Stream.sorted(...).forEach(...) issue (parallel streams)

[Paul Sandoz]

Hi Georgiy,

This is not a bug.

Stream.forEach when executed for a parallel stream will report elements in a temporal order not in encounter order of the upstream (if any).

Temporal order is when elements are reported concurrently by one or more evaluating threads. Basically a thread will report an element as and when it is ready to do so without regard to any other threads also reporting elements.

Thus you are adding concurrently to a non-concurrent mutable data structure.

If there is one thread (essentially sequential evaluation) then temporal order matches upstream encounter order (if any).

You need to do this:

  Arrays.parallelStream(baseData).sorted(cmp).sequential().forEach(sorted::add);

or do a collect:

  Arrays.parallelStream(baseData).sorted(cmp).collect(Collectors.toList());

to preserve order in the list.

Paul.

On Jan 30, 2013, at 2:45 PM, Georgiy Rakov <georgiy.rakov at oracle.com> wrote:

> Hello,
> 
> it seems there is a bug in promoted b74 while chaining sorted and forEach being applied to parallel streams.
> 
> Let's consider following code:
> 
>   public class SortedIssue2 {
>        public static void main(String[]args) {
>            Integer[] baseData = new Integer[]{1, 4, 3, 2, 9, 7};
>            Comparator<Integer> cmp = (a, b) -> a - b;
> 
>            List<Integer> sorted = new ArrayList<>();
> 
>              Arrays.parallelStream(baseData).sorted(cmp).forEach(sorted::add);
>            System.out.println("incorrect: " +
>   Arrays.toString(sorted.toArray()));
> 
>            //Following line is just intended for providing the correct
>   result in order to compare with incorrect one
>            System.out.println("correct:   " +
>   Arrays.toString(Arrays.parallelStream(baseData).sorted(cmp).toArray()));
>        }
>   }
> 
> This code produces following output:
> 
>   incorrect: [1, _*4, 2*_, 7, _*9, 3*_]
>   correct:   [1, 2, 3, 4, 7, 9]
> 
> The result [1, _*4, 2*_, 7, _*9, 3*_]is obviously incorrect.
> BTW: the first line of the output is non deterministic. Provided the code is run several times different output is produced each time. For instance:
> 
>   incorrect: [3, 2, 7, 9, 4]
>   correct: [1, 2, 3, 4, 7, 9]
> 
>   incorrect: [1, 4, 7, 2, 9, 3]
>   correct: [1, 2, 3, 4, 7, 9]
> 
>   incorrect: [4, 2, 3, 7, 9]
>   correct: [1, 2, 3, 4, 7, 9]
> 
>   incorrect: [1, 2, 3, 7, 4]
>   correct: [1, 2, 3, 4, 7, 9]
> 
> After changing parallel mode to sequential the code works fine. This issue is reproduced on Windows 7x64.
> 
> Please confirm it's a bug and it will be fixed.
> 
> The code is attached for your convenience.
> 
> Thank you,
> Georgiy.
> <SortedIssue2.java>