Optimise Quarkus REST layer for closed world

[snowdrop-bot]

Profling a Quarkus application it's immediatley visible that the current RESTEasy extension isn't taking advantage of the closed world assumption:

Annotation scanning and type introspection:

Stack Trace	TLABs	Total TLAB Size(bytes)	Pressure(%)
java.lang.reflect.Method.getParameterTypes()	28	137,118,288	6.174
   sun.reflect.annotation.AnnotationInvocationHandler.invoke(Object, Method, Object[])	28	137,118,288	6.174
      com.sun.proxy.$Proxy17.annotationType()	16	77,359,528	3.483
         org.jboss.resteasy.spi.util.FindAnnotation.findAnnotation(Annotation[], Class)	16	77,359,528	3.483
            org.jboss.resteasy.plugins.providers.sse.SseEventSinkInterceptor.filter(ContainerRequestContext)	16	77,359,528	3.483

And more..

Stack Trace	TLABs	Total TLAB Size(bytes)	Pressure(%)
java.lang.Class.getInterfaces()	425	2,083,949,208	93.826
   org.jboss.resteasy.core.providerfactory.Utils.createHeaderDelegateFromInterfaces(Map, Class[])	219	1,071,276,576	48.233
   org.jboss.resteasy.core.providerfactory.Utils.createHeaderDelegate(Map, Class)	206	1,012,672,632	45.594

I suppose this implies we can improve it a lot still?

The above methods have been identified when looking for strong allocators; this implies memory consumption could be cut down by dodging these operations.

---

quarkusio#4345

---

$upstream:4345$

[geoand]

I will follow up with RESTEasy folks to see if there are plans about this.

[geoand]

This has been now moved to work on Quarkus REST

[geoand]

This has been delivered in the form of RESTEasy Reactive