Hystrix Feign针对500错误的重试机制失效，求排查帮助

Let's break down why your Feign retry logic isn't kicking in when the TEST-CONFIG service returns a 500 error, and walk through actionable fixes:

1. Hystrix Timeout is Cutting Off Retries Early #

Feign's retry process needs to complete before Hystrix triggers a timeout and fallback. By default, Hystrix has a 1-second timeout, which is likely too short for your 5 retry attempts (even with short initial delays). If Hystrix times out first, it will abort the Feign call entirely, skipping retries.

Fix: #

Increase the Hystrix timeout to give Feign enough time to complete all retries. Add this to your application.yml or application.properties:

hystrix:
  command:
    # Apply to all Hystrix commands, or target your specific client
    default:
      execution:
        isolation:
          thread:
            timeoutInMilliseconds: 30000  # 30 seconds to cover multiple retries
    # Optional: Target only your TestClient's test() method
    TestClient#test():
      execution:
        isolation:
          thread:
            timeoutInMilliseconds: 30000

2. Feign's Retryer Doesn't Work with Async Return Types (Observable) #

Your TestClient returns an Observable<String>, which is an asynchronous type. Feign's built-in Retryer is designed for synchronous calls only—it won't automatically retry async streams.

Fixes: #

• Option 1: Switch to a synchronous return type (if your use case allows) to leverage Feign's native retry:

  @FeignClient(name = "TEST-CONFIG", configuration = FeignRetryConfig.class, fallbackFactory = XYZClientFallbackFactory.class)
  public interface TestClient {
      @RequestMapping(value = "/test", method = RequestMethod.GET, consumes = MediaType.APPLICATION_JSON_VALUE)
      String test(); // Replace Observable<String> with String
  }
  

• Option 2: Add retry logic directly to the Observable stream using RxJava's retryWhen operator:

  // In your service class where you call TestClient
  Observable<String> testResponse = testClient.test()
      .retryWhen(RetryHandler.create(5, 1000)); // Retry 5 times with 1-second delays
  

3. Hystrix Retries Are Disabled by Default #

Hystrix might be blocking retries even if Feign is configured to retry. You need to explicitly enable Hystrix's retry functionality.

Fix: #

Add this to your configuration:

hystrix:
  command:
    default:
      execution:
        retry:
          enabled: true # Enable Hystrix-level retries
      fallback:
        enabled: true # Ensure fallback is enabled (your error shows fallback failed too, so verify fallback logic as well)

4. Your RetryableException May Be Missing Critical Context #

Your FeignErrorDecoder returns a RetryableException, but the constructor you're using doesn't pass the request method or retry timestamp. Some Feign versions require this context to recognize the exception as retryable.

Fix: #

Update your ErrorDecoder to use a more complete RetryableException constructor:

public class FeignErrorDecoder implements ErrorDecoder {
    private final ErrorDecoder defaultErrorDecoder = new Default();
    private static final Logger log = LoggerFactory.getLogger(FeignErrorDecoder.class);

    @Override
    public Exception decode(String methodKey, Response response) {
        Exception exception = defaultErrorDecoder.decode(methodKey, response);
        if (response.status() == 500) {
            log.error(String.format("##### Got %s response from %s #######", response.status(), methodKey));
            return new RetryableException(
                    response.status(),
                    exception.getMessage(),
                    response.request().httpMethod(),
                    exception,
                    Date.from(Instant.now().plusSeconds(1)), // Retry after 1 second
                    response.request()
            );
        }
        return exception;
    }
}

5. Verify Feign Configuration Priority #

Double-check that your FeignRetryConfig is actually being used:

• The @Primary annotation ensures it takes precedence over other Feign configurations, but confirm no other @Configuration classes are overriding it.
• Ensure your @FeignClient correctly references FeignRetryConfig.class in the configuration attribute (your code looks correct here, but it's worth a quick check).

Start with adjusting the Hystrix timeout first—it's the most common culprit here. If that doesn't work, test with a synchronous return type to rule out async-related issues.

内容的提问来源于stack exchange，提问作者user280960