如何正确实现多线程安全的singleton patterns
Singleton Pattern
Short Introduction
Singleton pattern, described in the GOF Design Patterns book, is one of the most easily understandable and on of the most frequently used pattern. The goal of the singleton pattern is to make sure that there is only one instance of this class in the system and allow other classes access this instance.
Case study
Today I tried to extend a exist MVC system in our project. Because only one instance of the Controller is needed in the system (typical Singleton pattern), a traditional singleton pattern was implemented here. The code is looked like:
public class Controller {
private static Controller instance;
private Controller()
{
// do something
}
public final static Controller getInstance(){
public final static Controller getInstance(){
(A)
if (instance == null) {
(B)
instance = new Controller();
}
return instance;
}
// some more methods
}
The above shown code is, just like it described in almost all Design patterns books, is beautiful and correct based on the traditional implantation of singleton pattern.
Just at this time, our Team leader came to me, saw the implementation and told me at the first second that the traditional implementation is NOT thread safe! Yes! It is thread unsafe, after I read the code once again. Why? Let us make try: two threads T1 and T2 try to call the class Controller. When T1 goes to the position (B) in the above shown code, it sleeps. Then comes T2 to the position (A) and checks whether an instance of the Controller exists. Because T1 sleeps at the position (B) before an instance will be created, T2 can go into the block and create a new instance of Controller. Now T1 is awake, and what will it do? Just create another instance of Controller, because it is already in the block! So, great! TWO instance are created! It is no more singleton!
The first simple idea that I got in the first second, as you estimated, is adding the key word synchronized before the method getInstance(). But this brings bad Performance. As we knew, the normally used solution for such a problem is to moving the synchronizing into the method. That means to building a synchronized block in the method. The code should be looked like this:
public class Controller {
private static Controller instance;
private Controller()
{
// do something
}
public final static Controller getInstance(){
(A)
if (instance == null) {
(B)
synchronized {
instance = new Controller();
}
}
return instance;
}
// some more methods
}
But This code is still thread unsafe! Why? Because it is only synchronized when an instance is created, the above described problem DOES still exist. T1 sleeps at (B), T2 goes through and create a new instance, T1 wakes up and create another one! If we move the synchronized block up to contain the whole if block, there is nothing difference with a synchronized method. So, synchronized is a bad idea.
We must find another way.
The Current most used thread safe solution for this problem is to creating an instance of the class as class’s private property when the class is loaded. The instance can be accessed by calling the public getInstance() method. The correct code should be looked like:
public class Controller {
private static Controller instance = new Controller();
private Controller()
{
// do something
}
public final static Controller getInstance(){
return instance;
}
// some more methods
}
One drawback of this solution is that an instance is created anyway, even if the instance will be never used.