int sum = widgets.stream()
.filter(b -> b.getColor() == RED)
.mapToInt(b -> b.getWeight())
.sum();
Here we use widgets
, a Collection<Widget>
,
as a source for a stream, and then perform a filter-map-reduce on the stream
to obtain the sum of the weights of the red widgets. (Summation is an
example of a reduction
operation.)
The key abstraction introduced in this package is stream. The
classes Stream
, IntStream
,
LongStream
, and DoubleStream
are streams over objects and the primitive int
, long
and
double
types. Streams differ from collections in several ways:
- No storage. A stream is not a data structure that stores elements; instead, it conveys elements from a source such as a data structure, an array, a generator function, or an I/O channel, through a pipeline of computational operations.
- Functional in nature. An operation on a stream produces a result,
but does not modify its source. For example, filtering a
Stream
obtained from a collection produces a newStream
without the filtered elements, rather than removing elements from the source collection. - Laziness-seeking. Many stream operations, such as filtering, mapping,
or duplicate removal, can be implemented lazily, exposing opportunities
for optimization. For example, "find the first
String
with three consecutive vowels" need not examine all the input strings. Stream operations are divided into intermediate (Stream
-producing) operations and terminal (value- or side-effect-producing) operations. Intermediate operations are always lazy. - Possibly unbounded. While collections have a finite size, streams
need not. Short-circuiting operations such as
limit(n)
orfindFirst()
can allow computations on infinite streams to complete in finite time. - Consumable. The elements of a stream are only visited once during
the life of a stream. Like an
Iterator
, a new stream must be generated to revisit the same elements of the source.
- From a
Collection
via thestream()
andparallelStream()
methods; - From an array via
stream(Object[])
; - From static factory methods on the stream classes, such as
of(Object[])
,range(int, int)
oriterate(Object, UnaryOperator)
;
Additional stream sources can be provided by third-party libraries using these techniques.
Stream operations and pipelines
Stream operations are divided into intermediate and
terminal operations, and are combined to form stream
pipelines. A stream pipeline consists of a source (such as a
Collection
, an array, a generator function, or an I/O channel);
followed by zero or more intermediate operations such as
Stream.filter
or Stream.map
; and a terminal operation such
as Stream.forEach
or Stream.reduce
.
Intermediate operations return a new stream. They are always
lazy; executing an intermediate operation such as
filter()
does not actually perform any filtering, but instead
creates a new stream that, when traversed, contains the elements of
the initial stream that match the given predicate. Traversal
of the pipeline source does not begin until the terminal operation of the
pipeline is executed.
Terminal operations, such as Stream.forEach
or
IntStream.sum
, may traverse the stream to produce a result or a
side-effect. After the terminal operation is performed, the stream pipeline
is considered consumed, and can no longer be used; if you need to traverse
the same data source again, you must return to the data source to get a new
stream. In almost all cases, terminal operations are eager,
completing their traversal of the data source and processing of the pipeline
before returning. Only the terminal operations iterator()
and
spliterator()
are not; these are provided as an "escape hatch" to enable
arbitrary client-controlled pipeline traversals in the event that the
existing operations are not sufficient to the task.
Processing streams lazily allows for significant efficiencies; in a pipeline such as the filter-map-sum example above, filtering, mapping, and summing can be fused into a single pass on the data, with minimal intermediate state. Laziness also allows avoiding examining all the data when it is not necessary; for operations such as "find the first string longer than 1000 characters", it is only necessary to examine just enough strings to find one that has the desired characteristics without examining all of the strings available from the source. (This behavior becomes even more important when the input stream is infinite and not merely large.)
Intermediate operations are further divided into stateless
and stateful operations. Stateless operations, such as filter
and map
, retain no state from previously seen element when processing
a new element -- each element can be processed
independently of operations on other elements. Stateful operations, such as
distinct
and sorted
, may incorporate state from previously
seen elements when processing new elements.
