Is iterating ConcurrentHashMap values thread safe?

Question

What does it mean?

That means that each iterator you obtain from a ConcurrentHashMap is designed to be used by a single thread and should not be passed around. This includes the syntactic sugar that the for-each loop provides.

What happens if I try to iterate the map with two threads at the same time?

It will work as expected if each of the threads uses it's own iterator.

What happens if I put or remove a value from the map while iterating it?

It is guaranteed that things will not break if you do this (that's part of what the "concurrent" in ConcurrentHashMap means). However, there is no guarantee that one thread will see the changes to the map that the other thread performs (without obtaining a new iterator from the map). The iterator is guaranteed to reflect the state of the map at the time of it's creation. Futher changes may be reflected in the iterator, but they do not have to be.

In conclusion, a statement like

for (Object o : someConcurrentHashMap.entrySet()) {
    // ...
}

will be fine (or at least safe) almost every time you see it.

You may use this class to test two accessing threads and one mutating the shared instance of ConcurrentHashMap:

import java.util.Map;
import java.util.Random;
import java.util.UUID;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;

public class ConcurrentMapIteration
{
  private final Map<String, String> map = new ConcurrentHashMap<String, String>();

  private final static int MAP_SIZE = 100000;

  public static void main(String[] args)
  {
    new ConcurrentMapIteration().run();
  }

  public ConcurrentMapIteration()
  {
    for (int i = 0; i < MAP_SIZE; i++)
    {
      map.put("key" + i, UUID.randomUUID().toString());
    }
  }

  private final ExecutorService executor = Executors.newCachedThreadPool();

  private final class Accessor implements Runnable
  {
    private final Map<String, String> map;

    public Accessor(Map<String, String> map)
    {
      this.map = map;
    }

    @Override
    public void run()
    {
      for (Map.Entry<String, String> entry : this.map.entrySet())
      {
        System.out.println(
            Thread.currentThread().getName() + " - [" + entry.getKey() + ", " + entry.getValue() + ']'
        );
      }
    }
  }

  private final class Mutator implements Runnable
  {

    private final Map<String, String> map;
    private final Random random = new Random();

    public Mutator(Map<String, String> map)
    {
      this.map = map;
    }

    @Override
    public void run()
    {
      for (int i = 0; i < 100; i++)
      {
        this.map.remove("key" + random.nextInt(MAP_SIZE));
        this.map.put("key" + random.nextInt(MAP_SIZE), UUID.randomUUID().toString());
        System.out.println(Thread.currentThread().getName() + ": " + i);
      }
    }
  }

  private void run()
  {
    Accessor a1 = new Accessor(this.map);
    Accessor a2 = new Accessor(this.map);
    Mutator m = new Mutator(this.map);

    executor.execute(a1);
    executor.execute(m);
    executor.execute(a2);
  }
}

No exception will be thrown.

Sharing the same iterator between accessor threads can lead to deadlock:

import java.util.Iterator;
import java.util.Map;
import java.util.Random;
import java.util.UUID;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;

public class ConcurrentMapIteration
{
  private final Map<String, String> map = new ConcurrentHashMap<String, String>();
  private final Iterator<Map.Entry<String, String>> iterator;

  private final static int MAP_SIZE = 100000;

  public static void main(String[] args)
  {
    new ConcurrentMapIteration().run();
  }

  public ConcurrentMapIteration()
  {
    for (int i = 0; i < MAP_SIZE; i++)
    {
      map.put("key" + i, UUID.randomUUID().toString());
    }
    this.iterator = this.map.entrySet().iterator();
  }

  private final ExecutorService executor = Executors.newCachedThreadPool();

  private final class Accessor implements Runnable
  {
    private final Iterator<Map.Entry<String, String>> iterator;

    public Accessor(Iterator<Map.Entry<String, String>> iterator)
    {
      this.iterator = iterator;
    }

    @Override
    public void run()
    {
      while(iterator.hasNext()) {
        Map.Entry<String, String> entry = iterator.next();
        try
        {
          String st = Thread.currentThread().getName() + " - [" + entry.getKey() + ", " + entry.getValue() + ']';
        } catch (Exception e)
        {
          e.printStackTrace();
        }

