Java 8 Streams and Rx Java Comparison

@JosePaumard
RxJava
J8 Stream
Comparison:
patterns
performances

Why should we be
interested in the
Streams API?

@JosePaumard#Devoxx #J8Stream
It’s all about data processing
Up to 2014: only one tool, the Collection Framework
Also 3rd party API: Common Collections, …

Up to 2014: only one tool, the Collection Framework
Also 3rd party API: Common Collections, …
From 2014: so many new APIs

Why so?
Because data processing is more and more important

Why so?
and more and more complex!

Why so?
bigger and bigger amount of data to process

Why so?
controlled response time

Why so?
controlled response time
complex algorithms

So data processing needs high level primitives

Able to access data wherever it is

That provides map / filter / reduce functions

And efficient implementations of those!

And efficient implementations of those!
Most probably implemented in parallel

Agenda
1) To present two API: the Java 8 Stream API and RxJava
• Fundamentals
• Implemented functionalities
• Patterns!

Agenda
1) To present two API: the Java 8 Stream API and RxJava
• Fundamentals
• Implemented functionalities
• Patterns!
2) And to compare those APIs
• From the developer point of view
• Performances!

Questions?
#J8Stream

API Stream
What is a Java 8 Stream?
An object that connects to a source

API Stream
That does not hold any data

API Stream
That implements the map / filter / reduce pattern

API Stream
That implements the map / filter / reduce pattern
A new concept in JDK 8

Definition of a Stream
Two things about streams:
1) A Stream does not hold any data
2) A Stream does not modify the data it gets from the source

API Stream
Example
List<String> list = Arrays.asList("one", "two", "three") ;
list.stream()
.map(s -> s.toUpperCase())
.max(Comparator.comparing(s -> s.length()))
.ifPresent(s -> System.out.println(s)) ;

API Stream
Example
list.stream() // creation of a new Stream object

API Stream
Example
list.stream()
.map(s -> s.toUpperCase()) // to upper case
.filter(s -> s.length() < 20)

API Stream
Example
list.stream()
.filter(s -> s.length() < 20) // remove strings longer than 20

API Stream
Example
list.stream()
.max(Comparator.comparing(s -> s.length())) // take the longest s

API Stream
Example
list.stream()
.ifPresent(s -> System.out.println(s)) ; // and print the result

API Stream
Example
list.stream()
.map(String::toUpperCase)
.max(Comparator.comparing(String::length))
.ifPresent(System.out::println) ; // and print the result

Collectors
We can use collectors
List<Person> list = ... ;
list.stream()
.filter(person -> person.getAge() > 30)
.collect(
Collectors.groupingBy(
Person::getAge, // key extractor
Collectors.counting() // downstream collector
)
) ;

Collectors
Map<
list.stream()
.collect(
)
) ;

Collectors
Map<Integer,
list.stream()
.collect(
)
) ;

Collectors
Map<Integer, Long> map =
list.stream()
.collect(
)
) ;

Collectors
And we can go parallel!
list.stream().parallel()
.collect(
)
) ;

Sources of a Stream
Many ways of connecting a Stream to a source of data

Sources of a Stream
First patterns to create a Stream
List<Person> people = Arrays.asList(p1, p2, p3);

Sources of a Stream
Stream<Person> stream = people.stream();

Sources of a Stream
Stream<Person> stream = people.stream();
Stream<Person> stream = Stream.of(p1, p2, p3);

More patterns
Stream on String
String crazy = "supercalifragilisticexpialidocious";
IntStream letters = crazy.chars();
long count = letters.distinct().count();

More patterns
Stream on String
long count = letters.distinct().count();
> count = 15

More patterns
Stream on String
letters.boxed().collect(
Function.identity(),
Collectors.counting()
)
);

More patterns
Stream on String
Map<
)
);

More patterns
Stream on String
Map<Integer,
)
);

More patterns
Stream on String
)
);

More patterns
Stream on String
)
);
a -> 3
c -> 3
d -> 1
e -> 2
f -> 1
g -> 1
i -> 7
l -> 3
o -> 2
p -> 2
r -> 2
s -> 3
t -> 1
u -> 2
x -> 1

More patterns
Stream built on the lines of a file
String book = "alice-in-wonderland.txt";
Stream<String> lines = Files.lines(Paths.get(book)); // autocloseable

More patterns
Stream built on the lines of a file
Stream built the words of a String
String book = "alice-in-wonderland.txt";
Stream<String> lines = Files.lines(Paths.get(book)); // autocloseable
String line = "Alice was beginning to get very tired of";
Stream<String> words = Pattern.compile(" ").splitAsStream(line);

Flatmap
How to mix both to get all the words of a book?
Memory-efficient implementations built on lazy reading of the
source of data
Function<String, Stream<String>> splitToWords =
line -> Pattern.compile(" ").splitAsStream(line);
Stream<String> lines = Files.lines(path);
Stream<String> words = lines.flatMap(splitToWords);

Flatmap
Analyzing the result
long count =
words.filter(word -> word.length() > 2)
.map(String::toLowerCase)
.distinct()
.count();

Flatmap
Another analysis of the result
words.filter(word -> word.length() > 2)
.map(String::toLowerCase)
// .distinct()
.collect(
String::length,
)
);

Extracting the max from a Map
Yet another analysis of the result
map.entrySet().stream()
.sorted(
Entry.<Integer, Long>comparingByValue().reversed()
)
.limit(3)
.forEach(System.out::println);

Connecting a Stream on a source
Connecting a Stream on a non-standard source is possible

Connecting a Stream on a non-standard source is possible
A Stream is built on 2 things:
- A Spliterator (comes from split and iterator)
- A ReferencePipeline (the implementation)

The Spliterator is meant to be overriden
public interface Spliterator<T> {
boolean tryAdvance(Consumer<? super T> action) ;
Spliterator<T> trySplit() ;
long estimateSize();
int characteristics();
}

The Spliterator is meant to be overriden
boolean tryAdvance(Consumer<? super T> action) ;
Spliterator<T> trySplit(); // not needed for non-parallel operations
long estimateSize(); // can return 0
int characteristics(); // returns a constant
}

Examples of custom Spliterators
Suppose we have a Stream:
[1, 2, 3, 4, 5, …]
We want to regroup the elements:
[[1, 2, 3], [4, 5, 6], [7, 8, 9], …]
Let us build a GroupingSpliterator to do that

GroupingSpliterator
[1, 2, 3, 4, 5, …] -> [[1, 2, 3], [4, 5, 6], …]
public class GroupingSpliterator<E> implements Spliterator<Stream<E>>
{
private final long grouping ;
private final Spliterator<E> spliterator ;
// implementation
}

GroupingSpliterator
[1, 2, 3, 4, 5, …] -> [[1, 2, 3], [4, 5, 6], …]
public class GroupingSpliterator<E> implements Spliterator<Stream<E>>
{
private final long grouping ;
private final Spliterator<E> spliterator ;
// implementation
}
GroupingSpliterator<Integer> gs =
new GroupingSpliterator(spliterator, grouping);

