达到 api 端点时实施 "retries" 的最佳方式？

Question

我正在使用 ApacheHttpClient

我有一个 Java 方法（在 Java 微服务中）向外部端点（端点）发出 Http POST 请求我不拥有）。一切通常都运行良好，但有时端点出现故障。代码看起来像这样（简化）：

private HttpResponseData postRequest(String url) throws Exception {
        HttpResponseData response = null;
        try (InputStream key = MyAPICaller.class.getResourceAsStream(keyPath)) {
            MyAPICaller.initializeApiClient(Username, PassString, key);
            int attempts = REQUEST_RETRY_COUNT; // starts at 3

            while (attempts-- > 0) {
                try {
                    response = MyAPICaller.getInstance().post(url);
                    break;
                } catch (Exception e) {
                    log.error("Post Request to {} failed. Retries remaining {}", url, attempts);
                    Thread.sleep(REQUEST_RETRY_DELAY * 1000);
                }
            }

            if (response == null)
                throw new Exception("Post request retries exceeded. Unable to complete request.");

        }
        return response;
    }

我没有写原始代码，但正如你所看到的，它看起来像是在发出请求，而 REQUEST_RETRY_COUNT 大于 0 （看起来总是如此），它将尝试将 post 转换为 url。貌似那里不知为何有个断点，所以跳进try块后，会一直断，没有重试机制。

Java 中是否有通用设计模式来实现重试模式以命中外部端点？我知道在 Java 脚本中你可以使用带有 returns 承诺的 Fetch API，是否有与 Java 类似的东西？

Answer 1

像 GCP 和 AWS 这样的云平台通常都有自己的重试策略，这应该是首选方法。

如果您想使用自己的重试策略，指数回退可能是一个很好的起点。

它可以是基于注释的，您可以在其中注释您的客户端方法。例如，您将 API 方法注释如下：

@Retry(maxTries = 3, retryOnExceptions = {RpcException.class}) public UserInfo getUserInfo(String userId);

@Documented
@Retention(RetentionPolicy.RUNTIME)
@Target(ElementType.METHOD)
public @interface Retry {

  int maxTries() default 0;

  /**
   * Attempt retry if one of the following exceptions is thrown.
   * @return array of applicable exceptions
   */
  Class<? extends Throwable> [] retryOnExceptions() default {Throwable.class};
}

方法拦截器可以实现如下：

public class RetryMethodInterceptor implements MethodInterceptor {

  private static final Logger logger = Logger.getLogger(RetryMethodInterceptor.class.getName());

  @Override
  public Object invoke(MethodInvocation methodInvocator) throws Throwable {
    Retry retryAnnotation = methodInvocator.getMethod().getAnnotation(Retry.class);
    Set<Class<? extends Throwable>> retriableExceptions =
        Sets.newHashSet(retryAnnotation.retryOnExceptions());

    String className = methodInvocator.getThis().getClass().getCanonicalName();
    String methodName = methodInvocator.getMethod().getName();

    int tryCount = 0;
    while (true) {
      try {
        return methodInvocator.proceed();
      } catch (Throwable ex) {
        tryCount++;
        boolean isExceptionInAllowedList = isRetriableException(retriableExceptions, ex.getClass());
        if (!isExceptionInAllowedList) {
          System.out.println(String.format(
              "Exception not in retry list for class: %s - method: %s - retry count: %s",
              className, methodName, tryCount));
          throw ex;
        } else if (isExceptionInAllowedList && tryCount > retryAnnotation.maxTries()) {
          System.out.println(String
                  .format(
                      "Exhausted retries, rethrowing exception for class: %s - method: %s - retry count: %s",
                      className, methodName, tryCount));
          throw ex;
        }
        System.out.println(String.format("Retrying for class: %s - method: %s - retry count: %s",
            className, methodName, tryCount));
      }
    }
  }

  private boolean isRetriableException(Set<Class<? extends Throwable>> allowedExceptions,
      Class<? extends Throwable> caughtException) {
    for (Class<? extends Throwable> look : allowedExceptions) {
      // Only compare the class names we do not want to compare superclass so Class#isAssignableFrom
      // can't be used.
      if (caughtException.getCanonicalName().equalsIgnoreCase(look.getCanonicalName())) {
        return true;
      }
    }
    return false;
  }
}

达到 api 端点时实施 "retries" 的最佳方式？

Best way to implement "retries" when hitting api endpoint?

java

api

http

endpoint