Backpressure: Window by Size

[benjchristensen]

The window operators that use time or Observable boundaries do not participate in backpressure ... they act as "temporal" operators for flow control and not reactive pull. The window(int size) variant however seems like it should work with reactive pull backpressure and behave similarly to groupBy.

In other words, it can be applied to a cold observable and correctly compose backpressure to emit windows and items within each window at a controlled rate.

Any thoughts or opinions on this?

[headinthebox]

Yes, anything with count sounds like a slamdunk to participate in back-pressure. In fact, I can imagine we would ask for chunks the size of the window.

[benjchristensen]

I'll try and get this done in time.

[benjchristensen]

I'm probably not going to get this in for 1.0.0 so it will have to come during the 1.0.x releases.

[pakoito]

In layman's terms, this feature request is a sliding window for the latest up to N elements over T time, right?

I'm going through the documentation and haven't found anything similar, and it's a common use component in this environments.

[benjchristensen]

@pakoito no, this is adding reactive pull backpressure to already existing functionality.

Go ahead and open a new issue with your use case if the operators are not working for you (see http://reactivex.io/RxJava/javadoc/rx/Observable.html#window(long,%20long,%20java.util.concurrent.TimeUnit,%20int,%20rx.Scheduler) )

[pakoito]

That operator is what I was after, thank you. I would add to the documentation/example that the windows for that case can overlap, unlike other count based window/buffer operations.

[zsxwing]

The window(int size) variant however seems like it should work with reactive pull backpressure and behave similarly to groupBy.

I think it's more like buffer. Only support to request the number of Windows., but don't support backpressure for each Window Observable. Right?

[benjchristensen]

I actually see this one as more similar to groupBy as it needs to provide backpressure on all windows, otherwise there isn't any backpressure.

[akarnokd]

Is this in the works? @zsxwing PR #1862 was closed do you want to revive it?

[zsxwing]

I'm busy with other work and may not go back to work on this one soon. If someone is interested in the issue, please go head.

[akarnokd]

Window supports backpressure on the outer Observable as of 1.0.14 and there is a PR that adds backpressure support to the inner observable: #3150.

[akarnokd]

Backpressure support added + there is a subsequent fix to this, #3678

[akarnokd]

Closing via #3678.