GroupingSpliterator
Implementing estimateSize() and characteristics()
public long estimateSize() {
return spliterator.estimateSize() / grouping ;
}
public int characteristics() {
return this.spliterator.characteristics();
}

GroupingSpliterator
Implementing trySplit()
public Spliterator<Stream<E>> trySplit() {
Spliterator<E> spliterator = this.spliterator.trySplit() ;
return new GroupingSpliterator<E>(spliterator, grouping) ;
}

GroupingSpliterator
Implementing tryAdvance()
public boolean tryAdvance(Consumer<? super Stream<E>> action) {
// should call action.accept() with the next element of the Stream
// and return true if more elements are to be consumed
return true ; // false when we are done
}

GroupingSpliterator
The structure of the resulting Stream is the following:
[[1, 2, 3], [4, 5, 6], [7, 8, 9], …]
Each element is a Stream built on the elements of the
underlying Stream

GroupingSpliterator
Build a Stream element by element is done with a
Stream.Builder
Stream.Builder<E> builder = Stream.builder() ;
for (int i = 0 ; i < grouping ; i++) {
spliterator.tryAdvance(element -> builder.add(element);
}
Stream<E> subStream = subBuilder.build(); // [1, 2, 3]

GroupingSpliterator
Are we done with the underlying Stream?
spliterator.tryAdvance(
element -> builder.add(element)
);
}

GroupingSpliterator
spliterator.tryAdvance( // when this call returns false
element -> builder.add(element)
);
}

GroupingSpliterator
boolean finished = false;
if (spliterator.tryAdvance(element -> builder.add(element)))
finished = true;
}

GroupingSpliterator
boolean finished = false;
if (spliterator.tryAdvance(element -> builder.add(element)))
finished = true;
}
action.accept(subStream) ;
return !finished ;
}

RollingSpliterator
What about building a Stream like this one:
[[1, 2, 3], [2, 3, 4], [3, 4, 5], …]
This time we need a ring buffer

RollingSpliterator
The tryAdvance() call on the underlying Stream becomes this:
bufferWriteIndex is an AtomicLong
private boolean advanceSpliterator() {
return spliterator.tryAdvance(
element -> {
buffer[bufferWriteIndex.get() % buffer.length] = element ;
bufferWriteIndex.incrementAndGet() ;
});
}

RollingSpliterator
Building the element streams from the ring buffer:
private Stream<E> buildSubstream() {
Stream.Builder<E> subBuilder = Stream.builder() ;
subBuilder.add(
(E)buffer[(i + bufferReadIndex.get()) % buffer.length]
) ;
}
bufferReadIndex.incrementAndGet() ;
Stream<E> subStream = subBuilder.build() ;
return subStream ;
}

RollingSpliterator
Putting things together to build the tryAdvance() call

boolean finished = false ;
if (bufferWriteIndex.get() == bufferReadIndex.get()) {
if (!advanceSpliterator()) {
finished = true ;
}
}
}
if (!advanceSpliterator()) {
finished = true ;
}
Stream<E> subStream = buildSubstream() ;
action.accept(subStream) ;
return !finished ;
}

Spliterator on Spliterator
We saw how to build a Stream on another Stream by
rearranging its elements
What about building a Stream by merging the elements of
other Streams?

Spliterator on Spliterators
Let us take two Streams:
[1, 2, 3, …]
[a, b, c, …]
And build a ZippingSpliterator:
[F[1, a], F[2, b], F[3, c], …]

What about
estimateSize()
trySplit()
characteristics()?
They are the same as the underlying streams

We then need to implement tryAdvance()
Where transform is a BiFunction
public boolean tryAdvance(Consumer<? super R> action) {
return spliterator1.tryAdvance(
e1 -> {
spliterator2.tryAdvance(e2 -> {
action.accept(tranform.apply(e1, e2)) ;
}) ;
}) ;
}

ZippingSpliterator
What about creating a Builder for this Spliterator?
ZippingSpliterator.Builder<String, String, String> builder =
new ZippingSpliterator.Builder();
ZippingSpliterator<String,String,String> zippingSpliterator = builder
.with(spliterator1)
.and(spliterator2)
.mergedBy((e1, e2) -> e1 + " - " + e2)
.build();

ZippingSpliterator
The complete pattern:
Stream<String> stream1 = Stream.of("one", "two", "three");
Stream<Integer> stream2 = Stream.of(1, 2, 3);
Spliterator<String> spliterator1 = stream1.spliterator();
Spliterator<Integer> spliterator2 = stream2.spliterator();
Stream<String> zipped =
StreamSupport.stream(zippingSpliterator, false);

What did we do with Streams so far?
We took one stream and built a stream by regrouping its
elements in some ways

We took to streams and we merged them, element by element,
using a bifunction

using a bifunction
What about taking one element at a time, from different
streams?

The WeavingSpliterator
We have N streams, and we want to build a stream that takes:
- its 1st element from the 1st stream
- its 2nd element from the 2nd stream, etc…

The estimateSize():
public long estimateSize() {
int size = 0 ;
for (Spliterator<E> spliterator : this.spliterators) {
size += spliterator.estimateSize() ;
}
return size ;
}

The tryAdvance():
private AtomicInteger whichOne = new AtomicInteger();
public boolean tryAdvance(Consumer<? super E> action) {
return spliterators[whichOne.getAndIncrement() %
spliterators.length]
.tryAdvance(action);
}

One last thing about Spliterators
The characteristics() method is in fact quite important
What does « characteristics » mean for a Spliterator?

The Spliterator
It is in fact a special word: characteristics
public static final int ORDERED = 0x00000010;
public static final int DISTINCT = 0x00000001;
public static final int SORTED = 0x00000004;
public static final int SIZED = 0x00000040;
public static final int NONNULL = 0x00000100;
public static final int IMMUTABLE = 0x00000400;
public static final int CONCURRENT = 0x00001000;
public static final int SUBSIZED = 0x00004000;
}

The Spliterator
The Spliterator holds a special word: characteristics
// ArrayListSpliterator
return Spliterator.ORDERED | Spliterator.SIZED |
Spliterator.SUBSIZED;
}
// HashMap.KeySpliterator
return (fence < 0 || est == map.size ? Spliterator.SIZED : 0) |
Spliterator.DISTINCT;
}

The Spliterator
This word is used for optimization
people.stream()
.sorted() // quicksort?
.collect(Collectors.toList());

The Spliterator
people.stream()
.sorted() // quicksort? It depends on SORTED == 0

The Spliterator
SortedSet<Person> people = ...;
people.stream()
.sorted() // SORTED == 1, no quicksort

The Spliterator
ArrayList<Person> people = ...;
people.stream()
.sorted() // SORTED == 0, quicksort

The characteristics can change
Each Stream object in a pipeline has its own characteristics

The characteristics can change
Each Stream object in a pipeline has its own characteristics
Method Set to 0 Set to 1
filter() SIZED -
map() DISTINCT, SORTED -
flatMap() DISTINCT, SORTED, SIZED -
sorted() - SORTED, ORDERED
distinct() - DISTINCT
limit() SIZED -
peek() - -
unordered() ORDERED -

Back to the Spliterator itself
What did we do with the spliterator, and with the Stream built
on it?

