I noticed that if I use StreamEx lib to parallel out my streams with a custom ForkJoinPool as below - the subsequent actions do run in parallel threads from that pool. However, if I add a map() operation and parallel the resulting stream - only one thread from the pool is used.
Below is the full code (without all imports) of a minimal working example that demonstrates this problem. The only difference between the executeAsParallelFromList() and the executeAsParallelAfterMap() methods is addition of .map(...) call before the .parallel() .
import one.util.streamex.StreamEx;
public class ParallelExample {
private static final Logger logger = LoggerFactory.getLogger(ParallelExample.class);
private static ForkJoinPool s3ThreadPool = new ForkJoinPool(3);
public static List<String> getTestList(){
int listSize = 10;
List<String> testList = new ArrayList<>();
for (int i=0; i<listSize; i++)
testList.add("item_" + i);
return testList;
}
public static void executeAsParallelFromList(){
logger.info("executeAsParallelFromList():");
List<String> testList = getTestList();
StreamEx<String> streamOfItems = StreamEx
.of(testList)
.parallel(s3ThreadPool);
logger.info("streamOfItems.isParallel(): {}", streamOfItems.isParallel());
streamOfItems.forEach(item -> handleItem(item));
}
public static void executeAsParallelAfterMap(){
logger.info("executeAsParallelAfterMap():");
List<String> testList = getTestList();
StreamEx<String> streamOfItems = StreamEx
.of(testList)
.map(item -> item+"_mapped")
.parallel(s3ThreadPool);
logger.info("streamOfItems.isParallel(): {}", streamOfItems.isParallel());
streamOfItems.forEach(item -> handleItem(item));
}
private static void handleItem(String item){
// do something with the item - just print for now
logger.info("I'm handling item: {}", item);
}
}
Unit test to execute both methods:
public class ParallelExampleTest {
@Test
public void testExecuteAsParallelFromList() {
ParallelExample.executeAsParallelFromList();
}
@Test
public void testExecuteAsParallelFromStreamEx() {
ParallelExample.executeAsParallelAfterMap();
}
}
Results of the execution:
08:49:12.992 [main] INFO marina.streams.ParallelExample - executeAsParallelFromList():
08:49:13.002 [main] INFO marina.streams.ParallelExample - streamOfItems.isParallel(): true
08:49:13.040 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_6
08:49:13.040 [ForkJoinPool-1-worker-2] INFO marina.streams.ParallelExample - I'm handling item: item_2
08:49:13.040 [ForkJoinPool-1-worker-3] INFO marina.streams.ParallelExample - I'm handling item: item_1
08:49:13.041 [ForkJoinPool-1-worker-2] INFO marina.streams.ParallelExample - I'm handling item: item_4
08:49:13.041 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_8
08:49:13.041 [ForkJoinPool-1-worker-3] INFO marina.streams.ParallelExample - I'm handling item: item_0
08:49:13.041 [ForkJoinPool-1-worker-2] INFO marina.streams.ParallelExample - I'm handling item: item_3
08:49:13.041 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_9
08:49:13.041 [ForkJoinPool-1-worker-3] INFO marina.streams.ParallelExample - I'm handling item: item_5
08:49:13.041 [ForkJoinPool-1-worker-2] INFO marina.streams.ParallelExample - I'm handling item: item_7
08:49:13.043 [main] INFO marina.streams.ParallelExample - executeAsParallelAfterMap():
08:49:13.043 [main] INFO marina.streams.ParallelExample - streamOfItems.isParallel(): true
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_0_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_1_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_2_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_3_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_4_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_5_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_6_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_7_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_8_mapped
08:49:13.044 [ForkJoinPool-1-worker-1] INFO marina.streams.ParallelExample - I'm handling item: item_9_mapped
As you can see, all three threads are being used when executing the executeAsParallelFromList(), but only one thread is used when executing the executeAsParallelAfterMap().
Why?
thanks!
Marina
NOTE: the example is deliberately simplistic - I tried to make it as minimal as possible to demo the issue. Obviously in real life there is much more going on in the map(), handleItem() , etc., and the input data is much more interesting (I'm trying to process AWS S3 buckets/prefixes in parallel).
Simple answer: it's a bug. I filed and fixed it. This was overlooked by tests as tests only check that all the operations are executed in the specified pool, but don't check whether different threads of pool are used (sometimes it's ok if parallelization does not work, e.g. for a stream of one element only).
A fix is available in 0.6.4 release. In previous releases to work-around the problem you may consider using .parallel().parallel(fjp)
: it should parallelize correctly.
