当你加载一个配置属性对象时，无论是从数据库还是从文件中，将它作为一个单例都会有所帮助;没有理由在服务器运行时重新读取不会改变的静态数据。

当管理对整个应用程序共享的资源的访问时，应该使用单例，并且可能存在同一个类的多个实例将是破坏性的。确保对共享资源的访问线程安全是这种模式至关重要的一个很好的例子。

当使用单例时，你应该确保你不会意外地隐藏依赖关系。理想情况下，单例对象(就像应用程序中的大多数静态变量一样)在应用程序的初始化代码执行期间设置(c#可执行文件为静态void Main()， java可执行文件为静态void Main())，然后传递给所有其他需要它的实例化类。这有助于维护可测试性。

Singleton类的主要目的是限制创建的实例数量，从而确保对资源的访问控制。

使用单例类不会产生内存空间浪费，因为它限制了实例的创建。因为对象的创建只会发生一次，而不是每次新请求时都创建。

单例类在内部使用互斥量，因此使其线程安全。这就是为什么多线程和数据库应用程序主要使用Java中的Singleton模式进行缓存、日志记录、线程池、配置设置等等

Singleton pattern is the most pervasive pattern in the Spring containerization approach. If we look at that in terms of architectural primitives - they form a blackboard graph of objects, to which every thread can read and write. They do the dramatic act of synchronizing between multiple threads. The very reason why multiple threads need to synchronize is because there are always resources that underlie a computational program, over which contention might occur. Consider what is called a 'last seat problem'. A flight is being booked, but there are multiple ways to do it. For simplicity lets assume that the data about the flight occupancy is stored in a flat file rather than a database. Now, if there are two threads, each functionally different (i.e represented by different endpoints in the webapp) and let one of these threads A, be the thread which a prospective passenger uses to make a booking and the other one B is which a flight manager uses to close the booking - virtually closing the boarding door. Then, if these threads do not use singleton, the flight object would be detached from the real resource out-there, which we say not the actual aeroplane but the entry in the flat file. The A thread would have reference to an object, while the passenger is still fighting a dilemma whether to fly or not and then finally when he makes up his mind, the B thread would already have closed the door. But the object referenced by the A thread would still show one more seat to go. Now, cutting out the RDBMS due to our initial assumption, the system would write a ticket for the passenger and issue it to him eventhough the boarding is closed. Now, in a singleton implementation, the moment the theread B accesses the system, the universal object Flight is updated with status closed. Hence, if the passenger finally makes up his mind and clicks confirm, he would get an error right away. All this would not have been possible without the singleton. Hence, singleton allows you to stay close to the resources and avoids thread contention.

首先，让我们区分一下单对象和单对象。后者是前者的许多可能实现之一。而且单对象的问题与单例的问题是不同的。单对象本身并不坏，有时是做事情的唯一方法。简而言之:

单对象-我只需要程序中对象的一个实例 Singleton -创建一个带有静态字段的类。添加一个返回此字段的静态方法。在第一次调用时惰性地实例化一个字段。总是返回相同的对象。

public class Singleton {
    private static Singleton instance;

    private Singleton() {}

    public static Singleton instance() {
        if (instance == null) {
            instance = new Singleton();
        }
        return instance;
    }
}

如您所见，规范形式的“单例”模式对测试不是很友好。不过，这个问题很容易解决:只要让Singleton实现一个接口。让我们称它为“可测试单例”:)

public class Singleton implements ISingleton {
    private static Singleton instance;

    private Singleton() {}

    public static ISingleton instance() {
        if (instance == null) {
            instance = new Singleton();
        }
        return instance;
    }
}

现在我们可以模拟Singleton了，因为我们通过接口来使用它。其中一项索赔消失了。让我们看看是否可以去掉另一个声明-共享全局状态。

如果我们剥离单例模式，它的核心是惰性初始化:

public static ISingleton instance() {
    if (instance == null) {
        instance = new Singleton();
    }
    return instance;
}

这就是它存在的全部原因。这就是单对象模式。我们把它拿走，放到工厂方法中，例如:

public class SingletonFactory {
    private static ISingleton instance;

    // Knock-knock. Single Object here
    public static ISingleton simpleSingleton() {
        if (instance == null) {
            instance = new Singleton();
        }
        return instance;
    }
}

与我们的可测试单例有什么不同?没有，因为这是单对象模式的本质——不管你是把它实现为单例、工厂方法还是服务定位器。你仍然有一些共享的全局状态。如果从多个线程访问它，这可能会成为一个问题。您将不得不使simpleSingleton()同步并处理所有多线程问题。

再说一遍:无论你选择什么方法，你都必须付出单一对象的代价。使用依赖注入容器只是将复杂性转移到框架，它将不得不处理单对象的固有问题。

回顾:

Most of people who mention Singleton mean Single Object One of the popular ways to implement it is the Singleton pattern It has its flaws that can be mitigated However, the most of Singleton's complexity roots in Single Object's complexity Regardless of how you instantiate your Single Object, it's still there, be it a Service Locator, a Factory Method or something else You can shift the complexity to a DI container which is (hopefully) well-tested Sometimes using the DI container is cumbersome - imagine injecting a LOGGER to every class

一个实例可以在Test::Builder中找到，这个类支持几乎所有现代Perl测试模块。Builder单例存储并代理测试流程的状态和历史记录(历史测试结果，计算运行的测试次数)以及测试输出流向等内容。这些都是协调由不同作者编写的多个测试模块以在单个测试脚本中一起工作所必需的。

The history of Test::Builder's singleton is educational. Calling new() always gives you the same object. First, all the data was stored as class variables with nothing in the object itself. This worked until I wanted to test Test::Builder with itself. Then I needed two Test::Builder objects, one setup as a dummy, to capture and test its behavior and output, and one to be the real test object. At that point Test::Builder was refactored into a real object. The singleton object was stored as class data, and new() would always return it. create() was added to make a fresh object and enable testing.

Currently, users are wanting to change some behaviors of Test::Builder in their own module, but leave others alone, while the test history remains in common across all testing modules. What's happening now is the monolithic Test::Builder object is being broken down into smaller pieces (history, output, format...) with a Test::Builder instance collecting them together. Now Test::Builder no longer has to be a singleton. Its components, like history, can be. This pushes the inflexible necessity of a singleton down a level. It gives more flexibility to the user to mix-and-match pieces. The smaller singleton objects can now just store data, with their containing objects deciding how to use it. It even allows a non-Test::Builder class to play along by using the Test::Builder history and output singletons.

似乎在数据的协调和行为的灵活性之间存在着一种推拉关系，这种关系可以通过在共享数据周围使用尽可能少的行为来缓解，以确保数据的完整性。