using a bifunction
We took a collection of streams and built a stream by taking
elements from them, in a given order

Wrap-up for the Stream API
The Java 8 Stream API is not just about implementing map /
filter / reduce or building hashmaps
We can use it to manipulate data in advanced ways:
- by deciding on what source we want to connect
- by deciding how we can consume the data from that source

RxJava
Open source API, available on Github
Developed by Netflix
RxJava is the Java version of ReactiveX
.NET
Python, Kotlin, JavaScript, Scala, Ruby, Groovy, Rust
Android
https://github.com/ReactiveX/RxJava

RxJava
RxJava is not an alternative implementation of Java 8
Streams, nor the Collection framework
It is an implementation of the Reactor pattern

RxJava
The central class is the Observable class

RxJava
It’s big: ~10k lines of code

RxJava
It’s complex: ~100 static methods, ~150 non-static methods

RxJava
It’s complex: ~100 static methods, ~150 non-static methods
It’s complex: not only because there is a lot of things in it, but
also because the concepts are complex

RxJava
Interface Observer
used to « observe » an observable

RxJava
Interface Observer
used to « observe » an observable
Interface Subscription
used to model the link between an observer and an
observable

Interface Observer
A simple interface
public interface Observer<T> {
public void onNext(T t);
public void onCompleted();
public void onError(Throwable e);
}

How to subscribe
Subscribing to an observable
Observable<T> observable = ... ;
Subscription subscription = observable.subscribe(observer) ;

How to subscribe
Subscribing to an observable
Subscription subscription = observable.subscribe(observer) ;
public interface Subscription {
public void unsubscribe();
public void isUnsubscribe();
}

How to subscribe
We can also just declare a callback
Observable<T> next = observable.doOnNext(System.out::println) ;
Observable<T> error = observable.doOnError(System.out::println) ;
Observable<T> each = observable.doOnEach(System.out::println) ;

RxJava – agenda
Patterns to create Observables
How to merge observables together
Hot observables / backpressure

How to create
an Observable

The usual ways
Observable from collections and arrays
Observable<String> obs1 = Observable.just("one", "two", "three") ;
List<String> strings = Arrays.asList("one", "two", "three") ;
Observable<String> obs2 = Observable.from(strings) ;

The usual ways
Observable useful for tests
never(): never emits anything
Observable<String> empty = Observable.empty() ;
Observable<String> never = Observable.never() ;
Observable<String> error = Observable.<String>error(exception) ;

The usual ways
Series and time series
Observable<Long> longs = Observable.range(1L, 100L) ;
// interval
Observable<Long> timeSerie1 =
Observable.interval(1L, TimeUnit.MILLISECONDS) ; // serie of longs
// initial delay, then interval
Observable<Long> timeSerie2 =
Observable.timer(10L, 1L, TimeUnit.MILLISECONDS) ; // one 0

The usual ways
The using() method
public final static <T, Resource> Observable<T> using(
final Func0<Resource> resourceFactory, // producer
final Func1<Resource, Observable<T>> observableFactory, // function
final Action1<? super Resource> disposeAction // consumer
) { }

How using() works
The using() method
1) Creates a resource that will provide the elements
2) Uses this resource to create the elements
3) Applies the 3rd argument to the resource
This 3rd element is in fact there to close the resource

How using() works
Suppose we want to create an observable on the lines of a text
file
1) We need a BufferedReader
2) The we need to wrap the lines one by one in Observables
3) Then we need to close the BufferedReader

How using() works
Let us do that:
Func0<BufferedReader> resourceFactory =
() -> new BufferedReader(new FileReader(new File(fileName)));
Action1<BufferedReader> disposeAction =
br -> br.close();

How using() works
The problem is that… we need to handle exceptions 
Func0<BufferedReader> resourceFactory =
() -> {
try {
return new BufferedReader(
new FileReader(new File(fileName)));
} catch (Exception e) {
e.printStackTrace();
}
return null;
};

How using() works
How can we create one Observable by file line?

How using() works
We could read the file line by line, put the lines in a Collection,
then build an Observable on that collection

How using() works
We could read the file line by line, put the lines in a Iterator,
then build an Observable on that iterator

How using() works
We could read the file line by line, put the lines in a Iterator,
then build an Observable on that iterator
Creating an Observable from an Iterator:
Observable.from(() -> iterator);

Func1<BufferedReader, String> observableFactory = bufferedReader -> {
Iterator<String> readerIterator = new Iterator<String>() {
private String currentLine = null;
private boolean finished;
{
currentLine = bufferedReader.readLine();
finished = currentLine == null;
}
public boolean hasNext() {
return !finished;
}
public String next() {
String line = currentLine;
currentLine = bufferedReader.readLine();
finished = currentLine == null;
return line;
}
};
return Observable.from(() -> readerIterator);
}

How using() works
The using() method
Made to build an Observable on a resource that needs to be
« closed » at the end of the generation of the stream

How using() works
The using() method
Made to build an Observable on a resource that needs to be
« closed » at the end of the generation of the stream
It is not built on the auto closeable mechanism because
RxJava aims to be Java 6 compatible

Schedulers

RxJava & executors
Some of those methods take a further argument
This Observable is executed in this scheduler
Scheduler is an interface (in fact an abstract class, to define
« default methods » in Java 7)
Schedulers should be used to create schedulers
Observable<Long> longs = Observable.range(0L, 100L, scheduler) ;

RxJava & executors
It allows for the execution of Observable in specialized pools of
Threads
Some Observables are executed in special schedulers (cf
Javadoc)

Schedulers
Factory Schedulers
public final class Schedulers {
public static Scheduler immediate() {...} // immediate
public static Scheduler newThread() {...} // new thread
public static Scheduler trampoline() {...} // queued in the current
// thread
}

Schedulers
Factory Schedulers
public final class Schedulers {
public static Scheduler computation() {...} // computation ES
public static Scheduler io() {...} // IO growing ES
public static Scheduler test() {...}
public static Scheduler from(Executor executor) {...}
}

Schedulers
Schedulers can be seen as Executors
Some of them are specialized (IO, Computation)
Observers are called in the thread that runs the Observable

Schedulers
The Schedulers.from(executor) is useful to call observers in
certain threads
For instance:
Scheduler swingScheduler =
Schedulers.from(
SwingUtilities::invokeLater
);

A 1st example
A simple example
Observable<Integer> range1To100 = Observable.range(1L, 100L) ;
range1To100.subscribe(System.out::println) ;

A 1st example
A simple example
Observable<Integer> range1To100 = Observable.range(1L, 100L) ;
range1To100.subscribe(System.out::println) ;
> 1 2 3 4 ... 100

A 2nd example
A not so simple example
Observable<Integer> timer = Observable.timer(1, TimeUnit.SECONDS) ;
timer.subscribe(System.out::println) ;

A 2nd example
A not so simple example
Nothing is printed
timer.subscribe(System.out::println) ;
>

A 2nd example
Let us modify this code
timer.subscribe(() -> {
System.out.println(Thread.currentThread().getName() + " " +
Thread.currentThread().isDaemon()) ;
}) ;
Thread.sleep(2) ;

A 2nd example
Let us modify this code
timer.subscribe(() -> {
System.out.println(Thread.currentThread().getName() + " " +
Thread.currentThread().isDaemon()) ;
}) ;
Thread.sleep(2) ;
> RxComputationThreadPool-1 - true

About the 1st & 2nd examples
The first example ran in the main thread

The second example ran in a daemon thread, that does not
prevent the JVM from exiting. Nothing was printed out,
because it did not have the time to be executed.

