虽然我在这里提到的答案中看到了许多很好的理论解释，但我在这里添加了一个实际的例子来解释:

1.

代码在不使用volatile的情况下运行

public class VisibilityDemonstration {

private static int sCount = 0;

public static void main(String[] args) {
    new Consumer().start();
    try {
        Thread.sleep(100);
    } catch (InterruptedException e) {
        return;
    }
    new Producer().start();
}

static class Consumer extends Thread {
    @Override
    public void run() {
        int localValue = -1;
        while (true) {
            if (localValue != sCount) {
                System.out.println("Consumer: detected count change " + sCount);
                localValue = sCount;
            }
            if (sCount >= 5) {
                break;
            }
        }
        System.out.println("Consumer: terminating");
    }
}

static class Producer extends Thread {
    @Override
    public void run() {
        while (sCount < 5) {
            int localValue = sCount;
            localValue++;
            System.out.println("Producer: incrementing count to " + localValue);
            sCount = localValue;
            try {
                Thread.sleep(1000);
            } catch (InterruptedException e) {
                return;
            }
        }
        System.out.println("Producer: terminating");
    }
}
}

在上面的代码中，有两个线程——生产者和消费者。

生产者线程在循环中迭代5次(睡眠时间为1000毫秒或1秒)。在每次迭代中，生产者线程将sCount变量的值增加1。因此，在所有迭代中，生产者将sCount的值从0更改为5

使用者线程处于一个常量循环中，每当sCount的值发生变化时，它就会打印，直到值达到5为止。

两个循环同时开始。因此，生产者和消费者都应该将sCount的值打印5次。

输出

Consumer: detected count change 0
Producer: incrementing count to 1
Producer: incrementing count to 2
Producer: incrementing count to 3
Producer: incrementing count to 4
Producer: incrementing count to 5
Producer: terminating

分析

In the above program, when the producer thread updates the value of sCount, it does update the value of the variable in the main memory(memory from where every thread is going to initially read the value of variable). But the consumer thread reads the value of sCount only the first time from this main memory and then caches the value of that variable inside its own memory. So, even if the value of original sCount in main memory has been updated by the producer thread, the consumer thread is reading from its cached value which is not updated. This is called VISIBILITY PROBLEM .

2.

代码使用volatile运行

在上面的代码中，用下面的代码替换声明了sCount的代码行:

private volatile  static int sCount = 0;

输出

Consumer: detected count change 0
Producer: incrementing count to 1
Consumer: detected count change 1
Producer: incrementing count to 2
Consumer: detected count change 2
Producer: incrementing count to 3
Consumer: detected count change 3
Producer: incrementing count to 4
Consumer: detected count change 4
Producer: incrementing count to 5
Consumer: detected count change 5
Consumer: terminating
Producer: terminating

分析

当我们声明一个变量为volatile时，这意味着所有对这个变量的读写操作都将直接进入主存。这些变量的值永远不会被缓存。

由于sCount变量的值永远不会被任何线程缓存，消费者总是从主内存中读取sCount的原始值(在那里由生产者线程更新)。因此，在这种情况下，输出是正确的，两个线程都打印了5次不同的sCount值。

通过这种方式，volatile关键字解决了可见性问题。

挥发性(vɒlətʌɪl):在常温下容易挥发

关于volatile的重要一点:

Synchronization in Java is possible by using Java keywords synchronized and volatile and locks. In Java, we can not have synchronized variable. Using synchronized keyword with a variable is illegal and will result in compilation error. Instead of using the synchronized variable in Java, you can use the java volatile variable, which will instruct JVM threads to read the value of volatile variable from main memory and don’t cache it locally. If a variable is not shared between multiple threads then there is no need to use the volatile keyword.