Please consider reporting StreamEx problems to official StreamEx issue tracker. I visit StackOverflow only occasionally these days, so may overlook the problems reported here.
The problem is that as soon as you call the map(...)
method, StreamEx creates the underlying Java 8 stream with the sequential/parallel configuration as of that point, (i.e. sequential), and calling parallel(...)
after that does not appear to update the underlying Java 8 stream.
The solution depends on what you are trying to achieve. If you're happy for your map(...)
operation to be run in parallel as well, then just move the parallel(...)
operation up so that it is the first thing after the of(...)
.
However, if you want some operations to be carried out sequentially, before some parallel operations, then you'd be better using two streams. For example, following the style of your sample code:
public static void executeAsParallelAfterMapV2() {
logger.info("executeAsParallelAfterMapV2():");
List<String> testList = getTestList();
StreamEx<String> sequentialStream = StreamEx
.of(testList)
.map(item -> {
logger.info("Mapping {}", item);
return item + "_mapped";
});
logger.info("sequentialStream.isParallel(): {}", sequentialStream.isParallel());
List<String> afterSequentialProcessing = sequentialStream.toList();
StreamEx<String> streamOfItems = StreamEx.of(afterSequentialProcessing)
.parallel(s3ThreadPool);
logger.info("streamOfItems.isParallel(): {}", streamOfItems.isParallel());
streamOfItems.forEach(item -> handleItem(item));
}
This gives something like:
20:43:36.835 [main] INFO scott.streams.ParallelExample - executeAsParallelAfterMapV2():
20:43:36.883 [main] INFO scott.streams.ParallelExample - sequentialStream.isParallel(): false
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_0
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_1
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_2
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_3
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_4
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_5
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_6
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_7
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_8
20:43:36.886 [main] INFO scott.streams.ParallelExample - Mapping item_9
20:43:36.886 [main] INFO scott.streams.ParallelExample - streamOfItems.isParallel(): true
20:43:36.889 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_6_mapped
20:43:36.889 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - I'm handling item: item_2_mapped
20:43:36.890 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - I'm handling item: item_8_mapped
20:43:36.890 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_5_mapped
20:43:36.890 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - I'm handling item: item_4_mapped
20:43:36.890 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - I'm handling item: item_9_mapped
20:43:36.890 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_1_mapped
20:43:36.890 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - I'm handling item: item_3_mapped
20:43:36.890 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - I'm handling item: item_7_mapped
20:43:36.890 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_0_mapped
An Aside...
Out of interest, if you create a Java 8 stream directly, (not using StreamEx), and put the parallel()
operation below the map(...)
, then it does update the type of the (whole) stream to be parallel:
public static void executeAsParallelAfterMapJava8Stream() throws InterruptedException {
logger.info("executeAsParallelAfterMapJava8Stream():");
List<String> testList = getTestList();
s3ThreadPool.submit(() -> {
Stream<String> streamOfItems = testList.stream()
.map(item -> {
logger.info("Mapping {}", item);
return item + "_mapped";
})
.parallel();
logger.info("streamOfItems.isParallel(): {}", streamOfItems.isParallel());
streamOfItems.forEach(item -> handleItem(item));
}).join();
}
If you create a similar unit test then you get something similar to:
20:36:23.469 [main] INFO scott.streams.ParallelExample - executeAsParallelAfterMapJava8Stream():
20:36:23.517 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - streamOfItems.isParallel(): true
20:36:23.520 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - Mapping item_6
20:36:23.520 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - Mapping item_2
20:36:23.520 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - Mapping item_8
20:36:23.520 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_6_mapped
20:36:23.520 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - I'm handling item: item_2_mapped
20:36:23.520 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - I'm handling item: item_8_mapped
20:36:23.520 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - Mapping item_5
20:36:23.520 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - Mapping item_4
20:36:23.520 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - Mapping item_9
20:36:23.520 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_5_mapped
20:36:23.520 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - I'm handling item: item_4_mapped
20:36:23.520 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - I'm handling item: item_9_mapped
20:36:23.520 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - Mapping item_1
20:36:23.520 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - Mapping item_3
20:36:23.520 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - Mapping item_7
20:36:23.521 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_1_mapped
20:36:23.521 [ForkJoinPool-1-worker-2] INFO scott.streams.ParallelExample - I'm handling item: item_3_mapped
20:36:23.521 [ForkJoinPool-1-worker-3] INFO scott.streams.ParallelExample - I'm handling item: item_7_mapped
20:36:23.521 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - Mapping item_0
20:36:23.521 [ForkJoinPool-1-worker-1] INFO scott.streams.ParallelExample - I'm handling item: item_0_mapped
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With