Java Stream: Part 2, Is a Count Always a Count?

DZone 's Guide to

Java Stream: Part 2, Is a Count Always a Count?

Since Java 9, the Stream::count implementation has been improved. Let's compare.

· Java Zone ·
Free Resource

In my previous article on the subject, we learned that JDK 8’s  stream()::count takes a longer time to execute the more elements there are in the Stream.

For more recent JDKs, such as Java 11, that is no longer the case for simple stream pipelines. Learn how things have gotten improved within the JDK itself.

Java 8

In my previous article, we could conclude that the operation list.stream().count() is O(N) under Java 8, i.e. the execution time depends on the number of elements in the original list. Read the article here.

Java 9 and Upwards

As rightfully pointed out by Nikolai Parlog (@nipafx) and Brian Goetz (@BrianGoetz) on Twitter, the implementation of  Stream::count was improved beginning from Java 9. Here is a comparison of the underlying  Stream::count code between Java 8 and later Java versions:

Java 8 (From the ReferencePipeline Class)

return mapToLong(e -> 1L).sum();                             

Java 9 and Later (From the ReduceOps Class)

if (StreamOpFlag.SIZED.isKnown(flags)) {
    return spliterator.getExactSizeIfKnown();

It appears Stream::count in Java 9 and later is O(1) for Spliterators of known size rather than being O(N). Let’s verify that hypothesis.


The big-O property can be observed by running the following JMH benchmarks under Java 8 and Java 11: 

public class CountBenchmark {

    private List<Integer> list;

    @Param({"1", "1000", "1000000"})
    private int size;

    public void setup() {
        list = IntStream.range(0, size)

    public long listSize() {
        return list.size();

    public long listStreamCount() {
        return list.stream().count();

    public static void main(String[] args) throws RunnerException {
        Options opt = new OptionsBuilder()

        new Runner(opt).run();



This will produce the following outputs on my laptop (MacBook Pro mid-2015, 2.2 GHz Intel Core i7):

JDK 8 (From my Previous Article)

Benchmark                        (size)   Mode  Cnt          Score           Error  Units
CountBenchmark.listSize               1  thrpt    5  966658591.905 ± 175787129.100  ops/s
CountBenchmark.listSize            1000  thrpt    5  862173760.015 ± 293958267.033  ops/s
CountBenchmark.listSize         1000000  thrpt    5  879607621.737 ± 107212069.065  ops/s
CountBenchmark.listStreamCount        1  thrpt    5   39570790.720 ±   3590270.059  ops/s
CountBenchmark.listStreamCount     1000  thrpt    5   30383397.354 ±  10194137.917  ops/s
CountBenchmark.listStreamCount  1000000  thrpt    5        398.959 ±       170.737  ops/s

JDK 11

Benchmark                                  (size)   Mode  Cnt          Score           Error  Units
CountBenchmark.listSize                         1  thrpt    5  898916944.365 ± 235047181.830  ops/s
CountBenchmark.listSize                      1000  thrpt    5  865080967.750 ± 203793349.257  ops/s
CountBenchmark.listSize                   1000000  thrpt    5  935820818.641 ±  95756219.869  ops/s
CountBenchmark.listStreamCount                  1  thrpt    5   95660206.302 ±  27337762.894  ops/s
CountBenchmark.listStreamCount               1000  thrpt    5   78899026.467 ±  26299885.209  ops/s
CountBenchmark.listStreamCount            1000000  thrpt    5   83223688.534 ±  16119403.504  ops/s

As can be seen, in Java 11, the  list.stream().count() operation is now  O(1) and not O(N).

Brian Goetz pointed out that some developers who were using Stream::peek method calls under Java 8 discovered that these methods were no longer invoked if the Stream::count terminal operation was run under Java 9 and onwards. This generated some negative feedback to the JDK developers. Personally, I think it was the right decision by the JDK developers and that this, instead, presented a great opportunity for Stream::peek users to get their code right.

More Complex Stream Pipelines

In this chapter, we will take a look at more complex stream pipelines.

JDK 11

Tagir Valeev concluded that pipelines like stream().skip(1).count() are not O(1) for List::stream.

This can be observed by running the following benchmark:

public long listStreamSkipCount() {
    return list.stream().skip(1).count();
CountBenchmark.listStreamCount                  1  thrpt    5  105546649.075 ±  10529832.319  ops/s
CountBenchmark.listStreamCount               1000  thrpt    5   81370237.291 ±  15566491.838  ops/s
CountBenchmark.listStreamCount            1000000  thrpt    5   75929699.395 ±  14784433.428  ops/s
CountBenchmark.listStreamSkipCount              1  thrpt    5   35809816.451 ±  12055461.025  ops/s
CountBenchmark.listStreamSkipCount           1000  thrpt    5    3098848.946 ±    339437.339  ops/s
CountBenchmark.listStreamSkipCount        1000000  thrpt    5       3646.513 ±       254.442  ops/s                             

Thus, list.stream().skip(1).count() is still O(N).


Some stream implementations are actually aware of their sources and can take appropriate shortcuts and merge stream operations into the stream source itself. This can improve performance massively, especially for large streams with more complex stream pipelines like  stream().skip(1).count().

The Speedment ORM tool allows databases to be viewed as Stream objects, and these streams can optimize away many stream operations like the Stream::countStream::skip, and Stream::limit  operation, as demonstrated in the benchmark below. I have used the open-source Sakila exemplary database as data input. The Sakila database is all about rental films, artists, etc.

public long rentalsSkipCount() {
    return rentals.stream().skip(1).count();

public long filmsSkipCount() {
    return films.stream().skip(1).count();

When run, the following output will be produced:

SpeedmentCountBenchmark.filmsSkipCount        N/A  thrpt    5   68052838.621 ±    739171.008  ops/s
SpeedmentCountBenchmark.rentalsSkipCount      N/A  thrpt    5   68224985.736 ±   2683811.510  ops/s

The “rental” table contains over 10,000 rows whereas the “film” table only contains 1,000 rows. Nevertheless, their  stream().skip(1).count() operations complete in almost the same time. Even if a table would contain a trillion rows, it would still count the elements in the same elapsed time. Thus, the stream().skip(1).count()  implementation has a complexity that is O(1) and not O(N).

Note: The benchmark above were run with "DataStore" in-JVM-memory acceleration. If run with no acceleration directly against a database, the response time would depend on the underlying database’s ability to execute a nested “ SELECT count(*) …” statement.



was significantly improved in Java 9.

There are stream implementations, such as Speedment, that are able to compute Stream::count in O(1) time even for more complex stream pipelines like  stream().skip(...).count() or even  stream.filter(...).skip(...).count().


Speedment Stream ORM Initializer: https://www.speedment.com/initializer/

Sakila: https://dev.mysql.com/doc/index-other.html or https://hub.docker.com/r/restsql/mysql-sakila


Published at DZone with permission of Per-Åke Minborg , DZone MVB. See the original article here.

Opinions expressed by DZone contributors are their own.

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

{{ parent.tldr }}

{{ parent.urlSource.name }}