingWebHandler]

配置方法

因此要在yml中設置合適的調用降級方法的線程池, 合理的配置可以杜絕網關500錯誤的發生。

hystrix:
  threadpool:
    group-accept:  #代碼裏面設置的HystrixCommandGroupKey.Factory.asKey("group-accept")
      coreSize: 50 #併發執行的最大線程數，默認10
      maxQueueSize: 1500 #BlockingQueue的最大隊列數
      #即便maxQueueSize沒有達到，達到queueSizeRejectionThreshold該值後，請求也會被拒絕
      queueSizeRejectionThreshold: 1400                                   

網關異常截獲

上面的異常後，沒有捕獲異常直接返回前端500錯誤，通常狀況下須要返回一個統一接口，好比：

@Data
@ToString
@EqualsAndHashCode
@Accessors(chain = true)
public class Result<T> implements Serializable {
    private Integer code;
    private String message;
    private T data;
    private String sign;

    public static final String SUCCESS = "成功";
    public static final String FAILURE = "失敗";


    public Result(int code, String message) {
        this.code = code;
        this.message = message;
    }

    public Result(int code, String message, T data) {
        this.code = code;
        this.message = message;
        this.data = data;
    }
    public Result(int code, String message, T data, String sign) {
        this.code = code;
        this.message = message;
        this.data = data;
        this.sign = sign;
    }

    public static Result<Object> success() {
        return new Result<Object>(200, SUCCESS);
    }
    public static Result<Object> success(Object data) {
        return new Result<Object>(200, SUCCESS, data);
    }
    public static Result<Object> success(Object data, String sign) {
        return new Result<Object>(200, SUCCESS, data, sign);
    }
    public static Result<Object> failure() {
        return new Result<Object>(400, FAILURE);
    }
    public static Result<Object> failure(Object data) {
        return new Result<Object>(400, FAILURE, data);
    }
    public static Result<Object> failure(Object data, String sign) {
        return new Result<Object>(400, FAILURE, data, sign);
    }


}

 

建立GlobalExceptionConfiguration 實現ErrorWebExceptionHandler（這一段是來者網友提供的）

@Slf4j
@Order(-1)
@Component
@RequiredArgsConstructor
public class GlobalExceptionConfiguration implements ErrorWebExceptionHandler {
    private final ObjectMapper objectMapper;

    @Override
    public Mono<Void> handle(ServerWebExchange exchange, Throwable ex) {
        ServerHttpResponse response = exchange.getResponse();

        if (response.isCommitted()) {
            return Mono.error(ex);
        }

        response.getHeaders().setContentType(MediaType.APPLICATION_JSON_UTF8);
        if (ex instanceof ResponseStatusException) {
            response.setStatusCode(((ResponseStatusException) ex).getStatus());
        }

        return response
                .writeWith(Mono.fromSupplier(() -> {
                    DataBufferFactory bufferFactory = response.bufferFactory();
                    try {
                        return bufferFactory.wrap(objectMapper.writeValueAsBytes(Result.failure(ex.getMessage())));
                    } catch (JsonProcessingException e) {
                        log.warn("Error writing response", ex);
                        return bufferFactory.wrap(new byte[0]);
                    }
                }));
    }
}

 

這樣，就會把網關異常統一包裝在接口中返回：如：

後臺日誌已經沒有以前的錯誤日誌了。

編碼實現，沒找到

因爲Spring Cloud Gateway 中的 Hystrix採用的是HystrixObservableCommand.Setter, 沒有采用 HystrixCommand.Setter， 在 HystrixCommand.Setter中是能夠編碼實現線程池配置的, 可是在HystrixObservableCommand.Setter沒有提供：

final public static class Setter {
        protected final HystrixCommandGroupKey groupKey;
        protected HystrixCommandKey commandKey;
        protected HystrixThreadPoolKey threadPoolKey;  //有屬性可是沒有set方法
        protected HystrixCommandProperties.Setter commandPropertiesDefaults;
        protected HystrixThreadPoolProperties.Setter threadPoolPropertiesDefaults; //有屬性沒有set方法

        protected Setter(HystrixCommandGroupKey groupKey) {
            this.groupKey = groupKey;
            // default to using SEMAPHORE for ObservableCommand
            commandPropertiesDefaults = setDefaults(HystrixCommandProperties.Setter());
        }

        public static Setter withGroupKey(HystrixCommandGroupKey groupKey) {
            return new Setter(groupKey);
        }

        public Setter andCommandKey(HystrixCommandKey commandKey) {
            this.commandKey = commandKey;
            return this;
        }

        public Setter andCommandPropertiesDefaults(HystrixCommandProperties.Setter commandPropertiesDefaults) {
            this.commandPropertiesDefaults = setDefaults(commandPropertiesDefaults);
            return this;
        }

        private HystrixCommandProperties.Setter setDefaults(HystrixCommandProperties.Setter commandPropertiesDefaults) {
            if (commandPropertiesDefaults.getExecutionIsolationStrategy() == null) {
                // default to using SEMAPHORE for ObservableCommand if the user didn't set it
                commandPropertiesDefaults.withExecutionIsolationStrategy(ExecutionIsolationStrategy.SEMAPHORE);
            }
            return commandPropertiesDefaults;
        }
    }

 

因爲本人水平有限，沒有找到Setter中設置HystrixThreadPoolKey和HystrixThreadPoolProperties.Setter的方法，因此只能在yml中配置。有知道的同窗告訴我一聲，不勝感激。

總結

因此在Spring Cloud Gateway網關的配置中，須要綜合考慮限流大小、網關調用後臺鏈接池設置大小、後臺服務的鏈接池以及空閒時間，包括網關調用降級方法的線程池配置，都須要在壓測中調整到一個合理的配置，才能發揮最大的功效。

本人水平有限，跟深刻的研究還在繼續，若是文章有表達錯誤或者不周，請你們指正，謝謝！