Sergei Egorov

Have you considered @testcontainers? :)

@forax yes, sorry, talking about SVM. > none of them works with native image because they all rely one way or another on java.lang.invoke reflection. oh, okay 😅 I thought...

Thing to keep in mind: `Hooks.onEachOperator`.

Simplified reproducer, fails on 3.3.x as well: ```java Flux.range(0, 3) // parallelism must be >= 2 .parallel(2) .doOnNext(i -> { throw new RuntimeException("boom"); }) .log("afterFlatMap") .ordered(Comparator.naturalOrder()) .blockFirst(Duration.ofSeconds(1)); ```

@Sage-Pierce you seem to be blocking the non-blocking threads. Maybe you wanted `publishOn`, not `subscribeOn` before `doOnNext`?

when I remove `Thread.sleep` from `doOnNext` everything seems to be fine. Since doing `Thread.sleep` there is incorrect, @Sage-Pierce please provide a reproducer that is not relying on it.

> FWIW, the use case itself is similar to a sort of deduplication functionality on an infinite stream. For deduplicating an infinite stream, I'd recommend something like this (in case...

@Sage-Pierce I still see `.subscribeOn(scheduler, true)` in the code - is it intentional? The problem is that you're blocking the parallel scheduler and it may be that the delays you're...

@Sage-Pierce this code still blocks the thread (by running the heavy computations). Consider using `publishOn` before `doOnNext`

Although I haven't fixed the issue yet, I think I got a much simpler reproducer: ```java final int total = 100; Long count = Flux.range(0, total) .groupBy(i -> (i /...