Sam Lightfoot
Sam Lightfoot
I do also see around 20% additional throughput for HTTP/1.1 connections using a parallelism of 20 in the above test (different parallelisms produce very similar results). HTTP/1.1 - 60k/s HTTP/2...
We have seen very similar behaviour in our WebFlux project and are currently trying to create a minimum reproducible sample. Which Reactor instrumentation mode are you using (default is DECORATE_ON_EACH)?...
I see the same behaviour if using DECORATE_QUEUES, but using DECORATE_ON_EACH as the Reactor Instrumentation type preserved the trace id.
@violetagg What feature specifically are you thinking here? I may be able to help out.
@Kieun @jzheaux For what it's worth, we implemented the caching of raw tokens as described earlier to great success. We were spending 10% of our CPU time decoding (mostly) and...