The second example ran in a daemon thread, that does not
prevent the JVM from exiting. Nothing was printed out,
because it did not have the time to be executed.
Observable are ran in their own schedulers (executors)

Merging Observables

Merging Observable together
RxJava provides a collection of methods to merge observables
together

Merging Observable together
RxJava provides a collection of methods to merge observables
together
With many different, very interesting semantics

Representing an Observable
RxJava uses « marble diagrams » to represent observables
Observable

On this example, onNext() is called 5 times
Observable

Here an exception has been thrown, no data will be generated
after it, the onError() method is called
Observable

Here the end of the Stream has been reached, the
onComplete() method is called
Observable

The ambiguous operator
The first operator is amb(), that stands for « ambiguous »
It takes the first Observable that produced data
©RxJava
O1
O2
result

The zip operator
The zip operator takes one element from each Observable and
combine them using a function
©RxJava
O1
O2
F(O1, O2)

The combine latest operator
The combineLatest is a non-strict zip operator, emits an item
each time an observable emits one
©RxJava
O1
O2
F(O1, O2)

Concat and merge
Concat: emits O1 and then O2, without mixing them,
merge stops on the emission of an error
©RxJava©RxJava

Observables of Observables
The methods we saw are defined on Iterables of Observables
public final static <T> Observable<T> merge(
Iterable<Observable<T>> listOfSequences) { }

The methods we saw are defined on Iterables of Observables
They are also defined on Observables of Observables
Observable<Observable<T>> sequenceOfSequences) { }

And the marble diagrams are different
©RxJava

Then comes the switchOnNext operator
public final static <T> Observable<T> switchOnNext(
©RxJava

Hot and cold
Observables

Cold and hot Observable
A cold Observable emits items if it is observered

A cold Observable emits items if it is observered
A hot Observable emits items when it is created

A 3rd example
Let us see an example
Observable<Long> timer = Observable.interval(1, TimeUnit.SECONDS);
Observable<String> manyStrings =
Observable.combineLatest(
timer, Observable.just("one"),
(i, s) -> i + " - " + s).forEach(System.out::println);

A 3rd example
> 0 - one
1 - one
2 – one ...

A 3rd example
Thread.sleep(5500);

A 3rd example
Thread.sleep(5500);
> 0 - one
1 - one
2 – one ...

A 3rd example
It means that the timer does not increment itself if not
observed
This Observable is cold

A 3rd example
It means that the timer does not increment itself if not
observed
This Observable is cold
How can we make it hot?

A 3rd example
Making it hot:
Observable<String> moreManyStrings =
timer, Observable.just("two"),

A 3rd example
It prints out the following:
> 0 - one
0 - two
1 – one
1 - two
2 – one
2 – two ...

A 3rd example
Making it hot:Observable<Long> timer = Observable.interval(1, TimeUnit.SECONDS);
Thread.sleep(5500);
Observable<String> moreManyStrings =
timer, Observable.just("two"),

A 3rd example
It then prints out the following!
> 0 - one
1 - one
2 - one
3 - one
4 - one
5 - one
5 - two
6 - one
6 - two
7 - one
7 – two ...

So in its first use, timer is a cold Observable
And in the second one, it becomes a hot Observable
If used more than once, a cold Observable will be seen hot by
the 2nd and later observers

Problem: what happens if we have many observers to
subscribe?

Problem: what happens if we have many observers to
subscribe?
Our application could miss the first values emitted by an
observable
How can we fix that?

We can use the cache method
cache(int capacity) { }
O1
O2

A cached timer
Creating a cached timer:
Observable<Long> cachedTimer = timer.cache(1_000);
...

ConnectableObservable
We can do better and use a ConnectableObservable
(which is an Observable)
// does not count as a subscription
ConnectableObservable<Long> publish = observable.publish();
// take the time to connect all the observers
// then call
publish.connect();

In fact the Observable returned by the connect() call is itself a
hot Observable
So this is also a way to create a hot Observable

Coffe break!Map, filter, reduce, collect
Repeating & retrying
Dealing with time
Backpressure
Combining Java 8 Streams
& RxJava
Comparisons
The future: reactive
support in Java 9

Map, filter,
reduce and collect

Map and filter
Map / filter, return an Observable
Emits only the elements that match the predicate
map(Func1<T, R> mapping) { }
cast(Class<R> clazz) { } // casts the elements
filter(Func1<T, Boolean> predicate) { }

Materializing
Materialize: special type of mapping
Emit a Notification object that wraps the item, with metadata
Useful for logging
Does the reverse:
materialize() { } // Observable<Notification<T>>
dematerialize() { } // Observable<T>

Selecting ranges
Selecting ranges of elements
first() { }
last() { }
skip(int n) { }
limit(int n) { }
take(int n) { }

Special filtering
Special functions
Emits a true then complete on the onComplete
if the Observable emitted at least one item
exists(Func1<T, Boolean> predicate) { }

Special filtering
Special functions
Emits the nth item then complete on the onNext
if the Observable emitted at least n items
elementAt(int n) { }

Special filtering
Special functions
Emits nothing then complete on the onComplete
ignoreElement() { }

Special filtering
Special functions
Emits the first item on the onComplete
or emits an error on the second item emitted
single() { }

Special filtering
Special functions
Emits the matching item if it has been seen, on onComplete
or emits an error if not, on onComplete
single(Func1<T, Boolean> predicate) { }

Reductions & collections
Classical reductions
all(Func1<T, Boolean> predicate) { } // Observable<Boolean>
count() { } // Observable<Long>
forEach(Action1<T> consumer) { } // void

Accumulations
Reduce: returns the reduction of all the items, on onComplete
Returns an error if the observable is empty
reduce(Func2<T, T, T> accumulator) { } // Observable<T>

Accumulations
Reduce: returns the reduction of all the items, on onComplete
In case the observable might be empty, use the other version
of reduce, that takes a seed
reduce(seed, Func2<T, T, T> accumulator) { } // Observable<T>

Accumulations
Scan: returns the reduction step by step, on onNext
Can also take a seed
scan(Func2<T, T, T> accumulator) { } // Observable<T>
reduce(seed, Func2<T, T, T> accumulator) { } // Observable<T>

Collecting data in a mutable container
Example: adding the elements to a list
collect(Func0<R> stateFactory, // Producer
Action2<R, T> collector) { } // BiConsumer
collect(ArrayList::new, // Producer
ArrayList::add) { } // BiConsumer

Ready to use collectors for lists:
toList() { } // Observable<List<T>>
toSortedList() { } // Observable<List<T>>

