Question

I am new to multi-threading and would like to avoid a race condition that is occurring in the below code. In the release() method there is a line available.add(resource) and in the remove() method there is a line available.remove(resource). So my question is how do I synchronize the ‘resource’ variable to avoid this race condition?

    package threadpool;
    import java.util.concurrent.BlockingQueue;
    import java.util.concurrent.ConcurrentHashMap;
    import java.util.concurrent.ConcurrentMap;
    import java.util.concurrent.CountDownLatch;
    import java.util.concurrent.LinkedBlockingQueue;
    import java.util.concurrent.TimeUnit;

    public class ResourcePoolImpl<R> implements ResourcePool<R> {

    private static final String CLOSED_POOL_EXCEPTION = "Pool is closed,cannot aquire resource.";

    private static final String  RELEASE_EXCEPTION = "Unaquired resource, cannot release it.";

    private volatile boolean open = false;

    private final BlockingQueue<R> available = new LinkedBlockingQueue<R>();

    private final ConcurrentMap<R, CountDownLatch> aquired = new ConcurrentHashMap<R,  CountDownLatch>();

    public R acquire() throws InterruptedException {
    if ( !open ) { throw new IllegalStateException( CLOSED_POOL_EXCEPTION ); }
    final R resource = available.take();
    if ( resource != null ) {
        aquired.put( resource, new CountDownLatch( 1 ) );
    }
    return resource;
    }


   public R acquire( final long timeout, final TimeUnit timeUnit ) throws InterruptedException {
    if ( !open ) { throw new IllegalStateException( CLOSED_POOL_EXCEPTION ); }

    final R resource = available.poll( timeout, timeUnit );
    if ( resource != null ) {
        aquired.put( resource, new CountDownLatch( 1 ) );
    }
    return resource;
    }


    public boolean add( final R resource ) 
    {
    return available.add( resource );
    } 

    public void close() throws InterruptedException {
    open = false;
    for ( final CountDownLatch latch : aquired.values() ) {
        latch.await();
    }
    }

    public void closeNow() {
    open = false;
    }

    public boolean isOpen() {
    return open;
    }

    public void open() {
    open = true;
    }

    public void release( final R resource ) 
    {
    final CountDownLatch latch = aquired.get( resource );
    if ( latch == null ) { throw new IllegalArgumentException( RELEASE_EXCEPTION ); }
    available.add( resource );
    latch.countDown();
    }

    public boolean remove( final R resource ) throws InterruptedException 
    {   

    final CountDownLatch latch = aquired.get( resource );
    if ( latch != null ) {
        latch.await();
    }
    return available.remove( resource );
    }


    public boolean removeNow( final R resource ) {
    return available.remove( resource );
    }

 }

Answer 1

Declare a

final Object mutex = new Object();

and have all methods that do read/write operations on shared collections aquire the mutex before doing the operations, or make decisions based on shared data, execute within a synchronized block:

synchronized (mutex) {
     // .. guaranteed single-threaded access here
     //   (for instance, contents of aquire() or release(); 
     //      also add() or any other collection access)
}

You can then use the simpler non-concurrent collection classes, as within the mutex-guarded areas, there cannot be any multi-threaded access.

Concurrent collections simply wrap their accesses within their own internal mutual-exclusion locks — but the problem, as you explain in your comments, is that aquired and available may be updated independently of each other, which you definitely do not want.

Therefore: simplify your code by declaring and using a single mutex for all critical-area accesses.