Stateful operations may need to process the entire input before producing a result. For example, one cannot produce any results from sorting a stream until one has seen all elements of the stream. As a result, under parallel computation, some pipelines containing stateful intermediate operations may require multiple passes on the data or may need to buffer significant data. Pipelines containing exclusively stateless intermediate operations can be processed in a single pass, whether sequential or parallel, with minimal data buffering.
Further, some operations are deemed short-circuiting operations. An intermediate operation is short-circuiting if, when presented with infinite input, it may produce a finite stream as a result. A terminal operation is short-circuiting if, when presented with infinite input, it may terminate in finite time. Having a short-circuiting operation in the pipeline is a necessary, but not sufficient, condition for the processing of an infinite stream to terminate normally in finite time.
Parallelism
Processing elements with an explicit for-
loop is inherently serial.
Streams facilitate parallel execution by reframing the computation as a pipeline of
aggregate operations, rather than as imperative operations on each individual
element. All streams operations can execute either in serial or in parallel.
The stream implementations in the JDK create serial streams unless parallelism is
explicitly requested. For example, Collection
has methods
stream()
and parallelStream()
,
which produce sequential and parallel streams respectively; other
stream-bearing methods such as range(int, int)
produce sequential streams but these streams can be efficiently parallelized by
invoking their parallel()
method.
To execute the prior "sum of weights of widgets" query in parallel, we would
do:
int sumOfWeights = widgets.
parallelStream()
.filter(b -> b.getColor() == RED) .mapToInt(b -> b.getWeight()) .sum();
The only difference between the serial and parallel versions of this
example is the creation of the initial stream, using "parallelStream()
"
instead of "stream()
". When the terminal operation is initiated,
the stream pipeline is executed sequentially or in parallel depending on the
orientation of the stream on which it is invoked. Whether a stream will execute in serial or
parallel can be determined with the isParallel()
method, and the
orientation of a stream can be modified with the
sequential()
and
parallel()
operations. When the terminal
operation is initiated, the stream pipeline is executed sequentially or in
parallel depending on the mode of the stream on which it is invoked.
Except for operations identified as explicitly nondeterministic, such
as findAny()
, whether a stream executes sequentially or in parallel
should not change the result of the computation.
Most stream operations accept parameters that describe user-specified
behavior, which are often lambda expressions. To preserve correct behavior,
these behavioral parameters must be non-interfering, and in
most cases must be stateless. Such parameters are always instances
of a functional interface such
as Function
, and are often lambda expressions or
method references.
Non-interference
Streams enable you to execute possibly-parallel aggregate operations over a variety of data sources, including even non-thread-safe collections such asArrayList
. This is possible only if we can prevent
interference with the data source during the execution of a stream
pipeline. Except for the escape-hatch operations iterator()
and
spliterator()
, execution begins when the terminal operation is
invoked, and ends when the terminal operation completes. For most data
sources, preventing interference means ensuring that the data source is
not modified at all during the execution of the stream pipeline.
The notable exception to this are streams whose sources are concurrent
collections, which are specifically designed to handle concurrent modification.
Concurrent stream sources are those whose Spliterator
reports the
CONCURRENT
characteristic.
Accordingly, behavioral parameters in stream pipelines whose source might not be concurrent should never modify the stream's data source. A behavioral parameter is said to interfere with a non-concurrent data source if it modifies, or causes to be modified, the stream's data source. The need for non-interference applies to all pipelines, not just parallel ones. Unless the stream source is concurrent, modifying a stream's data source during execution of a stream pipeline can cause exceptions, incorrect answers, or nonconformant behavior. For well-behaved stream sources, the source can be modified before the terminal operation commences and those modifications will be reflected in the covered elements. For example, consider the following code:
List<String> l = new ArrayList(Arrays.asList("one", "two"));
Stream<String> sl = l.stream();
l.add("three");
String s = sl.collect(joining(" "));
collect
operation commenced the result will be a string of "one two three". All the
streams returned from JDK collections, and most other JDK classes,
are well-behaved in this manner; for streams generated by other libraries, see
Low-level stream
construction for requirements for building well-behaved streams.
