Reactive Java (33rd Degree)

TOMASZ KOWALCZEWSKI
REACTIVE JAVA

 Reactive view of the world
 Desgining interface for reactive interactions (the cat experiment)
 Rx Java as implementation of that interface
 Lessons learned
STUFF I WILL TALK ABOUT

REACTIVE
“readily responsive to a stimulus”
Merriam-Webster dictionary

SYNCHRONOUS PULL COMMUNICATION
Client Server
Request
Response
Server processing
Network latency

SERVICE RETURNING OBSERVABLE
public interface ShrödingersCat {
boolean alive();
}
Future<Boolean> alive();
}
Iterator<Boolean> alive();
}

SUBCRIPTIONS AND EVENTS
t
subscribe
onNext*
onCompleted | onError

PULL VS PUSH
Observer Observable
Subscribe
...
onNext
Server processing
Network latency
Maybe this one needs all the data...

RX JAVA BY NETFLIX
 Open source project with Apache License.
 Java implementation of Rx Observables from Microsoft
 The Netflix API uses it to make the entire service layer asynchronous
 Provides a DSL for creating computation flows out of asynchronous sources
using collection of operators for filtering, selecting, transforming and
combining that flows in a lazy manner
 These flows are called Observables – collection of events with push
semantics (as oposed to pull in Iterator)
 Targets the JVM not a language. Currently supports Java, Groovy, Clojure,
and Scala

OBSERVABLE
Observable<Boolean> alive();
}

}
cat
.alive()
.subscribe(status -> System.out.println(status));

}
cat
.alive()
.throttleWithTimeout(250, TimeUnit.MILLISECONDS)
.distinctUntilChanged()
.filter(isAlive -> isAlive)
.map(Boolean::toString)
.subscribe(status -> display.display(status));

 Maybe it executes its logic on subscriber thread?
 Maybe it delegates part of the work to other threads?
 Does it use NIO?
 Maybe its an actor?
 Does it return cached data?
 Observer does not care!
HOW IS THE OBSERVABLE IMPLEMENTED?

Observable<ShrödingersCat> cats = listAllCats();
cats
.flatMap(cat ->
Observable
.from(catService.getPicturesFor(cat))
.filter(image -> image.size() < 100 * 1000)
)
).subscribe();
FLATMAP(FUNC)

CACHE
Random random = new Random();
Observable<Integer> observable = Observable
.range(1, 100)
.map(random::nextInt)
.cache();
observable.subscribe(System.out::println);
observable.subscribe(System.out::println);
...
 Always prints same values

INJECTING CUSTOM OPERATORS USING LIFT
class InternStrings implements Observable.Operator<String, String> {
public Subscriber<String> call(Subscriber<String> subscriber) {
return new Subscriber<String>() {
public void onCompleted() { subscriber.onCompleted(); }
public void onError(Throwable e) { subscriber.onError(e); }
public void onNext(String s) {
subscriber.onNext(s.intern()); };
}
}
Observable.from("AB", "CD", "AB", "DE")
.lift(new InternStrings())
.subscribe();
 Valuable for instrumentation
 Inject debug code – see rxjava-contrib/rxjava-debug
 Inject performance counters

ERROR HANDLING
 Correctly implemented observable will not produce any events after
error notification
 Operators available for fixing observables not adhering to this rule
 Pass custom error handling function to subscribe
 Transparently substite failing observable with another one
 Convert error into regular event
 Retry subscription in hope this time it will work...

ESCAPING THE MONAD
Iterable<String> strings = Observable.from(1, 2, 3, 4)
.map(i -> Integer.toString(i))
.toBlockingObservable()
.toIterable();
// or (and many more)
T firstOrDefault(T defaultValue, Func1 predicate)
Iterator<T> getIterator()
Iterable<T> next()
 Inverses the dependency, will wait for next item, then execute
 Usually to interact with other, synchronous APIs
 While migrating to reactive approach in small increments
 To trigger early evaluation while debugging

OBSERVER
public interface Observer<T> {
void onCompleted();
void onError(Throwable e);
void onNext(T args);
}

CREATING OBSERVABLES
Observable<Boolean> watchTheCat =
Observable.create(observer -> {
observer.onNext(cat.isAlive());
observer.onCompleted();
});
 create accepts OnSubscribe function
 Executed for every subscriber upon subscription
 This example is not asynchronous

Future<?> brighterFuture = executorService.submit(() -> {
});
subscriber.add(Subscriptions.from(brighterFuture));
});
 Executes code in separate thread (from thread pool executorService)
 Stream of events is delivered by the executor thread
 Thread calling onNext() runs all the operations defined on observable
 Future is cancelled if client unsubscribes

Observable<Boolean> watchTheCat =
})
.subscribeOn(scheduler);
 Subscribe function is executed on supplied scheduler (thin wrapper
over java.util.concurrent.Executor)

SUBSCRIPTION
public interface Subscription {
void unsubscribe();
boolean isUnsubscribed();
}

UNSUBSCRIBING
Observable.create(subscriber -> {
for (long i = 0; !subscriber.isUnsubscribed(); i++) {
subscriber.onNext(i);
System.out.println("Emitted: " + i);
}
subscriber.onCompleted();
})
.take(10)
.subscribe(aLong -> {
System.out.println("Got: " + aLong);
});
 Take operator unsubscribes from observable after 10 iterations

CONCURRENCY
 Synchronous vs. asynchonous, single or multiple threaded is
implementation detail of service provider (Observable)
 As long as onNext calls are not executed concurrently
 So the framework does not have to synchronize everything
 Operators combining many Observables ensure serialized access
 In face of misbehaving observable serialize() operator forces
correct behaviour
 Passing pure functions to Rx operators is always the best bet

LESSONS LEARNED
 In our use cases performance profile is dominated by other system
components
 Performance depends on implementation of used operators and
may vary
 Contention points on operators that merge streams
 Carelessly creating 1000s threads (one for each task) when
NewThreadScheduler used. Reaching `ulimit –u` - and system
almost freezes :)
 Current version (0.19) has very sane defaults though
 Debugging and reasoning about subscriptions is not always easy.
 Insert doOnEach or doOnNext calls for debugging
 IDE support not satisfactory, problems in placing breakpoints inside
closures – IntelliJ IDEA 13 has smart step into closures which my
help

MORE INFORMATION
 https://github.com/Netflix/RxJava
 https://github.com/Netflix/RxJava/wiki
 http://www.infoq.com/author/Erik-Meijer
 React conference
http://www.youtube.com/playlist?list=PLSD48HvrE7-
Z1stQ1vIIBumB0wK0s8llY
 Cat picture taken from http://www.teckler.com/en/Rapunzel

source:
flatmapthatshit.com
REMEBER: DON’T ITERATE - FLATMAP

Reactive Java (33rd Degree)

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Reactive Java (33rd Degree) (20)

More from Tomasz Kowalczewski (11)

Recently uploaded (20)

Reactive Java (33rd Degree)

Editor's Notes