源

volatile用法示例:

public class Singleton {
    private static volatile Singleton _instance; // volatile variable
    public static Singleton getInstance() {
        if (_instance == null) {
            synchronized (Singleton.class) {
                if (_instance == null)
                    _instance = new Singleton();
            }
        }
        return _instance;
    }
}

我们在第一个请求到来时惰性地创建实例。

如果我们不使_instance变量为volatile，那么创建Singleton实例的线程就不能与其他线程通信。因此，如果线程A正在创建单例实例，在创建后，CPU损坏等，所有其他线程将无法看到_instance的值不为空，他们将认为它仍然被分配为空。

为什么会发生这种情况?因为读线程不做任何锁，直到写线程从同步块中出来，内存不会被同步，_instance的值也不会在主存中更新。使用Java中的Volatile关键字，这是由Java本身处理的，这样的更新将对所有读取线程可见。

结论:volatile关键字也用于线程之间的内存内容通信。

without volatile的用法示例:

public class Singleton {    
    private static Singleton _instance;   //without volatile variable
    public static Singleton getInstance() {   
        if (_instance == null) {  
            synchronized(Singleton.class) {  
                if (_instance == null) 
                    _instance = new Singleton(); 
            } 
        }
        return _instance;  
    }
}

The code above is not thread-safe. Although it checks the value of instance once again within the synchronized block (for performance reasons), the JIT compiler can rearrange the bytecode in a way that the reference to the instance is set before the constructor has finished its execution. This means the method getInstance() returns an object that may not have been initialized completely. To make the code thread-safe, the keyword volatile can be used since Java 5 for the instance variable. Variables that are marked as volatile get only visible to other threads once the constructor of the object has finished its execution completely. Source

Java中不稳定的用法:

快速失败迭代器通常使用list对象上的volatile计数器实现。

当列表更新时，计数器会递增。 创建Iterator时，计数器的当前值嵌入到Iterator对象中。 当执行Iterator操作时，该方法比较两个计数器值，如果不相同则抛出ConcurrentModificationException异常。

故障安全迭代器的实现通常是轻量级的。它们通常依赖于特定列表实现的数据结构的属性。没有一般的模式。

在我看来，除了停止线程之外，使用volatile关键字的两个重要场景是:

Double-checked locking mechanism. Used often in Singleton design pattern. In this the singleton object needs to be declared volatile. Spurious Wakeups. Thread may sometimes wake up from wait call even if no notify call has been issued. This behavior is called spurious wakeup. This can be countered by using a conditional variable (boolean flag). Put the wait() call in a while loop as long as the flag is true. So if thread wakes up from wait call due to any reasons other than Notify/NotifyAll then it encounters flag is still true and hence calls wait again. Prior to calling notify set this flag to true. In this case the boolean flag is declared as volatile.

使用volatile的一个常见示例是使用volatile布尔变量作为终止线程的标志。如果您已经启动了一个线程，并且希望能够安全地从另一个线程中断它，您可以让线程定期检查标志。要阻止它，将标志设置为true。通过将标志设置为易失性，可以确保正在检查它的线程在下次检查时看到它已经设置好，甚至不必使用同步块。

每个访问volatile字段的线程在继续之前都会读取它的当前值，而不是(可能)使用缓存的值。

只有成员变量可以是易变的或瞬变的。

从oracle文档页，需要volatile变量来修复内存一致性问题:

使用易失性变量可以降低内存一致性错误的风险，因为对易失性变量的任何写入都会与该变量的后续读取建立happens-before关系。

这意味着对volatile变量的更改总是对其他线程可见。这也意味着当线程读取一个volatile变量时，它不仅看到了volatile的最新更改，还看到了导致更改的代码的副作用。

正如在Peter Parker的回答中解释的那样，在没有volatile修饰符的情况下，每个线程的堆栈都可能有自己的变量副本。通过将变量设置为volatile，可以修复内存一致性问题。

为了更好地理解，请查看jenkov教程页面。

看一下相关的SE问题，了解更多关于volatile的细节和使用volatile的用例:

Java中volatile和synchronized的区别

一个实际的用例:

你有很多线程，它们需要以特定的格式打印当前时间，例如:java.text.SimpleDateFormat("HH-mm-ss")。可以有一个类，它将当前时间转换为SimpleDateFormat，并每一秒更新一次变量。所有其他线程都可以使用这个易失性变量在日志文件中打印当前时间。