Stateless behaviors
Stream pipeline results may be nondeterministic or incorrect if the behavioral parameters to the stream operations are stateful. A stateful lambda (or other object implementing the appropriate functional interface) is one whose result depends on any state which might change during the execution of the stream pipeline. An example of a stateful lambda is the parameter tomap()
in:
Set<Integer> seen = Collections.synchronizedSet(new HashSet<>());
stream.parallel().map(e -> { if (seen.add(e)) return 0; else return e; })...
Note also that attempting to access mutable state from behavioral parameters presents you with a bad choice with respect to safety and performance; if you do not synchronize access to that state, you have a data race and therefore your code is broken, but if you do synchronize access to that state, you risk having contention undermine the parallelism you are seeking to benefit from. The best approach is to avoid stateful behavioral parameters to stream operations entirely; there is usually a way to restructure the stream pipeline to avoid statefulness.
Side-effects
Side-effects in behavioral parameters to stream operations are, in general, discouraged, as they can often lead to unwitting violations of the statelessness requirement, as well as other thread-safety hazards.If the behavioral parameters do have side-effects, unless explicitly
stated, there are no guarantees as to the
visibility
of those side-effects to other threads, nor are there any guarantees that
different operations on the "same" element within the same stream pipeline
are executed in the same thread. Further, the ordering of those effects
may be surprising. Even when a pipeline is constrained to produce a
result that is consistent with the encounter order of the stream
source (for example, IntStream.range(0,5).parallel().map(x -> x*2).toArray()
must produce [0, 2, 4, 6, 8]
), no guarantees are made as to the order
in which the mapper function is applied to individual elements, or in what
thread any behavioral parameter is executed for a given element.
Many computations where one might be tempted to use side effects can be more
safely and efficiently expressed without side-effects, such as using
reduction instead of mutable
accumulators. However, side-effects such as using println()
for debugging
purposes are usually harmless. A small number of stream operations, such as
forEach()
and peek()
, can operate only via side-effects;
these should be used with care.
As an example of how to transform a stream pipeline that inappropriately uses side-effects to one that does not, the following code searches a stream of strings for those matching a given regular expression, and puts the matches in a list.
ArrayList<String> results = new ArrayList<>();
stream.filter(s -> pattern.matcher(s).matches())
.forEach(s -> results.add(s)); // Unnecessary use of side-effects!
ArrayList
would cause incorrect results, and
adding needed synchronization would cause contention, undermining the
benefit of parallelism. Furthermore, using side-effects here is completely
unnecessary; the forEach()
can simply be replaced with a reduction
operation that is safer, more efficient, and more amenable to
parallelization:
List<String>results =
stream.filter(s -> pattern.matcher(s).matches())
.collect(Collectors.toList()); // No side-effects!
Ordering
Streams may or may not have a defined encounter order. Whether
or not a stream has an encounter order depends on the source and the
intermediate operations. Certain stream sources (such as List
or
arrays) are intrinsically ordered, whereas others (such as HashSet
)
are not. Some intermediate operations, such as sorted()
, may impose
an encounter order on an otherwise unordered stream, and others may render an
ordered stream unordered, such as unordered()
.
Further, some terminal operations may ignore encounter order, such as
forEach()
.
If a stream is ordered, most operations are constrained to operate on the
elements in their encounter order; if the source of a stream is a List
containing [1, 2, 3]
, then the result of executing map(x -> x*2)
must be [2, 4, 6]
. However, if the source has no defined encounter
order, then any permutation of the values [2, 4, 6]
would be a valid
result.
For sequential streams, the presence or absence of an encounter order does not affect performance, only determinism. If a stream is ordered, repeated execution of identical stream pipelines on an identical source will produce an identical result; if it is not ordered, repeated execution might produce different results.
For parallel streams, relaxing the ordering constraint can sometimes enable
more efficient execution. Certain aggregate operations,
such as filtering duplicates (distinct()
) or grouped reductions
(Collectors.groupingBy()
) can be implemented more efficiently if ordering of elements
is not relevant. Similarly, operations that are intrinsically tied to encounter order,
such as limit()
, may require
buffering to ensure proper ordering, undermining the benefit of parallelism.
