HashMap with weak values

Question

I'm implementing a cache for Objects stored persistently. The idea is:

• Method getObjectFromPersistence(long id); ///Takes about 3 seconds
• Method getObjectFromCache(long id) //Instantly

And have a method: getObject(long id) with the following pseudocode:

synchronized(this){
    CustomObject result= getObjectFromCache(id)
    if (result==null){
       result=getObjectFromPersistence(id);
       addToCache(result);
    }
    return result;
}

But I need to allow the CustomObject to be collected by the garbage collector. Until now I was using an HashMap<Long,WeakReference<CustomObject> for the implementation. The problem is that over the time the HashMap becomes filled of empty WeakReferences.

I've checked WeakHashMap but there the keys are weak (and the values are still strong references) so having the longs with WeakReferences have no sense.

Whats the best solution for solving this problem? Is there some "inverse WeakHashMap" or something similar?

Thanks

Answer 1

You can use the Guava MapMaker for this:

ConcurrentMap<Long, CustomObject> graphs = new MapMaker()
   .weakValues()
   .makeMap();

You can even include the computation part by replacing makeMap() with this:

   .makeComputingMap(
       new Function<Long, CustomObject>() {
         public CustomObject apply(Long id) {
           return getObjectFromPersistence(id);
         }
       });

Since what you are writing looks a lot like a cache, the newer, more specialized Cache (built via a CacheBuilder) might be even more relevant to you. It doesn't implement the Map interface directly, but provides even more controls that you might want for a cache.

You can refer to this for a detailed how to work for CacheBuilder and here is an example for fast access:

LoadingCache<Integer, String> cache = CacheBuilder.newBuilder()
   .maximumSize(100)
   .expireAfterWrite(10, TimeUnit.MINUTES)
   .build(
       new CacheLoader<Integer, String>() {
           @Override
           public String load(Integer id) throws Exception {
               return "value";
           }
       }
   ); 

Answer 2

A WeakReference is added to its ReferenceQueue supplied at the construction time when its reference is collected.

You could poll the ReferenceQueue whenever you access the cache, and hold a HashMap<WeakReference<CustomObject>,Long> to know which entry to remove if a reference is found in the queue.

Alternatively, if the cache is not frequently used, you can watch the queue in a separate thread.

Answer 3

Have you tried android.util.LruCache (its a SDK11 class but it's also in the compatibility package as android.support.v4.util.LruCache). It does not implement java.util.Map but works like a Map and you can define how much memory will it take and it will flush old (unused cached objects by itself).

Answer 4

You could start a "cleanup" - Thread every once in a while. Perhaps if your map size exceeds a threshold but at most every 5 minutes .... something like that.

Keep the cleanup cycles short to not block the main functionality.

Answer 5

You can also test WeakValueHashMap from jboss-common http://docs.jboss.org/jbossas/javadoc/4.0.2/org/jboss/util/collection/WeakValueHashMap.java.html

Answer 6

I think the best option (if a dependency on Guava is undesirable) would be to use a custom subclass of WeakReference that remembers its ID, so that your cleanup thread can remove the weak values during cleanup of the WeakReferences.

The implementation of the weak reference, with the necessary ReferenceQueue and cleanup thread would look something like this:

class CustomObjectAccess {

    private static final ReferenceQueue<CustomObject> releasedCustomObjects = 
                                                                  new ReferenceQueue<>();

    static {
        Thread cleanupThread = new Thread("CustomObject cleanup thread")                  
            while (true) {
                CustomObjectWeakReference freed = (CustomObjectWeakReference) 
                                CustomObjectWeakReference.releasedCustomObjects.remove();
                cache.remove(freed.id);
            }
        };
        cleanupThread.start();
    }

    private Map<CustomObjectID, CustomObjectWeakReference> cache;

    public CustomObject get(CustomObjectID id) {
        synchronized(this){
            CustomObject result= getFromCache(id);
            if (result==null) {
                result=getObjectFromPersistence(id);
                addToCache(result);
            }
        }
        return result;
    }

    private addToCache(CustomObject co) {
        cache.put(CustomObject.getID(), new CustomObjectWeakReference(co));
    }

    private getFromCache(CustomObjectID id) {
        WeakReference<CustomObject> weak = cache.get(id);
        if (weak != null) {
            return weak.get();
        }
        return null;
    }

    class CustomObjectWeakReference extends WeakReference<CustomObject> {

        private final CustomObjectID id;

        CustomObjectWeakReference(CustomObject co) {
            super(co, releasedCustomObjects);
            this.id = co.getID();
        }
    }
}

Answer 7

I had the need to store tagged weak objects and figured instead of using WeakHashMap<String, T>, I could just use WeakHashMap<T, String> instead.

This is Kotlin, but should apply to Java equally:

abstract class InstanceFactory<T> {
    @Volatile
    private var instances: MutableMap<T, String> = WeakHashMap<T, String>()

    protected fun getOrCreate(tag: String = SINGLETON, creator: () -> T): T =
        findByTag(tag)?.let {
            it
        } ?: synchronized(this) {
            findByTag(tag)?.let {
                it
            } ?: run {
                creator().also {
                    instances[it] = tag
                }
            }
        }

    private fun findByTag(tag: String): T? = instances.entries.find { it.value == tag }?.key

    companion object {
        const val SINGLETON = "singleton"
    }
}

This can be used as follows:

class Thing(private val dependency: Dep) { ... }

class ThingFactory(private val dependency: Dep) : InstanceFactory<Thing>() {

    createInstance(tag: String): Thing = getOrCreate(tag) { Thing(dependency) }

}

Simple singletons can be done like this:

object ThingFactory {
    getInstance(dependency: Dependency): Thing = getOrCreate { Thing(dependency) }
}

Answer 8

There is ReferenceMap in Apache Commons Collections, this is a map implementation with hard keys and soft values (the opposite of WeakHashMap).