RxJava Fetching Observables In Parallel

Question

I need some help in implementing parallel asynchronous calls in RxJava. I have picked up a simple use case wherein the FIRST call fetches (rather searches) a list of product

Answer 1

The parallel operator proved to be a problem for almost all use cases and does not do what most expect from it, so it was removed in the 1.0.0.rc.4 release: https://github.com/ReactiveX/RxJava/pull/1716

A good example of how to do this type of behavior and get parallel execution can be seen here.

In your example code it is unclear if searchServiceClient is synchronous or asynchronous. It affects how to solve the problem slightly as if it is already async no extra scheduling is needed. If synchronous extra scheduling is needed.

First here are some simple examples showing synchronous and asynchronous behavior:

import rx.Observable;
import rx.Subscriber;
import rx.schedulers.Schedulers;

public class ParallelExecution {

    public static void main(String[] args) {
        System.out.println("------------ mergingAsync");
        mergingAsync();
        System.out.println("------------ mergingSync");
        mergingSync();
        System.out.println("------------ mergingSyncMadeAsync");
        mergingSyncMadeAsync();
        System.out.println("------------ flatMapExampleSync");
        flatMapExampleSync();
        System.out.println("------------ flatMapExampleAsync");
        flatMapExampleAsync();
        System.out.println("------------");
    }

    private static void mergingAsync() {
        Observable.merge(getDataAsync(1), getDataAsync(2)).toBlocking().forEach(System.out::println);
    }

    private static void mergingSync() {
        // here you'll see the delay as each is executed synchronously
        Observable.merge(getDataSync(1), getDataSync(2)).toBlocking().forEach(System.out::println);
    }

    private static void mergingSyncMadeAsync() {
        // if you have something synchronous and want to make it async, you can schedule it like this
        // so here we see both executed concurrently
        Observable.merge(getDataSync(1).subscribeOn(Schedulers.io()), getDataSync(2).subscribeOn(Schedulers.io())).toBlocking().forEach(System.out::println);
    }

    private static void flatMapExampleAsync() {
        Observable.range(0, 5).flatMap(i -> {
            return getDataAsync(i);
        }).toBlocking().forEach(System.out::println);
    }

    private static void flatMapExampleSync() {
        Observable.range(0, 5).flatMap(i -> {
            return getDataSync(i);
        }).toBlocking().forEach(System.out::println);
    }

    // artificial representations of IO work
    static Observable getDataAsync(int i) {
        return getDataSync(i).subscribeOn(Schedulers.io());
    }

    static Observable getDataSync(int i) {
        return Observable.create((Subscriber s) -> {
            // simulate latency
                try {
                    Thread.sleep(1000);
                } catch (Exception e) {
                    e.printStackTrace();
                }
                s.onNext(i);
                s.onCompleted();
            });
    }
}

Following is an attempt at providing an example that more closely matches your code:

import java.util.List;

import rx.Observable;
import rx.Subscriber;
import rx.schedulers.Schedulers;

public class ParallelExecutionExample {

    public static void main(String[] args) {
        final long startTime = System.currentTimeMillis();

        Observable searchTile = getSearchResults("search term")
                .doOnSubscribe(() -> logTime("Search started ", startTime))
                .doOnCompleted(() -> logTime("Search completed ", startTime));

        Observable populatedTiles = searchTile.flatMap(t -> {
            Observable reviews = getSellerReviews(t.getSellerId())
                    .doOnCompleted(() -> logTime("getSellerReviews[" + t.id + "] completed ", startTime));
            Observable imageUrl = getProductImage(t.getProductId())
                    .doOnCompleted(() -> logTime("getProductImage[" + t.id + "] completed ", startTime));

            return Observable.zip(reviews, imageUrl, (r, u) -> {
                return new TileResponse(t, r, u);
            }).doOnCompleted(() -> logTime("zip[" + t.id + "] completed ", startTime));
        });

        List allTiles = populatedTiles.toList()
                .doOnCompleted(() -> logTime("All Tiles Completed ", startTime))
                .toBlocking().single();
    }

    private static Observable getSearchResults(String string) {
        return mockClient(new Tile(1), new Tile(2), new Tile(3));
    }

    private static Observable getSellerReviews(int id) {
        return mockClient(new Reviews());
    }

    private static Observable getProductImage(int id) {
        return mockClient("image_" + id);
    }

    private static void logTime(String message, long startTime) {
        System.out.println(message + " => " + (System.currentTimeMillis() - startTime) + "ms");
    }

    private static  Observable mockClient(T... ts) {
        return Observable.create((Subscriber s) -> {
            // simulate latency
                try {
                    Thread.sleep(1000);
                } catch (Exception e) {
                }
                for (T t : ts) {
                    s.onNext(t);
                }
                s.onCompleted();
            }).subscribeOn(Schedulers.io());
        // note the use of subscribeOn to make an otherwise synchronous Observable async
    }

    public static class TileResponse {

        public TileResponse(Tile t, Reviews r, String u) {
            // store the values
        }

    }

    public static class Tile {

        private final int id;

        public Tile(int i) {
            this.id = i;
        }

        public int getSellerId() {
            return id;
        }

        public int getProductId() {
            return id;
        }

    }

    public static class Reviews {

    }
}

This outputs:

Search started  => 65ms
Search completed  => 1094ms
getProductImage[1] completed  => 2095ms
getSellerReviews[2] completed  => 2095ms
getProductImage[3] completed  => 2095ms
zip[1] completed  => 2096ms
zip[2] completed  => 2096ms
getProductImage[2] completed  => 2096ms
getSellerReviews[1] completed  => 2096ms
zip[3] completed  => 2096ms
All Tiles Completed  => 2097ms
getSellerReviews[3] completed  => 2097ms

I have made each IO call be simulated to take 1000ms so it is obvious where the latency is and that it is happening in parallel. It prints out the progress is makes in elapsed milliseconds.

The trick here is that flatMap merges async calls, so as long as the Observables being merged are async, they will all be executed concurrently.

If a call like getProductImage(t.getProductId()) was synchronous, it can be made asynchronous like this: getProductImage(t.getProductId()).subscribeOn(Schedulers.io).

Here is the important part of the above example without all the logging and boilerplate types:

    Observable searchTile = getSearchResults("search term");;

    Observable populatedTiles = searchTile.flatMap(t -> {
        Observable reviews = getSellerReviews(t.getSellerId());
        Observable imageUrl = getProductImage(t.getProductId());

        return Observable.zip(reviews, imageUrl, (r, u) -> {
            return new TileResponse(t, r, u);
        });
    });

    List allTiles = populatedTiles.toList()
            .toBlocking().single();

I hope this helps.