In cases where the stream has an encounter order, but the user does not
particularly care about that encounter order, explicitly de-ordering
the stream with unordered()
may
improve parallel performance for some stateful or terminal operations.
However, most stream pipelines, such as the "sum of weight of blocks" example
above, still parallelize efficiently even under ordering constraints.
Reduction operations
A reduction operation (also called a fold) takes a sequence of input elements and combines them into a single summary result by repeated application of a combining operation, such as finding the sum or maximum of a set of numbers, or accumulating elements into a list. The streams classes have multiple forms of general reduction operations, calledreduce()
and collect()
,
as well as multiple specialized reduction forms such as
sum()
, max()
,
or count()
.
Of course, such operations can be readily implemented as simple sequential loops, as in:
int sum = 0;
for (int x : numbers) {
sum += x;
}
int sum = numbers.stream().reduce(0, (x,y) -> x+y);
int sum = numbers.stream().reduce(0, Integer::sum);
These reduction operations can run safely in parallel with almost no modification:
int sum = numbers.parallelStream().reduce(0, Integer::sum);
Reduction parallellizes well because the implementation
can operate on subsets of the data in parallel, and then combine the
intermediate results to get the final correct answer. (Even if the language
had a "parallel for-each" construct, the mutative accumulation approach would
still required the developer to provide
thread-safe updates to the shared accumulating variable sum
, and
the required synchronization would then likely eliminate any performance gain from
parallelism.) Using reduce()
instead removes all of the
burden of parallelizing the reduction operation, and the library can provide
an efficient parallel implementation with no additional synchronization
required.
The "widgets" examples shown earlier shows how reduction combines with
other operations to replace for loops with bulk operations. If widgets
is a collection of Widget
objects, which have a getWeight
method,
we can find the heaviest widget with:
OptionalInt heaviest = widgets.parallelStream()
.mapToInt(Widget::getWeight)
.max();
In its more general form, a reduce
operation on elements of type
<T>
yielding a result of type <U>
requires three parameters:
<U> U reduce(U identity,
BiFunction<U, ? super T, U> accumulator,
BinaryOperator<U> combiner);
More formally, the identity
value must be an identity for
the combiner function. This means that for all u
,
combiner.apply(identity, u)
is equal to u
. Additionally, the
combiner
function must be associative and
must be compatible with the accumulator
function: for all u
and t
, combiner.apply(u, accumulator.apply(identity, t))
must
be equals()
to accumulator.apply(u, t)
.
The three-argument form is a generalization of the two-argument form, incorporating a mapping step into the accumulation step. We could re-cast the simple sum-of-weights example using the more general form as follows:
int sumOfWeights = widgets.stream()
.reduce(0,
(sum, b) -> sum + b.getWeight())
Integer::sum);
Mutable reduction
A mutable reduction operation accumulates input elements into a mutable result container, such as aCollection
or StringBuilder
,
as it processes the elements in the stream.
If we wanted to take a stream of strings and concatenate them into a single long string, we could achieve this with ordinary reduction:
String concatenated = strings.reduce("", String::concat)
We would get the desired result, and it would even work in parallel. However,
we might not be happy about the performance! Such an implementation would do
a great deal of string copying, and the run time would be O(n^2) in
the number of characters. A more performant approach would be to accumulate
the results into a StringBuilder
, which is a mutable
container for accumulating strings. We can use the same technique to
parallelize mutable reduction as we do with ordinary reduction.
The mutable reduction operation is called
collect()
,
as it collects together the desired results into a result container such
as a Collection
.
A collect
operation requires three functions:
a supplier function to construct new instances of the result container, an
accumulator function to incorporate an input element into a result
container, and a combining function to merge the contents of one result
container into another. The form of this is very similar to the general
form of ordinary reduction:
<R> R collect(Supplier<R> supplier,
BiConsumer<R, ? super T> accumulator,
BiConsumer<R, R> combiner);
As with reduce()
, a benefit of expressing collect
in this
abstract way is that it is directly amenable to parallelization: we can
accumulate partial results in parallel and then combine them, so long as the
accumulation and combining functions satisfy the appropriate requirements.
