同步方法中ConcurrentHashMap的并发级别

Question

I'm trying to fix a memory leak issue. Heap dump analysis shows that a ConcurrentHashMap is occupying around 98% of heap memory . Checked the code and it turns out that ConcurrentHashMap instantiation is using a constructor with no parameter. The default configuration for concurrencyLevel is 16. After this map instantiation I see a synchronized method call where data is being put in the map.

I would like to know that since data is being put only in synchronized method, is it safe to set concurrencyLevel of ConcurrentHashMap to 1?

Following is the sample code snippet:

private volatile Map<String, Integer> storeCache;

public void someMethod() {
    storeCache = new ConcurrentHashMap<String, Integer>();
    syncMethod();
}

private synchronized void syncMethod() {
    storeCache.put("Test", 1);
}

Answer 1

I would like to know that since data is being put only in synchronized method, is it safe to set concurrencyLevel of ConcurrentHashMap to 1?

It's certainly safe, in the sense that it's not going to cause any Map corruption. However, it's not going to fix your memory leak. In fact, you probably don't want to synchronize access to the ConcurrentHashMap, which already guarantees safe reads and writes from multiple threads. Synchronizing externally is going to single-thread access to your CHM, which is going to eliminate many of the benefits of the CHM over a HashMap. If you remove the synchronized and specify a concurrencyLevel equal to the estimated number of concurrent writes , you'll probably achieve much better performance.

As for your memory leak, the keys and values in the CHM are strong references, meaning the Java garbage collector won't collect them, even if they're no longer referenced anywhere else in your code. So if you're using the CHM as a cache for temporary values, you'll need to .remove() them when your application no longer needs them.

(If you want the semantics of a ConcurrentMap without the strong keys, you can't get that out-of-the-box, but Guava provides a pretty good alternative .)

You may also want to check that the keys that you're .put() ing into the map have properly implemented .equals() and .hashCode() .