Ready to use collectors for lists:
And for maps (toMap can lose items):
toList() { } // Observable<List<T>>
toSortedList() { } // Observable<List<T>>
toMultimap(Func1<T, K> keySelector, // Observable<
Func1<T, V> valueSelector) { } // Map<K, Collection<T>>
toMap(Func1<T, K> keySelector) { } // Observable<Map<K, T>>

A special kind of map:
The returned observable could hold a single map,
but it does not
groupBy(Func1<T, K> keySelector) { } // function

but it does not
It holds GroupedObservable items, which is an Observable
with a key

but it does not
Instead of returning an Iterable of key / value pairs
It returns an Observable of key / value pairs

Observable<Person> people = ...;
// Observable<GroupedObservable<Integer, Person>>
people.groupBy(Person::getAge)

// Observable<GroupedObservable<Integer, Person>>
.filter(groupedObs -> groupedObs.getKey() > 20)

// GroupedObservable<Integer, Person> extends Observable<Person>
.flatMap(groupedObs -> groupedObs)

// Observable<Integer>
.flatMap(groupedObs -> groupedObs)
.count();

Wrap-up on map / filter / reduce
All those methods return an Observable
The result is thus wrapped in an Observable
Chaining operations is made with flatMap()
Wrapping and unwrapping has a cost…

Repeating and
Retrying

The repeat method repeats the same Observable
repeat() { } // Observable<T>
repeat(long times) { } // Observable<T>

The repeat method repeats the same Observable
repeat() { } // Observable<T>
repeat(long times) { } // Observable<T>
repeatWhen(
Func1<Observable<Void>>, Observable<?>> notificationHandler
) { }

repeat: will repeat the same Observable again and again
repeatWhen:
- on the onComplete of the source Observable, the
notification handler is passed an Observable
- if the notification handler calls the onError or onComplete of
this passed Observable, then repeatWhen calls the same on
the returned Observable

repeat: will repeat the same Observable again and again
On the returned Observable, one can invoke:
onNext(), that triggers the repetition
onComplete() or onError(), which will trigger the same call on
the source Observable

This function takes an Observable<Void>
And returns an Observable<?>

This function takes an Observable<Void>
And returns an Observable<?>
When the Observable to be repeated calls onComplete
Then the passed Observable calls onNext with null
As a response the returned Observable should call:
- onNext to repeat
- onComplete to stop repeating

The first thing we need is an Observable on which we can emit
items on demand

The first thing we need is an Observable on which we can emit
items on demand
For that, we need to implement the OnSubscribe interface
final Set<Subscriber<? super Void>> subscribers = new HashSet<>();
OnSubscribe<Void> onSubscribe = new OnSubscribe<Void>() {
public void call(Subscriber<? super Void> subscriber) {
subscribers.add(subscriber);
}
};

Then create an Observable from this onSubscribe object
final Observable<Void> returnedObservable =
Observable.create(onSubscribe);
final Set<Subscriber<? super Void>> subscribers = new HashSet<>();
OnSubscribe<Void> onSubscribe = new OnSubscribe<Void>() {
public void call(Subscriber<? super Void> subscriber) {
subscribers.add(subscriber);
}
};

Then create the notification handler
Func1 notificationHandler =
new Func1<Observable<Void>, Observable<Void>>() {
final Observable<Void> returnedObservable = ...;
public Observable<Void> call(Observable<Void> observable) {
// implementation
}
return returnedObservable;
};

observable.subscribe(new Observer<Void>() {
private int count = 0;
public void onCompleted() {
}
public void onError(Throwable e) {
subscribers.forEach(sub -> sub.onError(e));
}

int count = 0;
public void onNext(Void t) {
if (count == 3) {
subscribers.forEach(sub -> sub.onCompleted()); // stops
} else {
count++;
Thread.sleep(1_000);
subscribers.forEach(sub -> sub.onNext(null)); // repeat
}
}

The complete pattern
Observable<Long> timer =
Observable.interval(1, TimeUnit.MILLISECONDS).take(5);
timer.repeatWhen(
(Func1<? super Observable<? extends Void>, ? extends Observable<?>>)
notificationHandler)
.forEach(System.out::println);
Thread.sleep(5_000);

The retry method will reset the Observable on error
The retryThen() function works in the same way as for the
repeat
retry() { } // Observable<T>
retry(long times) { } // Observable<T>
retryWhen(
Func1<Observable<Throwable>>, Observable<?>> notificationHandler
) { }

There is also a retry family of methods, that works on onError
instead of onComplete

Joining
Joins to Observable, based on the overlapping of durations
join(Observable<TRight> right,
Func1<T, Observable<TLeftDuration>> leftDurationSelector,
Func1<TRight, Observable<TRightDuration>> rightDurationSelector,
Func2<T, TRight, R> resultSelector) { }

Join
©RxJava
Left obs.
Right obs.
F(O1, O2)

GroupJoin
groupJoin(
Observable<T2> right,
Func1<T, Observable<D1>> leftDuration,
Func1<T2, Observable<D2>> rightDuration,
Func2<T, <T2>, R> resultSelector)
{ }

GroupJoin
©RxJava
Left obs.
Right obs.
F(O1, O2)

Dealing with time

Dealing with the time
RxJava has a set of methods to deal with time
Let us go back on our cold / hot Observables
It is easy to imagine a hot Observable that emits an onNext
event each second, thus playing the role of a clock

Sampling
With hot observables, RxJava can synchronize on a clock
sample(long period, TimeUnit timeUnit) { }
©RxJava

Sampling
Or synchronize on a reference Observable!
sample(Observable<U> sampler) { } // samples on emission & completion
©RxJava
O1
sampler

RxJava and synchronization
A very powerful feature:
RxJava can synchronize on a clock (real-time)
And on another Observable, it can then take its own time scale

Measuring time
If we can measure time, then we can emit it
Will emit an the amount of time between the two last onNext
events
timeInterval() { } // Observable<TimeInterval<T>>

Measuring time
If we can measure time, then we can delay an emission
Will reemit the items, with a delay
This delay can be computed from the items themselves
delay(long delay, TimeUnit timeUnit) ; // Observable<T>
delay(Func1<T, Observable<U> func1) ; // Observable<T>

Measuring time
If we can measure time, then we can timeout
Will emit an error if no item is seen during this time
timeout(long n, TimeUnit timeUnit) { }

Backpressure

Fast hot observables
What happens if a hot Observable emits too many items?
What happens when an Observer cannot keep up the pace of
an Observable?

Fast hot observables
What happens if a hot Observable emits too many items?
What happens when an Observer cannot keep up the pace of
an Observable?
This leads to the notion of backpressure

Backpressure methods
Backpressure is a way to slow down the emission of elements
It can act on the observing side
Several strategies:
- Buffering items
- Skipping items (sampling is a way to implement this)

Use cases can be very different!
1) The temperature sensors example
- Suppose our application handle a group of temperature
sensors
- Each sensor sends one measure every seconds

1) The temperature sensors example
- Suppose our application handle a group of temperature
sensors
- Each sensor sends one measure every seconds
Maybe we can just keep one measure every hour!
Loosing items ≠ loosing information!

