Java 8 parallel stream gotchas

When using parallel streams in Java you are using the ForkJoinPool.commonPool() singleton and therefore sharing a single thread pool across the entire program. It, by default, has one less threads as you have processors, as returned by Runtime.availableProcessors() (can be overriden with -Djava.util.concurrent.ForkJoinPool.common.parallelism).

As a consequence:

• You have no control over the pool properties/size.
• There is no isolation between pools so long running tasks or IO locking calls may cause performance problems.

References:

• https://dzone.com/articles/think-twice-using-java-8
• https://dzone.com/articles/whats-wrong-java-8-part-iii

The solution is executing the stream in a specific ForkJoinPool:

ForkJoinPool forkJoinPool = new ForkJoinPool(2);
forkJoinPool.submit(() ->
    IntStream.range(1, 1_000_000)
      .parallel()
      .filter(PrimesPrint::isPrime)
      .collect(toList())
).get();

References:

• https://github.com/vandekeiser/parallel-stream-fork-join-pool
• http://blog.krecan.net/2014/03/18/how-to-specify-thread-pool-for-java-8-parallel-streams/

Other solution using CompletableFuture:

ForkJoinPool forkJoinPool = new ForkJoinPool(2);
CompletableFuture<List<Integer>> primes = CompletableFuture.supplyAsync(() ->
  range(1, 1_000_000)
    .parallel()
    .filter(PrimesPrint::isPrime)
    .collect(toList()), 
  forkJoinPool
);

See next blog entry on  CompletableFuture.