Java parallelStream not showing correct result [duplicate]

Question

I was doing some performance evaluation of Java aggeragate operations to iterate over collections. I was doing an evaluation of performance stream and parallelStream. But I found that the output of parallelStream is wrong most of the times. For example in the following code i got wrong output from parallelStream more than 80% of the time:

    public class ParallelStreamPerformance {
    static int totaleven = 0;
    public static void main(String[] args) {
        List<Integer> randomList = new ArrayList<>();
        Random rnd = new Random();
        for(int i = 0 ;i < 1000;i++) {
            int r = rnd.nextInt(500000);
            randomList.add(r);
        }

        long s1 = System.currentTimeMillis();

        randomList.stream().filter(e -> e%2 ==0).forEach(e -> count());
        System.out.println("Even: "+totaleven);
        long e1 = System.currentTimeMillis();
        System.out.println(e1 - s1);

        totaleven = 0;
        long s2 = System.currentTimeMillis();

        randomList.parallelStream().filter(e -> e%2 ==0).forEach(e -> count());
        System.out.println("Even: "+totaleven);
        long e2 = System.currentTimeMillis();
        System.out.println(e2 - s2);
    }
    public static void count() {
        totaleven++;
    }
}

My question is: Am I using parallelStream in a wrong way? Is there any way to ensure the correctness of parallelStream. Thanks

Your benchmarking methodology is worse than worthless; it will give you answers, but they'll be wrong. See stackoverflow.com/questions/504103/… — Brian Goetz
– Brian Goetz, Commented Dec 9, 2017 at 15:04
When your actual question is about getting a wrong result, you shouldn’t clutter it with irrelevant benchmark stuff. Your code could have be shortened to ten lines and the prose about you doing a “performance evaluation” is obsolete as well. When your code doesn’t work correctly, you don’t need to measure its performance. — Holger
– Holger, Commented Dec 11, 2017 at 7:45

sayboras · Accepted Answer · 2017-12-08 17:00:19Z

4

I think your code having issue with count() method. As parallelStream will try to perform task concurrently. This method should be synchronized or you can make totaleven as Atomtic Integer. Hope it helps.

answered Dec 8, 2017 at 17:00

sayboras

5,1732 gold badges25 silver badges42 bronze badges

Sign up to request clarification or add additional context in comments.

6 Comments

Md Johirul Islam Over a year ago

Yes using synchronized method solve the problem. Thanks

Brian Goetz Over a year ago

So now your code is correct, but it is unlikely that the parallel code will be faster than the sequential (which is the only reason to use parallelism), since the threads will spend all their time waiting for a lock. What you want to do is use count() or reduce() rather than incrementing a shared mutable counter. This is easier to get right (no synchronization needed) and more parallel-friendly.

Holger Over a year ago

@BrianGoetz: perhaps it was a design mistake to add forEach to the Stream API…

Brian Goetz Over a year ago

@Holger We actually did discuss that, since the only purpose of any stream operation whose behavior parameters don't return anything is to do side-effects. But realistically, we would have (and should have) been laughed out of the room had we done so; that would have just been hiding our heads in the sand. (And then people would have just coopted some other terminal operation and done their side-effects inside, say, a map() call, and collected into an empty sink to simulate forEach().) For better or worse, side-effects are part of Java. At best we can discourage them.

Holger Over a year ago

@BrianGoetz: perhaps, a not-so-easy-to-find name had helped, e.g. see how much the existence of forEachOrdered gets ignored…

|

Bluurr · Accepted Answer · 2017-12-08 17:20:20Z

3

Instead of using the forEach to increment a counter you can use the terminal operation Stream:count

For example

totaleven = randomList.stream().filter(e -> e % 2 ==0).count();
totaleven = 0;
totaleven = randomList.parallelStream().filter(e -> e % 2 ==0).count();

totaleven would need to be changed to data type long or casting applied.

answered Dec 8, 2017 at 17:20

Bluurr

4675 silver badges6 bronze badges

Comments

Julian Rubin · Accepted Answer · 2017-12-08 17:01:38Z

2

What's wrong with parallel stream result? If it is too small then most likely you have problem with totaleven++ as it is not thread safe. Use AtomicInteger or any other thread safe solution.

answered Dec 8, 2017 at 17:01

Julian Rubin

1,2551 gold badge13 silver badges23 bronze badges

Collectives™ on Stack Overflow

Java parallelStream not showing correct result [duplicate]

3 Answers 3

6 Comments

Comments

Comments

Linked

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

6 Comments

Comments

Comments

Linked

Related