2) The YouTube example
- YouTube sends a HD video to your browser
- But my bandwidth is too low

Instead of sending HD, YouTube will send me 720p

Backpressure
Here we are looking at the backpressure from the observer
point of view
But it can also clog a network
Acting at the Observable level can be important too!

Buffering
Buffering records data in a buffer and emits it
buffer(int size) { } // Observable<List<T>>
buffer(long timeSpan, TimeUnit unit) { } // Observable<List<T>>
buffer(long timeSpan, TimeUnit unit, int maxSize) { }

Buffering
Buffering records data in a buffer (a list) and emits it
buffer(int size) { } // Observable<List<T>>
buffer(long timeSpan, TimeUnit unit) { } // Observable<List<T>>
buffer(long timeSpan, TimeUnit unit, int maxSize) { }
buffer(Observable<O> bufferOpenings, // Openings events
Func1<O, Observable<C>> bufferClosings) { } // Closings events

Windowing
Windowing acts as a buffer, but emits Observables instead of
lists of buffered items
window(int size) { } // Observable<Observable<T>>
window(long timeSpan, TimeUnit unit) { } // Observable<Observable<T>>
window(long timeSpan, TimeUnit unit, int maxSize) { }
window(Observable<O> bufferOpenings, // Openings events
Func1<O, Observable<C>> bufferClosings) { } // Closings events

Buffering & windowing
©RxJava
©RxJava

Throttling
Throttling is a sampler on the beginning or the end of a window
throttleFirst(long windowDuration, TimeUnit unit) { } // Observable<T>
throttleLast(long windowDuration, TimeUnit unit) { }
throttleWithTimeout(long windowDuration, TimeUnit unit) { }

Throttling
throttleFirst: emits the first item in a window of time
throttleLast: emits the last item from that window, on the next
tick of the clock

Debouncing
Debounce also limits the rate of the emission of the items, by
adding a delay before reemitting
debounce(long delay, TimeUnit timeUnit) ; // Observable<T>
debounce(Func1<T, Observable<U> func1) ; // Observable<T>

Wrap-up on RxJava
Complex API, many different concepts, many methods
Allows to process data in chosen threads, this is useful for IO,
computations, specialized threads (GUI threads)
Allows the synchronization of operations
on clocks
on application events
Works in pull mode, and also in push mode
backpressure

Getting the best of
both worlds?

Getting the best of both API?
What about connecting a J8 Stream to a Rx Observable?

If we have an Iterator, this is easy:
Iterator<T> iterator = ... ;
Observable<T> observable = Observable.from(() -> iterator) ;

If we have an Iterator, this is easy:
If we have a Spliterator, not much more complex:
Iterator<T> iterator = ... ;
Observable<T> observable = Observable.from(() -> iterator) ;
Spliterator<T> spliterator = ... ;
Observable<T> observable =
Observable.from(() -> Spliterators.iterator(spliterator)) ;

So if we have a Stream, we can easily build an Observable

What about the other way?

1) We can build an Iterator on an Observable
2) Then build a Spliterator on an Iterator

1) We can build an Iterator on an Observable
2) Then build a Spliterator on an Iterator
But we need to do that ourselves…

Iterating on an Observable
To implement an Iterator:
1) We need to implement next() and hasNext()
2) remove() is a default method in Java 8

The trick is that
an iterator pulls the date from a source
an observable pushes the data to callbacks

The trick is that
an iterator pulls the date from a source
an observable pushes the data to callbacks
So we need an adapter…

How has it been done in the JDK?
public static<T> Iterator<T>
iterator(Spliterator<? extends T> spliterator) {
class Adapter implements Iterator<T>, Consumer<T> {
// implementation
}
return new Adapter() ;
}

boolean valueReady = false ;
T nextElement;
public void accept(T t) {
valueReady = true ;
nextElement = t ;
}
if (!valueReady)
spliterator.tryAdvance(this) ; // calls accept()
return valueReady ;
}
public T next() {
if (!valueReady && !hasNext())
throw new NoSuchElementException() ;
else {
valueReady = false ;
return nextElement ;
}
}
}

Let us adapt this pattern!
public static<T> Iterator<T>
of(Observable<? extends T> observable) {
// implementation
}
return new Adapter() ;
}

T nextElement;
public void accept(T t) { // needs to get the data from the Observable
valueReady = true ;
nextElement = t ;
}
return valueReady ;
}
public T next() {
if (!valueReady && !hasNext())
throw new NoSuchElementException() ;
else {
valueReady = false ;
return nextElement ;
}
}
}

The accept method
T nextElement;
observable.subscribe(
element -> nextElement = element, // onNext
exception -> valueReady = false, // onError
() -> valueReady = false // onComplete
) ;
}
}

The accept method
T nextElement;
element -> nextElement = element, // onNext
exception -> valueReady = false, // onError
() -> valueReady = false // onComplete
) ;
}
}
final…

We can wrap those value in Atomic variable
AtomicBoolean valueReady = new AtomicBoolean(false) ;
AtomicReference<T> nextElement = new AtomicReference() ;
element -> nextElement.set(element), // onNext
exception -> valueReady.set(false), // onError
() -> valueReady.set(false) // onComplete
) ;
}
}

Cant we do better?
interface Wrapper<E> {
E get() ;
}
Wrapper<Boolean> wb = () -> true ;

Cant we do better?
E get() ;
}
Action1<Boolean> onNext = b -> wb.set(b) ; // should return Wrapper<T>

Cant we do better?
E get() ;
public default Wrapper<E> set(E e) {
// should return a wrapper of e
}
}

Cant we do better?
E get() ;
public default Wrapper<E> set(E e) {
return () -> e ;
}
}

We can wrap those value in Atomic variable
Wrapper<Boolean> valueReady = () -> false ;
Wrapper<T> nextElement ;
element -> nextElement.set(element), // onNext
exception -> valueReady.set(false), // onError
() -> valueReady.set(false) // onComplete
) ;
}
}

So we can build an Iterator on an Observable
And with it, a Spliterator on an Observable
it will work on cold observables

Getting the best of both API
The cold observables can be implemeted with Java 8 Streams
The hot observables can be implemented by combining both
API

Comparisons:
Patterns &
Performances

Let us do some comparisons
Let us take one use case
Implemented in Rx and J8 Streams
See the different ways of writing the same processings
And compare the processing times using JMH

Shakespeare plays Scrabble
Published in Java Magazine
Presented in Devoxx 2014
Used during Virtual Technology Summit
https://community.oracle.com/docs/DOC-916777
https://github.com/JosePaumard/jdk8-lambda-tour

Basically, we have the set of the words used by Shakespeare
The official Scrabble player dictionnary
And the question is: how good at Scrabble Shakespeare would
have been?

