ConcurrentHashMap has been a very popular data structure these days. In this talk, we will walk through the new features that got introduced in Java 8.

Java 8 introduced the forEach, search, and reduce methods, which are pretty much to support parallelism. These three operations are available in four forms: accepting functions with keys, values, entries, and key-value pair arguments.

All of those methods take a first argument called parallelismThreshold.

We will take the example of this map definition:

ConcurrentHashMap<String, Integer> hashMap = new ConcurrentHashMap<>();
hashMap.put("A", 1);
hashMap.put("B", 2);
hashMap.put("C", 3);
hashMap.put("D", 4);
hashMap.put("E", 5);
hashMap.put("F", 6);
hashMap.put("G", 7);

parallelismThreshold

This is to define how you wanted to execute the operations — sequentially or in parallel. Suppose you have given a parallelismThreshold as 2. So as long as there are fewer than two elements in your map, it would be sequential. Otherwise, it's parallel (depends on the JVM).

hashMap.forEach(2, (k, v) -> System.out.println("key->" + k + "is related with value-> " + v +", by thread-> "+ Thread.currentThread().getName())); 

It produced the below o/p on my machine (you can see two different threads in action — main and ForkJoinPool.commonPool-worker-1):

key->A is related with value-> 1, by thread-> main
key->D is related with value-> 4, by thread-> ForkJoinPool.commonPool-worker-1
key->B is related with value-> 2, by thread-> main
key->E is related with value-> 5, by thread-> ForkJoinPool.commonPool-worker-1
key->C is related with value-> 3, by thread-> main
key->F is related with value-> 6, by thread-> ForkJoinPool.commonPool-worker-1
key->G is related with value-> 7, by thread-> ForkJoinPool.commonPool-worker-1

forEach

This method has the signature:

public void forEach(long parallelismThreshold, BiConsumer action)

We have already seen the parallelismThreshold. Now, let's dive into BiConsumer. It's a FunctionalInterface that accepts two input arguments and returns no result. It has this definition:

@FunctionalInterface
public interface BiConsumer<T, U> {
  void accept(T t, U u);
}

So forEach takes parallelismThreshold and BiConsumer and calls this method:

new ForEachMappingTask (null, batchFor(parallelismThreshold), 0, 0, table, action).invoke();

The batchFor(parallelismThreshold) method is to get parallelism using the ForkJoinPool.getCommonPoolParallelism() method, as we have seen above in the parallelismThreshold example.

Here, action is the BiConsumer that we passed in forEach method.

ForEachMappingTask is a static final class that extends BulkTask, and BulkTask extends the abstract class CountedCompleter, which extends the abstract class ForkJoinTask. So in short, ForEachMappingTask is to support ForkJoinTask.

The ForEachMappingTask class has the method compute as below:

public final void compute() {
final BiConsumer<? super K, ? super V> action;
    if ((action = this.action) != null) {
      for (int i = baseIndex, f, h; batch > 0 &&
      (h = ((f = baseLimit) + i) >>> 1) > i;) {
        addToPendingCount(1);
        new ForEachMappingTask<K,V> (this, batch >>>= 1, 
            baseLimit = h, f, tab, action).fork();
      }
      for (Node<K,V> p; (p = advance()) != null; )
      action.accept(p.key, p.val);
 propagateCompletion();
    }
}

Each time we execute a sub-task, the addToPendingCount(int) method gets invoked as in the above code. This is just to have a counter, which is to determine if task execution is completed.

The compute method above is responsible for adding the sub-task to a queue (shared/unshared queue) using the fork method of ForkJoinTask, as seen in the above code.

Let's understand the BiConsumer's accept method with the example below:

import java.util.function.BiConsumer;
public class BiConsumerTest {
  public static void main(String[] args) {
    BiConsumer<String, String> biConsumer = (x, y) -> {
      System.out.println("Key => " + x + ", and value => "+ y);
    };
    biConsumer.accept("k", "arun");
  }
}

Other ForkJoinTasks, such as RecursiveTask and RecursiveAction, don't need any explicit call to finish their job, while in the case of CountedCompleter, an explicit call of propagateCompletion() is required.

It's the same thing you've seen in the compute method.

As we have seen above, calling the invoke method of ForEachMappingTask, since ForEachMappingTask extends ForkJoinTask; implementation as below -

public final V invoke() {
  int s;
  if ((s = doInvoke() & DONE_MASK) != NORMAL)
    reportException(s);
  return getRawResult();
}

Here, getRawResult() returns the result of the computation, which is null by default. Otherwise, it should be overridden.

Search

Let's see a search snippet:

String result = hashMap.search(1, (k, v) -> {
  System.out.println(Thread.currentThread().getName());
  if (k.equals("A"))
    return k +"-" +v;
  return null;
});
System.out.println("result => " +result);

O/P:

main
ForkJoinPool.commonPool-worker-2
ForkJoinPool.commonPool-worker-2
result => A-1

The method signature of the search method is:

The search method calls the invoke method of SearchMappingsTask. SearchMappingsTask is a subclass of BulkTask, which we have seen already. SearchmappingTask has a compute method as below:

public final void compute() {
  final BiFunction<? super K, ? super V, ? extends U> searchFunction;
  final AtomicReference<U> result;
  if ((searchFunction = this.searchFunction) != null &&
   (result = this.result) != null) {
  for (int i = baseIndex, f, h; batch > 0 &&
 (h = ((f = baseLimit) + i) >>> 1) > i;) {
      if (result.get() != null)
        return;
      addToPendingCount(1);
      new SearchMappingsTask<K,V,U>(this, batch >>>= 1, 
        baseLimit = h, f, tab, searchFunction, result).fork();
}
while (result.get() == null) {
      U u;
      Node<K,V> p;
      if ((p = advance()) == null) {
        propagateCompletion();
        break;
      }
      if ((u = searchFunction.apply(p.key, p.val)) != null) {
        if (result.compareAndSet(null, u))
          quietlyCompleteRoot();
        break;
      }
    }
  }
}

In above method, AtomicReference is used to have an object reference, which may be updated atomically, since it's a result. Each time a sub-task is executed, the addToPendingCount(int) method is getting invoked as in the above code. This is just to have a counter to determine if task execution is going well. And after that, it's calling the fork method to make it parallel.

