These are chat archives for ReactiveX/RxJava

1st
Dec 2016
Dorus
@Dorus
Dec 01 2016 01:28
@renanferrari Sorry i didnt read that close enough, will see if i can find some answers tomorrow (doesnt look like anyone else is around, weird normally you get more replies here). One question, do you want to change Firebase's SDK to use a different thread, or do you want to change Rx to make downstream operators use a different thread?
Dorus
@Dorus
Dec 01 2016 01:35
@renanferrari I looked into it a bit more, and it seems it's 100% pointless to use subscribeOn as firebase handles calling stuff in the background itself. However if you need to process data returned from firebase elsewhere (on another thread), you need to schedule it yourself. Use .observeOn for that.
Can you explain why observeOn does not solve your problem?
Renan Ferrari
@renanferrari
Dec 01 2016 11:56
@Dorus You're right, .observeOn() does give me the ability to process the data returned by Firebase on another thread. That's what I'm doing already, but it just isn't the point of the question. The point is: when I pass a Scheduler through .subscribeOn() I expect the upstream to conform to that Scheduler's thread but that doesn't happen when the Observable has an internal listener that is being triggered from a callback on a different thread. When that happens, I lose the .subscribeOn() guarantee.
The severity of this issue may not seem obvious at first, but what if that Observable was part of a library? What's the best practice there? Should the library enforce its clients to always call an .observeOn() after any call to that method? Should the library call an .observeOn() itself and call it a "default Scheduler"? In any of these cases the .subscribeOn() is just useless, and that doesn't seem right to me.
@Dorus And thank you for the answer! I have updated my question to clarify that.
Dorus
@Dorus
Dec 01 2016 18:36
@renanferrari First of all, subscribeOn, somehow (and especially on RxJava) is a overused operator that rarely does what's expected. And indeed, as you have experienced, it doesn't do what you expected.
In short, subscribeOn will register the events on the thread specified in subscribeOn, but that does not mean the callback from those events will happen on the same thread that was used ot register them. A proper library might indeed do so, but more often libraries wil use their own internal threads, or in case of Firebase, use the main thread apparently. This is a problem with Firebase, not Rx. If you implement something like this in Rx, it would be smart to use the trampoline sceduler (i.e. same thread scheduler) to prevent these problems.
observeOn is merely a hotfix that will ensure the main thread isnt tied up long while you process an event.

Oh and to come back to your question

Should the library enforce its clients to always call an .observeOn() after any call to that method?

No, instead it would be best practice for clients to not do any long computations on the event dispatcher. If you dont do anything long running, it's way more efficient to simply borrow the already active thread, as context switching is expensive on itself. If you do need to do long running computations, it's always a smart idea to schedule these in the background, preferably in parallelle.

Renan Ferrari
@renanferrari
Dec 01 2016 18:39
@Dorus I'm aware this is not a problem with RxJava, I just want to know how to make RxJava work, properly, with this kind of listener.
Dorus
@Dorus
Dec 01 2016 18:40
Oh and in Rx it's simple to do that: source.flatMap(e => longRunning(e), computationScheduler)
David Karnok
@akarnokd
Dec 01 2016 18:40
subscribeOn is provided as the means to move the subscription side effects to another thread. Many sources, such as most listener APIs don't have subscription side effects and subscribeOn is completely ineffective. When these sources emit, that can happen from any thread. UI-like sources emit from the main thread, dedicated sources emit from their own background thread. In order to process these the required place, you should use observeOn. In addition, you may want to run your own single-threaded scheduler via Schedulers.from(Executors.newSingleThreadedExecutor()) which guarantees observing on it from multiple stages will end up on the same thread in a serialized manner.
Dorus
@Dorus
Dec 01 2016 18:41
^^ on the RxJs gitter people are always impressed with my knowledge, but in here @akarnokd words things so much better.
David Karnok
@akarnokd
Dec 01 2016 18:47
@Dorus how does that work in RxJS? Whenever a source signals, it queues that value and schedules a task to drain that particular queue on the given scheduler, right?
Dorus
@Dorus
Dec 01 2016 18:47
@akarnokd Well the Js people care surprisingly little about threading :)
They always tell me Js is single-threaded.
I'm a bit of a weirdo there too, because i know a lot about Rx so i'm good help there, but my Js knowledge is limited as i'm more of a Java/C# programmer in origine.
David Karnok
@akarnokd
Dec 01 2016 18:53
There is a logical evolution in the Rx ecosystem that works the other direction as well. For example, if you know the fusion-enabled RxJava 2 deep enough, you can leave off "features" and end up with Rx.NET or RxJS.
Renan Ferrari
@renanferrari
Dec 01 2016 18:58
@akarnokd Great explanation of .subscribeOn()! Thanks for that. I'm just not so sure if I understood your suggested solution with .observeOn() and also what you meant by "guarantees observing from multiple stages".
David Karnok
@akarnokd
Dec 01 2016 19:05
When you have observeOn() multiple times in a sequence because you route computation back and forth. Schedulers.computation() will hand out a Scheduler that provides separate thread (from a fixed set) to each use of observeOn and it is very unlikely two observeOn will run on the same thread.