Java completableFuture：如何并行化多个 CompletableFuture

Question

Here my code:

final int startIndex = LoaderConstants.ServeiTerritorial.DEFAULT_START_INDEX; 
final long pageSize = LoaderConstants.ServeiTerritorial.DEFAULT_PAGE_SIZE;
final String tableName = LoaderConstants.ServeiTerritorial.TIPUS_VIA_TABLE_NAME;
final String ownerType = LoaderConstants.ServeiTerritorial.TIPUS_VIA_OWNER_TYPE;

LongStream
    .iterate(1, n -> n <= 1000 / pageSize, n -> n+1)
    .mapToObj(pageNumber -> this.buildCompletableFutureOfResultSetType(tableName, ownerType, pageNumber * pageSize, pageSize))
    .map(CompletableFuture::join)

What I'm trying to get is to parallelize each item.

First of all, for each page, I build a CompletableFuture :

/**
* Builds a {@link CompletableFuture} in order to get oid.
*/
private CompletableFuture<ResultSetType> buildCompletableFutureOfResultSetType(
    final String tableName,
    final String owner,
    final long pageNumber,
    final long pageSize
) {
    Supplier<ResultSetType> supplier = () -> this.serveiTerritorialCatalegsClientRepository.getCataleg(tableName, null, owner, null, null, null, pageNumber, pageSize);

    return CompletableFuture.supplyAsync(supplier, this.servidorTerminologicExecutor);
}

I thought it was working fine, when I've realized that it's performed sequentially. Those are my logs:

2021-07-21 12:46:35.894 DEBUG [hes-mpi-imdg-loader,9ffb4627802548bf,9ffb4627802548bf,false] 22536 --- [ool-1-thread-38] ServeiTerritorialOidClientRepositoryImpl : Preparant crida a servei-territorial.getOid (oid: 2.16.724.4.402)
2021-07-21 12:46:35.895 DEBUG [hes-mpi-imdg-loader,9ffb4627802548bf,9ffb4627802548bf,false] 22536 --- [ool-1-thread-38] ServeiTerritorialOidClientRepositoryImpl : Creant petició cap a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>13801</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:35.896 DEBUG [hes-mpi-imdg-loader,9ffb4627802548bf,9ffb4627802548bf,false] 22536 --- [ool-1-thread-38] ServeiTerritorialOidClientRepositoryImpl : Enviant petició a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>13801</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:35.985 DEBUG [hes-mpi-imdg-loader,501c638f07534a0d,501c638f07534a0d,false] 22536 --- [ool-1-thread-39] ServeiTerritorialOidClientRepositoryImpl : Preparant crida a servei-territorial.getOid (oid: 2.16.724.4.402)
2021-07-21 12:46:35.986 DEBUG [hes-mpi-imdg-loader,501c638f07534a0d,501c638f07534a0d,false] 22536 --- [ool-1-thread-39] ServeiTerritorialOidClientRepositoryImpl : Creant petició cap a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>13901</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:35.987 DEBUG [hes-mpi-imdg-loader,501c638f07534a0d,501c638f07534a0d,false] 22536 --- [ool-1-thread-39] ServeiTerritorialOidClientRepositoryImpl : Enviant petició a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>13901</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:36.057 DEBUG [hes-mpi-imdg-loader,d840cc01f137b810,d840cc01f137b810,false] 22536 --- [ool-1-thread-40] ServeiTerritorialOidClientRepositoryImpl : Preparant crida a servei-territorial.getOid (oid: 2.16.724.4.402)
2021-07-21 12:46:36.058 DEBUG [hes-mpi-imdg-loader,d840cc01f137b810,d840cc01f137b810,false] 22536 --- [ool-1-thread-40] ServeiTerritorialOidClientRepositoryImpl : Creant petició cap a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>14001</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:36.061 DEBUG [hes-mpi-imdg-loader,d840cc01f137b810,d840cc01f137b810,false] 22536 --- [ool-1-thread-40] ServeiTerritorialOidClientRepositoryImpl : Enviant petició a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>14001</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:36.141 DEBUG [hes-mpi-imdg-loader,9ec5d6687eb6e9aa,9ec5d6687eb6e9aa,false] 22536 --- [ool-1-thread-41] ServeiTerritorialOidClientRepositoryImpl : Preparant crida a servei-territorial.getOid (oid: 2.16.724.4.402)
2021-07-21 12:46:36.142 DEBUG [hes-mpi-imdg-loader,9ec5d6687eb6e9aa,9ec5d6687eb6e9aa,false] 22536 --- [ool-1-thread-41] ServeiTerritorialOidClientRepositoryImpl : Creant petició cap a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>14101</startIndex><pageSize>100</pageSize></request>)
2021-07-21 12:46:36.142 DEBUG [hes-mpi-imdg-loader,9ec5d6687eb6e9aa,9ec5d6687eb6e9aa,false] 22536 --- [ool-1-thread-41] ServeiTerritorialOidClientRepositoryImpl : Enviant petició a servei-territorial (request: <request><oid>2.16.724.4.402</oid><startIndex>14101</startIndex><pageSize>100</pageSize></request>)

I don't quite figure out what I'm doing wrong.

Any ideas?

Answer 1

You're running this sequential stream with a CompletableFutre.join call in an intermediate step.

The problem is: elements traverse the (sequential) stream one at a time, and go through each step. This, of course, includes the join() call. That means that for each element, the entire operation (including the "async" part, this.serveiTerritorialCatalegsClientRepository.getCataleg ) has to complete before the following element enters processing.

To solve this, force your pipeline to create futures for all elements before join() calls start being invoked. Something like this should work:

List<CompletableFuture<ResultSetType>> submittedTasks = LongStream
    .iterate(1, n -> n <= 1000 / pageSize, n -> n+1)
    .mapToObj(pageNumber -> this.buildCompletableFutureOfResultSetType(tableName, 
                                       ownerType, pageNumber * pageSize,
                                       pageSize))
    .collect(Collectors.toList());

That way, the terminal collect() will gather all submitted CompletableFuture objects without blocking. You don't need to collect to a list, but any terminal operation that is going to force the asynchronous tasks to be submitted should do; as long as you can manage to call join on each task separately.

After that, you can block, knowing that all asynchronous tasks have been started or at least queued.

submittedTasks.stream()
    .map(CompletableFuture::join)
    .forEach(...) //some terminal operation

This way, you avoid unnecessary blocking code.

---

I notice that I mentioned "sequential" a few times, but that's not to imply that running the same code on a parallel stream will solve it. The main problem will persist, although throughput may improve because you'd be blocking in parallel