Simon Baslé
Simon Baslé
another fundamental question / limitation is with any operator that combines multiple sources. it is not even always known in advance which sources will get combine: for example, what should...
In Netty, it looks like `IdleStateHandler` still uses some form of scheduling with a configurable `EventExecutor`... Not sure how this would translate in reactor-core and how different it would be....
Interesting side note: RxJava seems to have a close approach where timeout tasks also always trigger, but the id of the "real" timeout task is captured so that an outdated...
yes, pretty much. I just want to set expectation here in case somebody wants to contribute: it can still get rejected if the benchmark doesn't show enough improvement.
thanks for providing that @unoexperto. I had a bit of trouble comparing that code to the `FluxTimeout` one because of Kotlin, but it seems very close. I was surprised to...
@unoexperto @bruto1 looks like I missed the fact that timeout isn't rescheduled in `onNext` but in `doTimeout` (which currently is imprecise, but could be improved). So with that approach we...
@pderop is conducting some interesting experiments with HashedWheelTimers. we do see a measurable improvement so we might consider it. the only question is what to expose exactly, and I'm leaning...
@bruto1 @pderop observed in his benchmark there was more time spent in the implicit picking of a `Worker` in `Schedulers.parallel()`, which isn't the case with `Schedulers.single()`. The use of `single()`...
_From @smaldini on March 3, 2017 10:42_ Yes we might need to change to a monitor based model to wake up the blocked loop
after further experiments, exploring the alternative with factories / wrapping schedulers: ### hook + `ContextRunnnable` approach: The application of hook(s) can be done transparently using the old API, since the...