      }
    }
  }

  private final class Mutator implements Runnable
  {

    private final Map<String, String> map;
    private final Random random = new Random();

    public Mutator(Map<String, String> map)
    {
      this.map = map;
    }

    @Override
    public void run()
    {
      for (int i = 0; i < 100; i++)
      {
        this.map.remove("key" + random.nextInt(MAP_SIZE));
        this.map.put("key" + random.nextInt(MAP_SIZE), UUID.randomUUID().toString());
      }
    }
  }

  private void run()
  {
    Accessor a1 = new Accessor(this.iterator);
    Accessor a2 = new Accessor(this.iterator);
    Mutator m = new Mutator(this.map);

    executor.execute(a1);
    executor.execute(m);
    executor.execute(a2);
  }
}

As soon as you start sharing the same Iterator<Map.Entry<String, String>> among accessor and mutator threads java.lang.IllegalStateExceptions will start popping up.

import java.util.Iterator;
import java.util.Map;
import java.util.Random;
import java.util.UUID;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;

public class ConcurrentMapIteration
{
  private final Map<String, String> map = new ConcurrentHashMap<String, String>();
  private final Iterator<Map.Entry<String, String>> iterator;

  private final static int MAP_SIZE = 100000;

  public static void main(String[] args)
  {
    new ConcurrentMapIteration().run();
  }

  public ConcurrentMapIteration()
  {
    for (int i = 0; i < MAP_SIZE; i++)
    {
      map.put("key" + i, UUID.randomUUID().toString());
    }
    this.iterator = this.map.entrySet().iterator();
  }

  private final ExecutorService executor = Executors.newCachedThreadPool();

  private final class Accessor implements Runnable
  {
    private final Iterator<Map.Entry<String, String>> iterator;

    public Accessor(Iterator<Map.Entry<String, String>> iterator)
    {
      this.iterator = iterator;
    }

    @Override
    public void run()
    {
      while (iterator.hasNext())
      {
        Map.Entry<String, String> entry = iterator.next();
        try
        {
          String st =
              Thread.currentThread().getName() + " - [" + entry.getKey() + ", " + entry.getValue() + ']';
        } catch (Exception e)
        {
          e.printStackTrace();
        }

      }
    }
  }

  private final class Mutator implements Runnable
  {

    private final Random random = new Random();

    private final Iterator<Map.Entry<String, String>> iterator;

    private final Map<String, String> map;

    public Mutator(Map<String, String> map, Iterator<Map.Entry<String, String>> iterator)
    {
      this.map = map;
      this.iterator = iterator;
    }

    @Override
    public void run()
    {
      while (iterator.hasNext())
      {
        try
        {
          iterator.remove();
          this.map.put("key" + random.nextInt(MAP_SIZE), UUID.randomUUID().toString());
        } catch (Exception ex)
        {
          ex.printStackTrace();
        }
      }

    }
  }

  private void run()
  {
    Accessor a1 = new Accessor(this.iterator);
    Accessor a2 = new Accessor(this.iterator);
    Mutator m = new Mutator(map, this.iterator);

    executor.execute(a1);
    executor.execute(m);
    executor.execute(a2);
  }
}

It means that you should not share an iterator object among multiple threads. Creating multiple iterators and using them concurrently in separate threads is fine.

This might give you a good insight

ConcurrentHashMap achieves higher concurrency by slightly relaxing the promises it makes to callers. A retrieval operation will return the value inserted by the most recent completed insert operation, and may also return a value added by an insertion operation that is concurrently in progress (but in no case will it return a nonsense result). Iterators returned by ConcurrentHashMap.iterator() will return each element once at most and will not ever throw ConcurrentModificationException, but may or may not reflect insertions or removals that occurred since the iterator was constructed. No table-wide locking is needed (or even possible) to provide thread-safety when iterating the collection. ConcurrentHashMap may be used as a replacement for synchronizedMap or Hashtable in any application that does not rely on the ability to lock the entire table to prevent updates.

Regarding this:

However, iterators are designed to be used by only one thread at a time.

It means, while using iterators produced by ConcurrentHashMap in two threads are safe, it may cause an unexpected result in the application.

What does it mean?

It means that you should not try to use the same iterator in two threads. If you have two threads that need to iterate over the keys, values or entries, then they each should create and use their own iterators.

What happens if I try to iterate the map with two threads at the same time?

It is not entirely clear what would happen if you broke this rule. You could just get confusing behavior, in the same way that you do if (for example) two threads try to read from standard input without synchronizing. You could also get non-thread-safe behavior.

But if the two threads used different iterators, you should be fine.

What happens if I put or remove a value from the map while iterating it?

If the two threads are using the same iterator: see above. You are liable to get confusing and possibly non-thread-safe behavior.

If the threads are using different iterators, the javadoc section that you quoted adequately answers it. Basically, it is not defined whether one thread / iterator will see the effects of any concurrent insertions, updates or deletions made by the other thread / iterator. However, the insertions / updates / deletes will be made according to the concurrency properties of the map.