1) Build the histogram of the letters of a word
2) Number of blanks needed for a letter in a word
3) Number of blanks needed to write a word
4) Predicate to check is a word can be written with 2 blanks
5) Bonus for a doubled letter
6) Final score of a word
7) Histogram of the scores
8) Best word

1) Histogram of the letters
Java 8 Stream API – Rx Java
// Histogram of the letters in a given word
Function<String, Map<Integer, Long>> histoOfLetters =
word -> word.chars().boxed()
.collect(
)
) ;

// Histogram of the letters in a given word
Func1<String, Observable<HashMap<Integer, LongWrapper>>> histoOfLetters =
word -> toIntegerObservable.call(word)
.collect(
() -> new HashMap<Integer, LongWrapper>(),
(HashMap<Integer, LongWrapper> map, Integer value) -> {
LongWrapper newValue = map.get(value) ;
if (newValue == null) {
newValue = () -> 0L ;
}
map.put(value, newValue.incAndSet()) ;
}) ;

interface LongWrapper {
long get() ;
public default LongWrapper set(long l) {
return () -> l ;
}
public default LongWrapper incAndSet() {
return () -> get() + 1L ;
}
public default LongWrapper add(LongWrapper other) {
return () -> get() + other.get() ;
}
}

2) # of blanks for a letter
// number of blanks for a given letter
ToLongFunction<Map.Entry<Integer, Long>> blank =
entry ->
Long.max(
0L,
entry.getValue() -
scrabbleAvailableLetters[entry.getKey() - 'a']
) ;

2) # of blanks for a letter
// number of blanks for a given letter
Func1<Entry<Integer, LongWrapper>, Observable<Long>> blank =
entry ->
Observable.just(
Long.max(
0L,
entry.getValue().get() -
scrabbleAvailableLetters[entry.getKey() - 'a']

3) # of blanks for a word
// number of blanks for a given word
Function<String, Long> nBlanks =
word -> histoOfLetters.apply(word)
.entrySet().stream()
.mapToLong(blank)
.sum();

3) # of blanks for a word
// number of blanks for a given word
Func1<String, Observable<Long>> nBlanks =
word -> histoOfLetters.call(word)
.flatMap(map -> Observable.from(() ->
map.entrySet().iterator()))
.flatMap(blank)
.reduce(Long::sum) ;

4) Predicate for 2 blanks
// can a word be written with 2 blanks?
Predicate<String> checkBlanks = word -> nBlanks.apply(word) <= 2 ;
// can a word be written with 2 blanks?
Func1<String, Observable<Boolean>> checkBlanks =
word -> nBlanks.call(word)
.flatMap(l -> Observable.just(l <= 2L)) ;

5) Bonus for a doubled letter – 1
// Placing the word on the board
// Building the streams of first and last letters
Function<String, IntStream> first3 =
word -> word.chars().limit(3);

// Placing the word on the board
// Building the streams of first and last letters
Func1<String, Observable<Integer>> first3 =
word ->
Observable.from(
IterableSpliterator.of(
word.chars().boxed().limit(3).spliterator()
)
) ;

// Bonus for double letter
ToIntFunction<String> bonusForDoubleLetter =
word -> Stream.of(first3.apply(word), last3.apply(word))
.flatMapToInt(Function.identity())
.map(scoreOfALetter)
.max()
.orElse(0) ;

// Bonus for double letter
Func1<String, Observable<Integer>> bonusForDoubleLetter =
word -> Observable.just(first3.call(word), last3.call(word))
.flatMap(observable -> observable)
.flatMap(scoreOfALetter)
.reduce(Integer::max) ;

// score of the word put on the board
Function<String, Integer> score3 =
word ->
2*(score2.apply(word)
+ bonusForDoubleLetter.applyAsInt(word))
+ (word.length() == 7 ? 50 : 0);

// score of the word put on the board
Func1<String, Observable<Integer>> score3 =
word ->
Observable.just(
score2.call(word), score2.call(word),
bonusForDoubleLetter.call(word),
bonusForDoubleLetter.call(word),
Observable.just(word.length() == 7 ? 50 : 0)
)
.flatMap(observable -> observable)
.reduce(Integer::sum) ;

Function<Function<String, Integer>, Map<Integer, List<String>>>
buildHistoOnScore =
score -> shakespeareWords.stream().parallel()
.filter(scrabbleWords::contains)
.filter(checkBlanks)
.collect(
score,
() -> new TreeMap<>(Comparator.reverseOrder()),
Collectors.toList()
)
) ;

Func1<Func1<String, Observable<Integer>>, Observable<TreeMap<Integer, List<String>>>>
buildHistoOnScore =
score -> Observable.from(() -> shakespeareWords.iterator())
.filter(scrabbleWords::contains)
.filter(word -> checkBlanks.call(word).toBlocking().first())
.collect(
() -> new TreeMap<Integer,
List<String>>(Comparator.reverseOrder()),
(TreeMap<Integer, List<String>> map, String word) -> {
Integer key = score.call(word).toBlocking().first() ;
List<String> list = map.get(key) ;
if (list == null) {
list = new ArrayList<String>() ;
map.put(key, list) ;
}
list.add(word) ;
}) ;

8) Best word
// best key / value pairs
List<Entry<Integer, List<String>>> finalList =
buildHistoOnScore.apply(score3).entrySet()
.stream()
.limit(3)
.collect(Collectors.toList()) ;

8) Best word
List<Entry<Integer, List<String>>> finalList2 =
buildHistoOnScore.call(score3)
.take(3)
.collect(
() -> new ArrayList<Entry<Integer, List<String>>>(),
(list, entry) -> { list.add(entry) ; }
)
.toBlocking()
.first() ;

8) Best word
CountDownLatch latch = new CountDownLatch(3) ;
List<Entry<Integer, List<String>>> finalList2 =
buildHistoOnScore.call(score3)
.take(3)
.collect(
() -> new ArrayList<Entry<Integer, List<String>>>(),
(list, entry) -> { list.add(entry) ; latch.countDown() ; }
)
.forEach(...) ;
latch.await() ;

Patterns comparison
Java 8 Stream API: clean, simple, factory methods for
Collectors
RxJava: flatMap calls, lack of factory methods
Java 8 can easily go parallel, which is a plus

Performances
Let us use JMH
Standard tool for measuring code performance on the JVM
Developed as an Open JDK tool
By Aleksey Shipilev http://shipilev.net/
https://twitter.com/shipilev
http://openjdk.java.net/projects/code-tools/jmh/
http://openjdk.java.net/projects/code-tools/jcstress/
https://www.parleys.com/tutorial/java-microbenchmark-harness-the-lesser-two-evils

JMH
Easy to setup
<dependency>
<groupId>org.openjdk.jmh</groupId>
<artifactId>jmh-core</artifactId>
<version>1.11.1</version>
</dependency>

JMH
Easy to use
@Benchmark
@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.MILLISECONDS)
@Warmup(iterations=5)
@Measurement(iterations=5)
@Fork(3)
public List<Entry<Integer, List<String>>> measureAverage() {
// implementation to test
}

JMH
Launching a benchmark
> mvn clean install
> java –jar target/benchmark.jar

JMH
3 ways of measuring performances
average execution time
number of executions per second
quantiles diagrams

Performances
Average execution time
Benchmark Mode Cnt Score Error Units
NonParallelStreams avgt 100 29,027 ± 0,279 ms/op

Performances
RxJava avgt 100 253,788 ± 1,421 ms/op

Performances
RxJava avgt 100 253,788 ± 1,421 ms/op
ParallelStreams avgt 100 7,624 ± 0,055 ms/op

Performances
RxJava spends a lot of time openning observables, due to the
all flatMap patterns
RxJava avgt 100 253,788 ± 1,421 ms/op
ParallelStreams avgt 100 7,624 ± 0,055 ms/op

Performances
The bench is on Github:
https://github.com/JosePaumard/jdk8-stream-rx-comparison

Java 8 Stream API
Back to the definitions:
2) A Stream does not modify its data

Java 8 Stream API
Back to the definitions:
2) A Stream does not modify its data
How does a Stream work?
1) It connects to a source of data: one source = one stream
2) It consumes the data from the source: « pull mode »

Java 8 Stream API
What about:
- Connecting several streams to a single source?

