Feign clients default ConnectTimeout & ReadTimeout aren't overriden by Ribbon configuration

[codependent]

Given a Feign client with the following Ribbon configuration:

images-microservice:
  ribbon:
    ConnectTimeout: 250
    ReadTimeout: 1000
    OkToRetryOnAllOperations: true
    MaxAutoRetriesNextServer: 2
    MaxAutoRetries: 2

One would expect that ConnectTimeout and ReadTimeout are 250 and 1000. However they never override Feign's defaults.

In org.springframework.cloud.netflix.feign.ribbon.FeignLoadBalancer.execute()

    Request.Options options;
    if (configOverride != null) {
        options = new Request.Options(
                configOverride.get(CommonClientConfigKey.ConnectTimeout,
                        this.connectTimeout),
                (configOverride.get(CommonClientConfigKey.ReadTimeout,
                        this.readTimeout)));
    }

this.connectTimeout is correct (250), but what actually is gotten from configOverride.get() is 10000. This value comes from Feign.Options' default value:

public static class Options {

    private final int connectTimeoutMillis;
    private final int readTimeoutMillis;
    ...
    public Options() {
      this(10 * 1000, 60 * 1000);
    }

Shouldn't FeignLoadBalancer.execute() prioritize Ribbon's configuration over the default Feign's Request.Options? If not, how are we supposed to override it?

[spencergibb]

I don't see how Feign.Options can override those values in FeignLoadBalancer. https://github.com/spring-cloud/spring-cloud-netflix/blob/master/spring-cloud-netflix-core/src/main/java/org/springframework/cloud/netflix/feign/ribbon/FeignLoadBalancer.java#L66-L73

I haven't verified the problem, just that we don't invoke the empty Feign.Options.

[codependent]

Hi spencergibb, let me explain it further. The sequence is as follows:

1. feign.SynchronousMethodHandler.executeAndDecode(RequestTemplate template) is called. And here we have response = client.execute(request, options); where options is a Request.Options object with connectTimeout = 10000 and readTimeout = 60000, that is to say, Request.Options' default values.

2. client.execute is LoadBalancerFeignClient.execute(Request request, Request.Options options), where the next code is executed:

   return lbClient(clientName).executeWithLoadBalancer(ribbonRequest,
                   new FeignOptionsClientConfig(options)).toResponse();
   

   As we see, Request.Options options is wrapped in a FeignOptionsClientConfig object.

3. The next step is FeignLoadBalancer.executeWithLoadBalancer(final S request, final IClientConfig requestConfig). which eventually calls FeingLoadBalancer.execute(RibbonRequest request, IClientConfig configOverride):

@Override
    public RibbonResponse execute(RibbonRequest request, IClientConfig configOverride)
            throws IOException {
        Request.Options options;
        if (configOverride != null) {
            options = new Request.Options(
                    configOverride.get(CommonClientConfigKey.ConnectTimeout,
                            this.connectTimeout),
                    (configOverride.get(CommonClientConfigKey.ReadTimeout,
                            this.readTimeout)));
        }

Notice that IClientConfig requestConfig/configOverride is the object that wrapped the default Request.Options in step 2. Therefore, when
options = new Request.Options( configOverride.get(CommonClientConfigKey.ConnectTimeout, this.connectTimeout),
is executed, configOverride already has ConnectTimeout and ReadTimeout values (10000 and 60000) loaded from the former Request.Options, so this.connectTimeout (250) is never applied.

I hope I managed to explain myself...

[spencergibb]

Yes, very well explained. A workaround is to create a @Bean of type Request.Options and match that to the feign client using feign configurations. I was able to reproduce.

[pmvilaca]

👍

Just to give a +1 to this issue as I was also debugging the same issue yesterday and the workaround force the clients to re-write the interface that can be provided by the servers in a separated jar

[PedroAlvarado]

+1 just ran into this issue as well. As suggested by @spencergibb here is the workaround I'm using.

    @Bean
    Request.Options requestOptions(ConfigurableEnvironment env){
        int ribbonReadTimeout = env.getProperty("ribbon.ReadTimeout", int.class, 6000);
        int ribbonConnectionTimeout = env.getProperty("ribbon.ConnectTimeout", int.class, 3000);

        return new Request.Options(ribbonConnectionTimeout, ribbonReadTimeout);
    }

[asarkar]

@codependent What's the images-microservice in your question, is the the value in @FeignClient(name = images-microservice)?