Question

I wonder if the following would be a correct implementation

    ExecutorService pool = Executors.newFixedThreadPool(MAX_NSH_THREADS);
    Set<Future<Void>> futureRequest = new HashSet<Future<Void>>();

    for (String host : SomeCollection)) {
        Callable<Void> callable = new FileExtractor(j);
        Future<Void> future = pool.submit(callable);
        futureRequest.add(future);
    }

    for (Future<Void> future : futureRequest) {
        try {
            future.get(); 
        } catch (Exception e) {
            logger.error(e);
        }
    }

    pool.shutdown();

According to Javadoc, future.get() waits for execution to complete for each thread, which (as i understand it) means that for each of the future, we will wait to receive the results separately. Where is the benefit coming from then, or am i not doing it right?

Answer 1

You are doing it right.

Lets say that SomeCollection contains 100 items, and that FileExtractor takes 5 seconds to run, and that your ExecutorService thread pool contains 100 threads.

If you start things as you have implemented above, it is expected that the code would run for around 5 seconds because FileExtractor would likely be I/O bound. (assuming maximum CPU efficiency).

If you didn’t use Future‘s, and everything ran serially, this code would be running for about 500 seconds instead.

The key is that Future#get() waits for the result to be populated by the Thread started by submitting your Callable to the ExecutorService, rather than waiting at the ExecutorService#submit(Callable) method.