For example, to collect the String representations of the elements in a
stream into an ArrayList
, we could write the obvious sequential
for-each form:
ArrayList<String> strings = new ArrayList<>();
for (T element : stream) {
strings.add(element.toString());
}
ArrayList<String> strings = stream.collect(() -> new ArrayList<>(),
(c, e) -> c.add(e.toString()),
(c1, c2) -> c1.addAll(c2));
List<String> strings = stream.map(Object::toString)
.collect(ArrayList::new, ArrayList::add, ArrayList::addAll);
ArrayList constructor
, the accumulator adds the stringified element to an
ArrayList
, and the combiner simply uses addAll
to copy the strings from one container into the other.
The three aspects of collect
-- supplier, accumulator, and
combiner -- are tightly coupled. We can use the abstraction of a
Collector
to capture all three aspects. The
above example for collecting strings into a List
can be rewritten
using a standard Collector
as:
List<String> strings = stream.map(Object::toString)
.collect(Collectors.toList());
Packaging mutable reductions into a Collector has another advantage:
composability. The class Collectors
contains a
number of predefined factories for collectors, including combinators
that transform one collector into another. For example, suppose we have a
collector that computes the sum of the salaries of a stream of
employees, as follows:
Collector<Employee, ?, Integer> summingSalaries
= Collectors.summingInt(Employee::getSalary);
?
for the second type parameter merely indicates that we don't
care about the intermediate representation used by this collector.)
If we wanted to create a collector to tabulate the sum of salaries by
department, we could reuse summingSalaries
using
groupingBy
:
Map<Department, Integer> salariesByDept
= employees.stream().collect(Collectors.groupingBy(Employee::getDepartment,
summingSalaries));
As with the regular reduction operation, collect()
operations can
only be parallelized if appropriate conditions are met. For any partially
accumulated result, combining it with an empty result container must
produce an equivalent result. That is, for a partially accumulated result
p
that is the result of any series of accumulator and combiner
invocations, p
must be equivalent to
combiner.apply(p, supplier.get())
.
Further, however the computation is split, it must produce an equivalent
result. For any input elements t1
and t2
, the results
r1
and r2
in the computation below must be equivalent:
A a1 = supplier.get();
accumulator.accept(a1, t1);
accumulator.accept(a1, t2);
R r1 = finisher.apply(a1); // result without splitting
A a2 = supplier.get();
accumulator.accept(a2, t1);
A a3 = supplier.get();
accumulator.accept(a3, t2);
R r2 = finisher.apply(combiner.apply(a2, a3)); // result with splitting
Here, equivalence generally means according to equals(Object)
.
but in some cases equivalence may be relaxed to account for differences in
order.
Reduction, concurrency, and ordering
With some complex reduction operations, for example acollect()
that
produces a Map
, such as:
Map<Buyer, List<Transaction>> salesByBuyer
= txns.parallelStream()
.collect(Collectors.groupingBy(Transaction::getBuyer));
Map
into another by
key) can be expensive for some Map
implementations.
Suppose, however, that the result container used in this reduction
was a concurrently modifiable collection -- such as a
ConcurrentHashMap
. In that case, the parallel
invocations of the accumulator could actually deposit their results
concurrently into the same shared result container, eliminating the need for
the combiner to merge distinct result containers. This potentially provides
a boost to the parallel execution performance. We call this a
concurrent reduction.
A Collector
that supports concurrent reduction is
marked with the CONCURRENT
characteristic. However, a concurrent collection also has a downside. If
multiple threads are depositing results concurrently into a shared container,
the order in which results are deposited is non-deterministic. Consequently,
a concurrent reduction is only possible if ordering is not important for the
stream being processed. The collect(Collector)
implementation will only perform a concurrent reduction if
- The stream is parallel;
- The collector has the
CONCURRENT
characteristic, and; - Either the stream is unordered, or the collector has the
UNORDERED
characteristic.
unordered()
method. For example:
Map<Buyer, List<Transaction>> salesByBuyer
= txns.parallelStream()
.unordered()
.collect(groupingByConcurrent(Transaction::getBuyer));
groupingByConcurrent(Function super T, ? extends K>)
is the
concurrent equivalent of groupingBy
).
