I have a parallel stream like this:
mylist.stream().parallel().map(transform::docPDFTypeToItem)
.map(fdUtil::createEdiLine).forEach(document::add)
this is working on my two CPUs
now I'm trying to oversize the forkjoinpool because I have a lot of IO as described here:
new ForkJoinPool(20).submit(() ->
mylist.stream().parallel().map(transform::docPDFTypeToItem)
.map(fdUtil::createEdiLine).parallel().forEach(document::add)).get())
but this is not working well, as I have only 5 threads used:
08:54:48.423 [ForkJoinPool-1-worker-18]
08:54:48.426 [ForkJoinPool-1-worker-25]
08:54:48.427 [ForkJoinPool-1-worker-29]
08:54:48.430 [ForkJoinPool-1-worker-15]
08:54:48.431 [ForkJoinPool-1-worker-11]
08:54:50.426 [ForkJoinPool-1-worker-25]
08:54:50.427 [ForkJoinPool-1-worker-29]
08:54:48.423 [ForkJoinPool-1-worker-18]
08:54:48.426 [ForkJoinPool-1-worker-25]
08:54:48.427 [ForkJoinPool-1-worker-29]
08:54:48.430 [ForkJoinPool-1-worker-15]
08:54:48.431 [ForkJoinPool-1-worker-11]
08:54:50.426 [ForkJoinPool-1-worker-25]
08:54:50.427 [ForkJoinPool-1-worker-29]
...
If I use
System.setProperty("java.util.concurrent.ForkJoinPool.common.parallelism", "20");
it is working as expected (20 threads reused).
Do you know what's wrong? Does this .submit() sound like a hack?
I know I could use completableFuture like here but I think it is less readable.
