Feign retries event if Retryer.NEVER_RETRY is configured in the AsyncFeign.<HttpClientContext>builder()

[jimartinmoreno]

Hello everyone!

We have run several tests with Feign async clients and even though we are configuring the AsyncFeign builder to not use retries. It worked correctly in most of the cases but when a HttpStatus SERVICE_UNAVAILABLE or TOO_MANY_REQUESTS is returned in the response it retries at least once:

• org.springframework.boot:spring-boot-starter-parent: 3.0.2
• org.springframework.cloud:spring-cloud-dependencies: 2022.0.1

Sample of code

import static feign.Retryer.NEVER_RETRY;

@Bean
    @Scope(ConfigurableBeanFactory.SCOPE_PROTOTYPE)
    public AsyncFeign.AsyncBuilder<HttpClientContext> defaultAsyncBuilder(AsyncClient<HttpClientContext> httpClient, Decoder decoder, Encoder encoder, Contract contract, ErrorDecoder errorDecoder, GlobalFeignRequestInterceptor globalFeignRequestInterceptor) {
              return AsyncFeign.<HttpClientContext>builder()
                          .client(httpClient)
                          .decoder(decoder)
                .errorDecoder(errorDecoder)
                .encoder(encoder)
                .dismiss404()
                .retryer(NEVER_RETRY) // Disable Feign retries
                .contract(contract)
                .requestInterceptor(globalFeignRequestInterceptor);
    }

cloud:
    openfeign:
      httpclient:
        max-connections-per-route: 5
        max-connections: 20
      client:
        config:
          default:
            loggerLevel: full 
            connectTimeout: 60
            readTimeout: 300
            dismiss404: true
          productFeignClient:
            connectTimeout: 10
            readTimeout: 20
            logger-level: full

Are we missing something or is this a known or unknown issue?

Thank you in advance for your answer

[velo]

That is unexpected. Which httpClient you using? It might not be feign, but the client feign uses behind the scenes that is retrying

[jimartinmoreno]

this are our dependencies:

• implementation 'io.github.openfeign:feign-hc5:12.1'
• implementation("org.springframework.cloud:spring-cloud-starter-openfeign") {
  exclude group: "com.fasterxml.jackson.core", module: "jackson-databind"
  }

[vitalijr2]

I think it is not a bug.

I have tested this case:

• with NEVER_RETRY this throws exception RetryableException instead of ServiceUnavailable and TooManyRequests but does not try to retry a request,
• with Default this also throws exception RetryableException and does to retry a request twice (I decrease maxAttempts value).

The instance of RetryableException has a real exception (ServiceUnavailable or TooManyRequests) in its getCause().

I have made the example https://github.com/radio-rogal/spring-boot-never-retry, there are two fast proofs:

• ./mvnw for NEVER_RETRY
• NEVER_RETRY=false ./mvnw for the retryer Default with two attempts.

Impornant note:

A remote server must send the header Retry-After. If the response does not have this header Feign throws regular exceptions ServiceUnavailable or TooManyRequests.

[ajtuss]

The feign-hc5 client contains enabled retry for SERVICE_UNAVAILABLE and TOO_MANY_REQUESTS by default.
It can be disabled by setting automaticRetriesDisabled property in the HttpClientBuilder:

 @Bean
  public HttpClientBuilderCustomizer customHttpClientBuilderCustomizer() {
    return HttpClientBuilder::disableAutomaticRetries;
  }