Note that if it is important that the elements for a given key appear in the order they appear in the source, then we cannot use a concurrent reduction, as ordering is one of the casualties of concurrent insertion. We would then be constrained to implement either a sequential reduction or a merge-based parallel reduction.
Associativity
An operator or functionop
is associative if the following
holds:
(a op b) op c == a op (b op c)
a op b op c op d == (a op b) op (c op d)
(a op b)
in parallel with (c op d)
, and
then invoke op
on the results.
Examples of associative operations include numeric addition, min, and max, and string concatenation.
Low-level stream construction
So far, all the stream examples have used methods likestream()
or stream(Object[])
to obtain a stream. How are those stream-bearing methods implemented?
The class StreamSupport
has a number of
low-level methods for creating a stream, all using some form of a
Spliterator
. A spliterator is the parallel analogue of an
Iterator
; it describes a (possibly infinite) collection of
elements, with support for sequentially advancing, bulk traversal, and
splitting off some portion of the input into another spliterator which can
be processed in parallel. At the lowest level, all streams are driven by a
spliterator.
There are a number of implementation choices in implementing a
spliterator, nearly all of which are tradeoffs between simplicity of
implementation and runtime performance of streams using that spliterator.
The simplest, but least performant, way to create a spliterator is to
create one from an iterator using
spliteratorUnknownSize(java.util.Iterator, int)
.
While such a spliterator will work, it will likely offer poor parallel
performance, since we have lost sizing information (how big is the
underlying data set), as well as being constrained to a simplistic
splitting algorithm.
A higher-quality spliterator will provide balanced and known-size
splits, accurate sizing information, and a number of other
characteristics
of the
spliterator or data that can be used by implementations to optimize
execution.
Spliterators for mutable data sources have an additional challenge;
timing of binding to the data, since the data could change between the time
the spliterator is created and the time the stream pipeline is executed.
Ideally, a spliterator for a stream would report a characteristic of
IMMUTABLE
or CONCURRENT
; if not it should be
late-binding. If a source
cannot directly supply a recommended spliterator, it may indirectly supply
a spliterator using a Supplier
, and construct a stream via the
Supplier
-accepting versions of
stream()
.
The spliterator is obtained from the supplier only after the terminal
operation of the stream pipeline commences.
These requirements significantly reduce the scope of potential interference between mutations of the stream source and execution of stream pipelines. Streams based on spliterators with the desired characteristics, or those using the Supplier-based factory forms, are immune to modifications of the data source prior to commencement of the terminal operation (provided the behavioral parameters to the stream operations meet the required criteria for non-interference and statelessness). See Non-Interference for more details.
Interfaces
BaseStream<T, S extends BaseStream<T, S>> | Base interface for streams, which are sequences of elements supporting sequential and parallel aggregate operations. |
Collector<T, A, R> | A mutable reduction operation that accumulates input elements into a mutable result container, optionally transforming the accumulated result into a final representation after all input elements have been processed. |
DoubleStream | A sequence of primitive double-valued elements supporting sequential and parallel aggregate operations. |
DoubleStream.Builder | A mutable builder for a DoubleStream . |
IntStream | A sequence of primitive int-valued elements supporting sequential and parallel aggregate operations. |
IntStream.Builder | A mutable builder for an IntStream . |
LongStream | A sequence of primitive long-valued elements supporting sequential and parallel aggregate operations. |
LongStream.Builder | A mutable builder for a LongStream . |
Stream<T> | A sequence of elements supporting sequential and parallel aggregate operations. |
Stream.Builder<T> | A mutable builder for a Stream . |
Classes
Collectors | Implementations of Collector that implement various useful reduction
operations, such as accumulating elements into collections, summarizing
elements according to various criteria, etc. |
StreamSupport | Low-level utility methods for creating and manipulating streams. |
Enums
Collector.Characteristics | Characteristics indicating properties of a Collector , which can
be used to optimize reduction implementations. |