Java 同步块 VS Collections.synizedMap

是否设置了以下代码来正确同步 synchronizedMap上的调用？

public class MyClass {
private static Map<String, List<String>> synchronizedMap = Collections.synchronizedMap(new HashMap<String, List<String>>());


public void doWork(String key) {
List<String> values = null;
while ((values = synchronizedMap.remove(key)) != null) {
//do something with values
}
}


public static void addToMap(String key, String value) {
synchronized (synchronizedMap) {
if (synchronizedMap.containsKey(key)) {
synchronizedMap.get(key).add(value);
}
else {
List<String> valuesList = new ArrayList<String>();
valuesList.add(value);
synchronizedMap.put(key, valuesList);
}
}
}
}

根据我的理解，我需要 addToMap()中的同步块来阻止另一个线程在我通过对 put()的调用之前调用 remove()或 containsKey()，但是我不需要 doWork()中的同步块，因为在 remove()返回之前，另一个线程不能进入 addToMap()中的同步块，因为我最初用 Collections.synchronizedMap()创建了 Map。是这样吗？还有更好的办法吗？

111451 次浏览
小开

If you are using JDK 6 then you might want to check out ConcurrentHashMap

Note the putIfAbsent method in that class.

小开

That looks correct to me. If I were to change anything, I would stop using the Collections.synchronizedMap() and synchronize everything the same way, just to make it clearer.

Also, I'd replace

  if (synchronizedMap.containsKey(key)) {
synchronizedMap.get(key).add(value);
}
else {
List<String> valuesList = new ArrayList<String>();
valuesList.add(value);
synchronizedMap.put(key, valuesList);
}

with

List<String> valuesList = synchronziedMap.get(key);
if (valuesList == null)
{
valuesList = new ArrayList<String>();
synchronziedMap.put(key, valuesList);
}
valuesList.add(value);
小开

Check out Google Collections' Multimap, e.g. page 28 of this presentation.

If you can't use that library for some reason, consider using ConcurrentHashMap instead of SynchronizedHashMap; it has a nifty putIfAbsent(K,V) method with which you can atomically add the element list if it's not already there. Also, consider using CopyOnWriteArrayList for the map values if your usage patterns warrant doing so.

小开
最佳答案

Collections.synchronizedMap() guarantees that each atomic operation you want to run on the map will be synchronized.

Running two (or more) operations on the map however, must be synchronized in a block. So yes - you are synchronizing correctly.

小开

There is the potential for a subtle bug in your code.

[UPDATE: Since he's using map.remove() this description isn't totally valid. I missed that fact the first time thru. :( Thanks to the question's author for pointing that out. I'm leaving the rest as is, but changed the lead statement to say there is potentially a bug.]

In doWork() you get the List value from the Map in a thread-safe way. Afterward, however, you are accessing that list in an unsafe matter. For instance, one thread may be using the list in doWork() while another thread invokes synchronizedMap.get(key).add(value) in addToMap(). Those two access are not synchronized. The rule of thumb is that a collection's thread-safe guarantees don't extend to the keys or values they store.

You could fix this by inserting a synchronized list into the map like

List<String> valuesList = new ArrayList<String>();
valuesList.add(value);
synchronizedMap.put(key, Collections.synchronizedList(valuesList)); // sync'd list

Alternatively you could synchronize on the map while you access the list in doWork():

  public void doWork(String key) {
List<String> values = null;
while ((values = synchronizedMap.remove(key)) != null) {
synchronized (synchronizedMap) {
//do something with values
}
}
}

The last option will limit concurrency a bit, but is somewhat clearer IMO.

Also, a quick note about ConcurrentHashMap. This is a really useful class, but is not always an appropriate replacement for synchronized HashMaps. Quoting from its Javadocs,

This class is fully interoperable with Hashtable in programs that rely on its thread safety but not on its synchronization details.

In other words, putIfAbsent() is great for atomic inserts but does not guarantee other parts of the map won't change during that call; it guarantees only atomicity. In your sample program, you are relying on the synchronization details of (a synchronized) HashMap for things other than put()s.

Last thing. :) This great quote from Java Concurrency in Practice always helps me in designing an debugging multi-threaded programs.

For each mutable state variable that may be accessed by more than one thread, all accesses to that variable must be performed with the same lock held.

小开

Yes, you are synchronizing correctly. I will explain this in more detail. You must synchronize two or more method calls on the synchronizedMap object only in a case you have to rely on results of previous method call(s) in the subsequent method call in the sequence of method calls on the synchronizedMap object. Let’s take a look at this code:

synchronized (synchronizedMap) {
if (synchronizedMap.containsKey(key)) {
synchronizedMap.get(key).add(value);
}
else {
List<String> valuesList = new ArrayList<String>();
valuesList.add(value);
synchronizedMap.put(key, valuesList);
}
}

In this code

synchronizedMap.get(key).add(value);

and

synchronizedMap.put(key, valuesList);

method calls are relied on the result of the previous

synchronizedMap.containsKey(key)

method call.

If the sequence of method calls were not synchronized the result might be wrong. For example thread 1 is executing the method addToMap() and thread 2 is executing the method doWork() The sequence of method calls on the synchronizedMap object might be as follows: Thread 1 has executed the method

synchronizedMap.containsKey(key)

and the result is "true". After that operating system has switched execution control to thread 2 and it has executed 

synchronizedMap.remove(key)

After that execution control has been switched back to the thread 1 and it has executed for example

synchronizedMap.get(key).add(value);

believing the synchronizedMap object contains the key and NullPointerException will be thrown because synchronizedMap.get(key) will return null. If the sequence of method calls on the synchronizedMap object is not dependent on the results of each other then you don't need to synchronize the sequence. For example you don't need to synchronize this sequence:

synchronizedMap.put(key1, valuesList1);
synchronizedMap.put(key2, valuesList2);

Here

synchronizedMap.put(key2, valuesList2);

method call does not rely on the results of the previous

synchronizedMap.put(key1, valuesList1);

method call (it does not care if some thread has interfered in between the two method calls and for example has removed the key1).

小开

The way you have synchronized is correct. But there is a catch

  1. Synchronized wrapper provided by Collection framework ensures that the method calls I.e add/get/contains will run mutually exclusive.

However in real world you would generally query the map before putting in the value. Hence you would need to do two operations and hence a synchronized block is needed. So the way you have used it is correct. However.

  1. You could have used a concurrent implementation of Map available in Collection framework. 'ConcurrentHashMap' benefit is

a. It has a API 'putIfAbsent' which would do the same stuff but in a more efficient manner.

b. Its Efficient: dThe CocurrentMap just locks keys hence its not blocking the whole map's world. Where as you have blocked keys as well as values.

c. You could have passed the reference of your map object somewhere else in your codebase where you/other dev in your tean may end up using it incorrectly. I.e he may just all add() or get() without locking on the map's object. Hence his call won't run mutually exclusive to your sync block. But using a concurrent implementation gives you a peace of mind that it can never be used/implemented incorrectly.


提交答案