Martin Kouba

The https://github.com/quarkusio/quarkus/pull/40183 is related to this issue.

> I would call the method `getInjectionTarget()`, not `getInjectedTarget()`. I wouldn't because then a caller could expect the `jakarta.enterprise.inject.spi.InjectionTarget` to be returned which is not the case. Hm, we could...

> But we could call it `getProgramElement()`, because it does return a representation of the program element corresponding to the injection point. How about just `getAnnotationTarget()`?

Well, we might also try `getInjectionPointTarget()` but from my point of view, all these names (`getAnnotationTarget()`, `getActualTarget()` and `getInjectionPointTarget()`) are ok so let Matej decide (and take on the naming...

CC @cescoffier @geoand @brunobat @Ladicek @maxandersen

Note: the current concurrency model needs to be revisited. First we should decide if we limit the concurrency per connection...

### Concurrency model There are 3 different execution models used to execute an endpoint callback: 1. Event loop - no parallel processing, ordering not guaranteed 2. Worked thread - depends...

> One important thing to mention is that the concurrency is per connection - not global. @cescoffier Yes, that's a good point. And the concurrency limiter mentioned in the "Current...

I have refactored the contexts (Vert.x, CDI) as follows: - we do not associate a single duplicated context with the connection anymore, - instead, we create a new duplicated context...

This pull request is superseded by https://github.com/quarkusio/quarkus/pull/39142. However, I will keep this branch to preserve the history...