Categories
Java

Multi-threading: Pitfalls when initializing static variables

As a developer you often have to initialize static variables in a multi-threaded environment. The basic solution that most programmers apply is:

1
2
3
4
5
6
7
8
9
private static Object staticVar = null;
 
public static synchronized Object getStaticVar() {
    if (staticVar == null) {
        // initialize
        staticVar = ...
    }
    return staticVar;
}

This is a simple but expensive method. Each thread that needs the variable must synchronize with each other although the variable has long been initialized. So, usually the next step is to synchronize less:

1
2
3
4
5
6
7
8
9
10
11
12
private static Object SYNCHRONIZER = new Object();
private static Object staticVar = null;
 
public static Object getStaticVar() {
    if (staticVar == null) {
        synchronized (SYNCHRONIZER) {
            // initialize
            staticVar = ...
        }
    }
    return staticVar;
}

Yep. That does it, doesn’t it? The answer is: half! Imagine two simultaneous threads entering the method at the same time. If both will see staticVar being null then both will try to enter the synchronized block. And of course, both will initialize the variable nevertheless another thread did it before. So, we add another evaluation to ensure that only one thread will initialize:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
private static Object SYNCHRONIZER = new Object();
private static Object staticVar = null;
 
public static Object getStaticVar() {
    if (staticVar == null) {
        synchronized (SYNCHRONIZER) {
            if (staticVar == null) {
               // initialize
               staticVar = ...
            }
        }
    }
    return staticVar;
}

Most books end here but omit a very crucial part. The code works perfect as long as the initialization is a simple operation only. Let’s make the initialization a bit more sophisticated:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
private static Object SYNCHRONIZER = new Object();
private static List<String> staticVar = null;
 
public static List<String> getStaticVar() {
    if (staticVar == null) {
        synchronized (SYNCHRONIZER) {
            if (staticVar == null) {
               // initialize
               staticVar = new ArrayList<String>();
               staticVar.add("value 1");
               staticVar.add("value 2");
               staticVar.add("value 3");
            }
        }
    }
    return staticVar;
}

The first glance doesn’t reveal anything. But it happened several times in one of my own applications that two threads were not correctly synchronized. Occasionally, one thread found the list to be empty. What happened?

The magic is that staticVar is being set at the very first beginning of the synchronized block. Meanwhile another thread was entering the method and saw the variable not being null. It immediately started using the list although it was not yet initialized completely. The correct solution is therefore:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
private static Object SYNCHRONIZER = new Object();
private static List<String> staticVar = null;
 
public static List<String> getStaticVar() {
    if (staticVar == null) {
        synchronized (SYNCHRONIZER) {
            if (staticVar == null) {
               // initialize
               List<String> tmp = new ArrayList<String>();
               tmp.add("value 1");
               tmp.add("value 2");
               tmp.add("value 3");
               staticVar = tmp;
            }
        }
    }
    return staticVar;
}

For readability, we could refactor the method:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
private static Object SYNCHRONIZER = new Object();
private static List<String> staticVar = null;
 
public static List<String> getStaticVar() {
    if (staticVar == null) {
        synchronized (SYNCHRONIZER) {
            if (staticVar == null) {
               staticVar = createList();
            }
        }
    }
    return staticVar;
}
 
private static List<String> createList() {
   // initialize
   List<String> tmp = new ArrayList<String>();
   tmp.add("value 1");
   tmp.add("value 2");
   tmp.add("value 3");
   return tmp;
}
Categories
Java

Synchronizing Reader and Writer Threads

Here are two functions that you should use when you want two threads, producer and consumer, to be synchronized. I used these functions mainly to ensure that the reader will stop until an object is ready to read. An advantage is that you can control how many objects are in memory at the same time.

 
protected List<Object> availableObjects = new ArrayList<Object>();
 
/**
 * Delivers the next object.
 * Used by the reader/consumer thread.
 */
public synchronized Object next() {
	Object rc;
 
	while (availableObjects.isEmpty()) {
		try {
			wait();
		} catch (InterruptedException e) { }
	}
	rc = availableObjects.remove(0);
 
	notify();
 
	return rc;
}
 
/**
 * Adds a new object to the list of available objects.
 * Used by the writer/producer thread.
 */
public synchronized void addObject(Object o) {
	while (availableObjects.size() >= 20) {
		try {
			wait();
		} catch (InterruptedException e) {}
	}
	availableObjects.add(o);
 
	notify();
}

The main idea was taken from Silberschatz’ book about Operating Systems. You have to make sure that you never call the next() method when the last object was read. So be careful when your number of objects produced is limited.