Java 8 Stream API
What about:
- Connecting several sources to a single stream?

Java 8 Stream API
What about:
- Having a source that produces data whether or not a stream
is connected to it

Java 8 Stream API
What about:
- Having a source that produces data whether or not a stream
is connected to it
Clearly, the Stream API has not been made to handle this

Reactive Stream API
This leads to the « reactive stream » API
3rd party API: Rx Java (and several other languages)
Implementations available as a preview of JDK 9
Everything takes place in java.util.concurrent.Flow
Available in the current JDK 9

Push mode stream
Let us write a model for the source of data
public interface Publisher<T> {
public ... subscribe(Subscriber<T> subscriber);
}

Push mode stream
As a subscriber I will want to unsubscribe
So I need an object from the publisher
on which I can call cancel()
public ... subscribe(Subscriber<T> subscriber);
}

Push mode stream
The first idea that could come to mind is to return a
Subscription object
public Subscription subscribe(Subscriber<T> subscriber);
}

Push mode stream
But it will be a callback, to stay in an asynchronous world
public void subscribe(Subscriber<T> subscriber);
}

Push mode stream
Callback in the subscriber to get a subscription
public interface Subscriber<T> {
public void onSubscribe(Subscription subscription);
}

Push mode stream
}
public void cancel();
}

Push mode stream
The publisher might look like this
public class SimplePublisher<T> implements Publisher<T> {
private Set<Subscriber<T>> subs = ConcurrentHashMap.newKeySet();
public void subscribe(Subscriber<T> subscriber) {
if (subs.add(subscriber)) {
Subscription subscription = new SimpleSubscription();
subscriber.onSubscribe(subscription);
}
}
}

Push mode stream
In the subscribing code
public class SimpleSubscriber<T> implements Subscriber<T> {
private Subscription subscription;
@Override
public void onSubscribe(Subscription subscription) {
this.subscription = subscription;
}
}

Push mode stream
In the running code
Publisher<String> publisher = ...;
Subscriber<String> subscriber = ...;
publisher.subscribe(subscriber);
// some more code
subscriber.getSubscription().cancel();

Push mode stream
I also need callbacks to get the data itself
}

Push mode stream
public void onNext(T item);
}

Push mode stream
public void onComplete();
}

Push mode stream
public void onComplete();
public void onError(Throwable throwable);
}

Push mode stream
Having a source that produces data independantly from its
consumers implies to work in an asynchronous mode
The API is built on callbacks

Several streams per source
In « pull mode », it would not work, or would require the
streams to be synchronized
map Filter 1 Filter 2 Average
map Filter 1 Histogram
Data

Several streams per source
In « pull mode », it would not work, or would require the
streams to be synchronized
In « push mode », it does not raise any problem
map Filter 1 Histogram
Data

Several sources for a stream
In « pull mode », it requires a special Spliterator
Data 1
Data 2

Several sources for a stream
In « pull mode », it requires a special Spliterator
In « push mode », since both sources are not synchronized,
we may have problems
Data 1
Data 2

Push mode with several sources
At some point in our data processing pipeline we want to see
both sources as one, ie merged in some way

How can we merge them if one source is faster than the other?

How can we merge them if one source is faster than the other?
We have all the strategies defined in RxJava

Merging sources in push mode
1) Decide to follow one of the streams, the first one
2) Combine the two last seen items,
- everytime a new item is generated
- or synchronized on a clock, or another source

Merging sources in push mode
1) Decide to follow one of the streams, the first one
2) Combine the two last seen items,
- everytime a new item is generated
- or synchronized on a clock, or another source
Sampler, Debouncer, Throttler, …

The question of backpressure
What will happen if a source is « too fast »?
That is, a consumer cannot process data fast enough
It leads to the same question of « backpressure »

Backpressure
Several strategies:
1) Create a buffer
2) Synchronize on a clock, or a gate, that could be generated
by the slow observer and sample, or windows, or
debounce, or…
3) Try to slow down the source (can be done if I have the
hand on both the producer and the consumer)

Backpressure
There is code for that in the Subscription object
The request() method is there to give information to the
producer
public void cancel();
public void request(long n);
}

Backpressure
There is code for that in the Subscription object
The request() method is there to give information to the
producer
public void onNext(String element) {
// process the element
this.subscription.request(1L);
}

Backpressure
Several strategies:
1) Create a buffer
2) Synchronize on a clock, or a gate, that could be generated
by the slow observer and sample, or windows, or
debounce, or…
3) Try to slow down the source (can be done if I have the
hand on both the producer and the consumer)
4) Have several observers in parallel and then merge the
results

Reactive Streams
New concept (at least in the JDK)
New complexity, several use cases are possible
Still under work (in the JDK and in 3rd party)

Reactive Streams links
Some references on the reactive streams:
http://www.reactive-streams.org/
http://reactivex.io/
https://github.com/reactive-streams/
http://openjdk.java.net/jeps/266
http://gee.cs.oswego.edu/dl/jsr166/dist/docs/index.html

Reactive Streams links
In the classes currently available in the JSR 166 package:
- The class Flow has the Publisher, Subscriber and
Subscription interfaces, and the Processor interface
- The class SubmissionPublisher, that implements Publisher,
meant to be overriden or used as a component of a complete
implementation
Just made it to the OpenJDK 9

Conclusion
Functional programming is in the mood
From the pure performance point of view, things are not that
simple
Java 8 Streams have adopted a partial functional approach,
which is probably a good trade off

Conclusion
Java 8 Stream API: great API to process data in a « pull »
mode
The introduction of a « push » mode allows for many
improvements (synchronization, backpressure)
The backpressure question is relevant

Conclusion
RxJava: rich and complex API
many patterns are available in Java 8 Streams
can run in Java 7 applications
the « push » approach is very interesting
choose your use case carefully, to avoir performance hits
http://www.reactive-streams.org/
http://reactivex.io/
https://github.com/reactive-streams/

Conclusion
Java 8 Streams: part of the JDK
good performances, efficient memory footprint
parallelization
extensible through the Spliterator patterns

Conclusion
Streams & Reactive Streams are very active topics
Java Stream has been released with Java 8, improvements
will be added in Java 9 and beyond
Reactive Streams has several 3rd party implementations
(RxJava) in several languages
Will be part of Java 9

Java 8 Streams and Rx Java Comparison

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Java 8 Streams and Rx Java Comparison (20)

More from José Paumard (17)

Recently uploaded (20)

Java 8 Streams and Rx Java Comparison