Performance discussion about default openfeign client

Resume
Discussion about default client performance of spring-cloud-openfeign

Versions
Java 17
Springboot 2.7.18
spring-cloud-dependencies 2021.0.8

Scenario

During a performance tests in a microservice that contain a 3 feign clients, it was possible to saw a high GC usage (i.e. high cpu utilization as well) in two methods, as we can see at profiler image bellow

We have a high tps scenario here - 110tps. The situation was absolutely unstable and we has too much resources to handle with that.

After studying the issue I decided to use lib {io.github.openfeign:feign-httpclient:11.8} and do a new evaluation. To my surprise, we had a significant improvement in resource usage, as feign started using other classes and no longer generated as many objects, as shown bellow

So, my question: Is this the expected behavior of the default implementation or is there something I did incorrectly, because we've a big difference here.

Code

I don't have a high customization clients, only a custom decoder and base64 auth, as bellow

@FeignClient(value = "...", url = "...",
        configuration = ClientConfig.class)
public interface .. {

    @PostMapping("..")
    SpecificObject post(
            @RequestHeader("..") Boolean ..,
            @RequestBody SpecificObject ..
    );
}

Begin with the default client path used by spring-cloud-openfeign under Java 17, Spring Boot 2.7.18, and spring-cloud-dependencies 2021.0.8, then compare a 110 TPS run with feign-httpclient 11.8. Use the profiler evidence and the stated custom decoder and basic-auth configuration to isolate the allocation difference. Done means a reproducible explanation of whether the default behavior is expected and a documented conclusion or actionable change.