Running CompletableFuture.thenAccept from the calling thread?

Question

I want to call CompletableFuture.supplyAsync() to delegate a blocking task to another thread. Once that task completes I would like for the CompletableFuture.thenAccept consumer to run in the context of the calling thread.

For example:

// Thread 1

CompletableFuture.supplyAsync(() -> {
   // Thread 2

   return BlockingMethod();
}).thenAccept((
   Object r) -> {

   // Thread 1
});

The following code suggests that CompletableFuture.thenAccept runs in its own thread; probably the same pool as CompletableFuture.supplyAsync as I get the same thread ID when I run it:

System.out.println("Sync thread supply " + Thread.currentThread().getId());

CompletableFuture.supplyAsync(() -> {

   System.out.println("Async thread " + Thread.currentThread().getId());

   try {
      Thread.sleep(2000);
   }
   catch (Exception e) {
      e.printStackTrace();
   }

   return true;
}).thenAccept((
   Boolean r) -> {

   System.out.println("Sync thread consume " + Thread.currentThread().getId());
});

Thread.sleep(3000);

Is it possible to have CompletableFuture.thenAccept run concurrently with the calling thread?

Answer 1

The CompletableFuture will only execute the Consumer you register with thenAccept when the receiver CompletableFuture (one returned by supplyAsync ) is completed, since it needs the value it was completed with.

If the receiver CompletableFuture is complete when thenAccept is invoked, then the Consumer will be executed in the calling thread. Otherwise, it will execute on whatever thread completes the Supplier submitted to supplyAsync .

Is it possible to have CompletableFuture.thenAccept run concurrently with the calling thread?

This is a confusing question because a thread can only run one thing at a time. There's no concurrently for a single thread. Concurrently is a property that spans multiple threads.

If you want the Consumer to run on the same thread that invoked thenAccept , then join on the CompletableFuture , blocking this thread until the future is completed. You can then execute the Consumer yourself or call thenAccept to execute it for you.

For example

CompletableFuture<Boolean> receiver = CompletableFuture.supplyAsync(() -> {
    System.out.println("Async thread " + Thread.currentThread().getId());

    try {
        Thread.sleep(2000);
    } catch (Exception e) {
        e.printStackTrace();
    }

    return true;
});

receiver.join();
Consumer<Boolean> consumer = (Boolean r) -> {
    System.out.println("Sync thread consume " + Thread.currentThread().getId());
};

consumer.accept(receiver.get());

(Exception handling omitted.)

---

If you want the Consumer to run in parallel with the Supplier supplied to supplyAsync , that's not possible. That Consumer is meant to consume the value produced by the Supplier . That value isn't available until the Supplier is finished.

Answer 2

If I understand your idea right that you want to fork I/O-intensive tasks but do all processing in the "event loop" (think Javascript), then your code could be translated to

Executor eventLoop = Executors.newSingleThreadExecutor();
Executor ioMultiplexor = Executors.newCachedThreadPool();

eventLoop.execute(() -> {
    System.out.println("event loop thread supply " + Thread.currentThread().getId());
    CompletableFuture.supplyAsync(() -> {
        System.out.println("I/O multiplexor thread " + Thread.currentThread().getId());
        try {
            Thread.sleep(2000);
        } catch (Exception e) {
            e.printStackTrace();
        }
        return true;
    }, ioMultiplexor).thenAcceptAsync((Boolean r) -> {
        System.out.println("event loop thread consume " + Thread.currentThread().getId());
    }, eventLoop);
});

Thread.sleep(3000);
// shut down executors

This prints

event loop thread supply 10
I/O multiplexor thread 11
event loop thread consume 10

If this code is used for some request handling where there can be many concurrent requests, you would probably want to have one global eventLoop and one global ioMultiplexer , and you would also want to release the main request handling thread once it finishes submitting the task onto eventLoop .