java - ConcurrentHashMap and compound operations

Question

Hashtable and Collections.synchronizedMap are thread safe but still compound operations like

if (!map_obj.containsKey(key)) {
   map_obj.put(key, value);
}

needs external synchronization as:

synchronized(map_obj) {
    if (!map_obj.containsKey(key)) {
       map_obj.put(key, value);
    }
}

Suppose we have ConcurrentHashMap(CHM) instead of Hashtable or HashMap. CHM provides an alternative putIfAbsent() method for the above compound operation, thus removing the need of external synchronization.

But suppose there is no putIfAbsent() provided by CHM. Then can we write following code:

synchronized(concurrenthashmap_obj) {
    if (!concurrenthashmap_obj.containsKey(key)) {
       concurrenthashmap_obj.put(key, value);
    }
}

I mean can we use external synchronization on CHM object?Will it work?

For above compound operation there is putIfAbsent() method in CHM but how can we achieve thread safety for other compound operations if we are using CHM. I mean can we use external synchronization on CHM object?

Answer 1

No, you cannot use external synchronization to ensure atomicity of compound operations over ConcurrentHashMap.

To be precise, you can use external synchronization to ensure atomicity of compound operations, but only if all operations with ConcurrentHashMap are synchronized over the same lock as well (though use of ConcurrentHashMap won't make sense in this case - you can replace it with regular HashMap).

Approach with external synchronization works with Hashtable and Collections.synchronizedMap() only because they guarantee that their primitive operations are synchronized over these objects as well. Since ConcurrentHashMap doesn't provide such a guarantee, primitive operations may interfere with execution of your compound operations, breaking their atomicity.

However, ConcurrentHashMap provides number of methods that can be used to implement compound operations in optimistic manner:

• putIfAbsent(key, value)
• remove(key, value)
• replace(key, value)
• replace(key, oldValue, newValue)

You can use these operation to implement certain compound operations without explict synchronization, the same way as you would do for AtomicReference, etc.