Over a million developers have joined DZone.
{{announcement.body}}
{{announcement.title}}

Eager Subscription: RxJava FAQs

DZone's Guide to

Eager Subscription: RxJava FAQs

Learning RxJava can be a challenge, what with it being a fairly substantial shift in how you think. In this post, we take a look at some FAQs that come up.

· Java Zone ·
Free Resource

How do you break a Monolith into Microservices at Scale? This ebook shows strategies and techniques for building scalable and resilient microservices.

While teaching and mentoring RxJava, as well as after authoring a book, I noticed some areas are especially problematic. I decided to publish a bunch of short tips that address the most common pitfalls. This is the first part.

Observables and Flowables are lazy by nature. This means that no matter what heavy or long-running logic you place inside your Flowable, it will get evaluated only when someone subscribes. And also as many times as someone subscribes. This is illustrated by the following code snippet:

private static String slow() throws InterruptedException {
    logger.info("Running");
    TimeUnit.SECONDS.sleep(1);
    return "abc";
}

//...

Flowable<String> flo = Flowable.fromCallable(this::slow);
logger.info("Created");
flo.subscribe();
flo.subscribe();
logger.info("Done");


Such an Observable or Flowable will inevitably print:

19:37:57.368 [main] - Created
19:37:57.379 [main] - Running
19:37:58.383 [main] - Running
19:37:59.388 [main] - Done


Notice that you pay the price of sleep() twice (double subscription). Moreover, all logic runs in the client (main) thread — there is no implicit threading in RxJava unless requested with subscribeOn() or implicitly available with asynchronous streams. The question is: Can we force running subscription logic eagerly so that whenever someone subscribes, the stream is already precomputed, or at least the computation has started?

Totally Eager Evaluation

The most obvious, but flawed, solution is to eagerly compute whatever the stream returns and simply wrap it with a fixed Flowable:

Flowable<String> eager() {
    final String slow = slow();
    return Flowable.just(slow);
}


Unfortunately, this substantially defeats the purpose of RxJava. First of all, operators like subscribeOn() no longer work and it becomes impossible to off-load computation to a different thread. Even worse, even though eager() returns a Flowable, it will always, by definition, block client threads. It is harder to reason, compose, and manage such streams. You should generally avoid such patterns and prefer lazy-loading, even when eager evaluation is necessary.

Using cache() Operator

The next example does just that with cache() operator:

Flowable<String> eager3() throws InterruptedException {
    final Flowable<String> cached =
        Flowable
            .fromCallable(this::slow)
            .cache();
    cached.subscribe();
    return cached;
}


The idea is simple: wrap computation with a lazy Flowable and make it cached. The cache() operator remembers all emitted events upon the first subscription so that when the second Subscriber appears, it will receive the same cached sequence of events. However, the cache() operator (like most others) is lazy, so we must forcibly subscribe for the first time. Calling subscribe() will prepopulate the cache. Moreover, if the second subscriber appears before the slow() computation finishes, it will wait for it as well, rather than starting it for the second time.

This solution works, but keep in mind that subscribe() will actually block because no Scheduler was involved. If you want to prepopulate your Flowable in the background, try subscribeOn():

Flowable<String> eager3() throws InterruptedException {
    final Flowable<String> cached =
        Flowable
            .fromCallable(this::slow)
            .subscribeOn(justDontAlwaysUse_Schedulers.io())
            .cache();
    cached.subscribe();
    return cached;
}


Yes, using Schedulers.io() is problematic and hard to maintain on production systems, so please avoid it in favor of custom thread pools.

Error Handling

Sadly it's surprisingly easy to swallow exceptions in RxJava. That's what can happen in our last example if the slow() method fails. The exception isn't swallowed entirely, but by default, if no one was interested, its stack trace is printed on System.err. Also, the unhandled exception is wrapped with OnErrorNotImplementedException. Not very convenient and most likely lost if you are doing any form of centralized logging. You can use the doOnError() operator for logging, but it still passes the exception downstream, and RxJava considers it unhandled as well, once more time wrapping with OnErrorNotImplementedException. So let's implement an onError callback in subscribe():

Flowable<String> eager3() throws InterruptedException {
    final Flowable<String> cached =
        Flowable
            .fromCallable(this::slow)
            .cache();
    cached.subscribe(
            x -> {/* ignore */},
            e -> logger.error("Prepopulation error", e));
    return cached;
}


We don't want to handle actual events, just errors in subscribe(). At this poin, you can safely return such Flowables. It's eager and, chances are, that whenever you subscribe to it, data will already be available. Notice that, for example, the observe() method from Hystrix is eager as well, as opposed to toObservable(), which is lazy. The choice is yours.

How do you break a Monolith into Microservices at Scale? This ebook shows strategies and techniques for building scalable and resilient microservices.

Topics:
rxjava ,subscription ,java ,lazy evaluation ,tutorial

Published at DZone with permission of

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}