Other ForkJoinTasks, such as RecursiveTask and RecursiveAction, don't need any explicit call to finish their job, while in the case of CountedCompleter, an explicit call of propagateCompletion() is required. Later, the quietlyCompleteRoot() method is completes the task normally.

Merge

As per the Javadoc, "If the specified key is not already associated with a (non-null) value, associates it with the given value. Otherwise, replaces the value with the results of the given remapping function, or removes if null. The entire method invocation is performed atomically. Some attempted update operations on this map by other threads may be blocked while computation is in progress, so the computation should be short and simple, and must not attempt to update any other mappings of this Map."

Let us look at the code snippet for the same:

ConcurrentHashMap<String, String> map = new ConcurrentHashMap<>();
map.put("X", "x");
System.out.println("1st ==> " +map);
System.out.println("2nd ==> " + map.merge("X", "x", (v1, v2) -> null));
System.out.println("3rd ==> " +map);
map.put("Y", "y");
map.put("X", "x1");
System.out.println("4th ==> " +map.merge("X", "x1", (v1, v2) -> "z"));
System.out.println("5th ==> " +map);
System.out.println("6th ==> " +map.merge(
 "X", "x1", (v1, v2) -> v2.concat("z")));
System.out.println("7th ==> " +map);

O/P:

1st ==> {X=x}
2nd ==> null
3rd ==> {}
4th ==> z
5th ==> {X=z, Y=y}
6th ==> x1z
7th ==> {X=x1z, Y=y}

The method signature of the merge method is:

Here, remappingFunction is the function that recomputes a value if present.

getOrDefault

As per the Javadoc, "It returns the value to which the specified key is mapped, or the given default value if this map contains no mapping for the key."

Let us look at the code snippet:

ConcurrentHashMap<String, Integer> defaultMap = 
  new ConcurrentHashMap<String, Integer>();
defaultMap.put("X", 30);
System.out.println(defaultMap);
System.out.println(defaultMap.getOrDefault("Y", 21));

O/P:

{X=30}
21

Compute

Generally, we do some computation on map values and store it back. In the concurrent model, it's difficult to manage, and that's the reason Java introduced the compute method. The entire method invocation is performed atomically.

The compute and computeIfPresent methods take a remapping function as an argument to compute a value, and remapping is of type BiFunction. The computeIfAbsent method takes an argument as mappingFunction to compute a value, and hence mappingFunction is of type Function.

Let us look at the code snippet to understand this:

ConcurrentHashMap<String, Integer> map1 = new ConcurrentHashMap<>();
map1.put("A", 1);
map1.put("B", 2);
map1.put("C", 3);
// Compute a new value for the existing key
System.out.println("1st print => " +map1.compute("A",
  (k, v) -> v == null ? 42 : v + 40));
System.out.println("2nd print => " + map1);
// This will add a new (key, value) pair
System.out.println("3rd print => " + map1.compute("X",
  (k, v) -> v == null ? 42 : v + 41));
System.out.println("4th print => " + map1);

//computeIfPresent method
System.out.println("5th print => " + map1.computeIfPresent("X", (k, v) -> v == null ? 42 : v + 10));
System.out.println("6th print => " + map1);

//computeIfAbsent method
System.out.println("7th print => " + map1.computeIfAbsent("Y", (k) -> 90));
System.out.println("8th print => " + map1);

O/P:

1st print => 41
2nd print => {A=41, B=2, C=3}
3rd print => 42
4th print => {A=41, B=2, C=3, X=42}
5th print => 52
6th print => {A=41, B=2, C=3, X=52}
7th print => 90
8th print => {A=41, B=2, C=3, X=52, Y=90}

Reduce 

The reduce method signature is:

public U reduce(long parallelismThreshold, BiFunction transformer, BiFunction reducer)

Here, transformer is a function returning the transformation for an element, or null if there is no transformation (in which case it is not combined), and reducer is a commutative associative combining function.

The reduce method calls the MapReduceMappingsTask's invoke method. MapReduceMappingsTask extends BulkTask, and we have seen this already.

Let us look at the example below:

ConcurrentHashMap<String, Integer> reducedMap = new ConcurrentHashMap<>();
reducedMap.put("One", 1);
reducedMap.put("Two", 2);
reducedMap.put("Three", 3);
System.out.println("reduce example => " 
 +reducedMap.reduce(2, (k, v) -> v*2, (total, elem) -> total + elem)); 

System.out.println("reduceKeys example => " 
 +reducedMap.reduceKeys(2, (key1, key2) -> key1.length() > key2.length() ? key1 + "-"+key2 : key2 + "-"+key1)); 

System.out.println("reduceValues example => " 
 +reducedMap.reduceValues(2, (v) -> v*2 , (value1, value2) -> value1 > value2 ? value1 - value2 : value2 - value1));
System.out.println("After reduce => " +reducedMap);

O/P:

reduce example => 12
reduceKeys example => Three-Two-One
reduceValues example => 0
After reduce => {One=1, Two=2, Three=3}

Have a great day learning. Enjoy!