写这个系列的文章,只为把所学的设计模式再系统的整理一遍。错误和不周到的地方欢迎大家批评。点击这里下载源代码。
什么时候使用单例模式
在程序运行时,某种类型只需要一个实例时,一般采用单例模式。为什么需要一个实例?第一,性能,第二,保持代码简洁,比如程序中通过某个配置类A读取配置文件,如果在每处使用的地方都new A(),才能读取配置项,一个是浪费系统资源(参考.NET垃圾回收机制),再者重复代码太多。
单例模式的实现
实现单例模式,方法非常多,这里我把见过的方式都过一遍,来体会如何在支持并发访问、性能、代码简洁程度等方面不断追求极致。(单击此处下载代码)
实现1:非线程安全
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Threading.Tasks;
7:
8: namespace SingletonPatternNotTheadSafe
9: {
10: public sealed class Singleton
11: {
12: private static Singleton instance = null;
13:
14: private Singleton()
15: {
16: }
17:
18: public static Singleton Instance
19: {
20: get
21: {
22: if (instance == null)
23: {
24: Thread.Sleep(1000);
25: instance = new Singleton();
26: Console.WriteLine(string.Format(
27: "[{0}]创建Singleton {1}" , Thread.CurrentThread.ManagedThreadId, instance.GetHashCode()));
28: }
29:
30: Console.WriteLine(string.Format(
31: "[{0}]获得Singleton {1}", Thread.CurrentThread.ManagedThreadId, instance.GetHashCode()));
32: return instance;
33: }
34: }
35: }
36: }
为了能够在下面的测试代码中展示上面代码的问题,这里在创建对象前,让线程休息1秒,并且在控制台打印出当前线程ID、对象的hashcode(一般不同对象的hashcode是不一样的,但可能重复)。
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Threading.Tasks;
7:
8: namespace SingletonPatternNotTheadSafe
9: {
10: class Program
11: {
12: private static void Main(string[] args)
13: {
14: Thread t1 = new Thread(new ThreadStart(Compute));
15:
16: t1.Start();
17:
18: Compute();
19:
20: Console.ReadLine(); // 阻止主线程结束
21: }
22:
23: private static void Compute()
24: {
25: Singleton o1 = Singleton.Instance;
26: }
27: }
28: }
执行结果如下:
分析:
Singleton.Instance的get方法中创建instance并未考虑并发访问的情况,导致可能重复创建Singleton对象。下面的实现方法修复了此问题。
实现2:简单线程安全
要解决上面的问题,最简单的方法就是在创建对象的时候加锁。
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Threading.Tasks;
7:
8: namespace SingletonSimpleThreadSafe
9: {
10: public sealed class Singleton
11: {
12: private static Singleton instance = null;
13: private static readonly object _lock = new object();
14:
15: private Singleton()
16: {
17: }
18:
19: public static Singleton Instance
20: {
21: get
22: {
23: lock (_lock)
24: {
25: if (instance == null)
26: {
27: Thread.Sleep(1000);
28: instance = new Singleton();
29: Console.WriteLine(string.Format(
30: "[{0}]创建Singleton {1}", Thread.CurrentThread.ManagedThreadId, instance.GetHashCode()));
31: }
32: }
33:
34: Console.WriteLine(string.Format(
35: "[{0}]获得Singleton {1}", Thread.CurrentThread.ManagedThreadId, instance.GetHashCode()));
36: return instance;
37: }
38: }
39: }
40: }
测试代码如下:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Diagnostics;
7: using System.Threading.Tasks;
8:
9: namespace SingletonSimpleThreadSafe
10: {
11: class Program
12: {
13: private static void Main(string[] args)
14: {
15: SingletonTest();
16: }
17:
18: private static void SingletonTest()
19: {
20: Thread t1 = new Thread(new ThreadStart(Compute));
21:
22: t1.Start();
23:
24: Compute();
25:
26: Console.ReadLine(); // 阻止主线程结束
27: }
28:
29: private static void Compute()
30: {
31: Singleton o1 = Singleton.Instance;
32: }
33: }
34: }
我们再看看执行效果:
创建Singleton只执行一次。但是这种写法性能并不高,每次通过Singleton.Instance获得实例对象都需要判断锁是否别别的线程占用。
这里我们修改一下Singleton,把代码中的Thread.Sleep和Console.Writeline都去掉,这里我重新创建了一个Singleton2 class,两个线程中循环调用100000000次,看一下这么实现的性能:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Threading.Tasks;
7:
8: namespace SingletonSimpleThreadSafe
9: {
10: public sealed class Singleton2
11: {
12: private static Singleton2 instance = null;
13: private static readonly object _lock = new object();
14:
15: private Singleton2()
16: {
17: }
18:
19: public static Singleton2 Instance
20: {
21: get
22: {
23: lock (_lock)
24: {
25: if (instance == null)
26: {
27: instance = new Singleton2();
28: }
29: }
30:
31: return instance;
32: }
33: }
34: }
35: }
测试代码如下:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Diagnostics;
7: using System.Threading.Tasks;
8:
9: namespace SingletonSimpleThreadSafe
10: {
11: class Program
12: {
13: private static void Main(string[] args)
14: {
15: Singleton2Test();
16: }
17:
18: private static void Singleton2Test()
19: {
20: Thread t1 = new Thread(new ThreadStart(Compute2));
21:
22: t1.Start();
23:
24: Compute2();
25:
26: Console.ReadLine(); // 阻止主线程结束
27: }
28:
29: private static void Compute2()
30: {
31: Stopwatch sw1 = new Stopwatch();
32:
33: sw1.Start();
34:
35: for (int i = 0; i < 100000000; i++)
36: {
37: Singleton2 instance = Singleton2.Instance;
38: }
39:
40: sw1.Stop();
41:
42: Console.WriteLine(string.Format("[{0}]耗时:{1}毫秒",
43: Thread.CurrentThread.ManagedThreadId,
44: sw1.ElapsedMilliseconds));
45: }
46: }
47: }
执行结果:
我们先不讨论结果,接着往下看看双检锁方式的性能。
实现3:双检锁实现的线程安全
Singleton双检锁实现:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading;
6: using System.Threading.Tasks;
7:
8: namespace SingletonDoubleCheckThreadSafe
9: {
10: public sealed class Singleton2
11: {
12: private static Singleton2 instance = null;
13: private static readonly object _lock = new object();
14:
15: private Singleton2()
16: {
17: }
18:
19: public static Singleton2 Instance
20: {
21: get
22: {
23: if (instance == null)
24: {
25: lock (_lock)
26: {
27: if (instance == null)
28: {
29: instance = new Singleton2();
30: }
31: }
32: }
33:
34: return instance;
35: }
36: }
37: }
38: }
测试代码和上面的一样,结果如下:
性能提高了(7571+7465-1410-1412)/ (7571+7465) * 100% = 81.2%。(实际项目中为了减少误差,应该跑多遍测试得到多个结果的平均值和方差,这里为了方便,我只把一次测试结果贴出来。)
双检锁机制在lock外又检查了一次instance是否为null,这样在第一次访问使instance创建后,后面的调用都无需检查lock是否被占用。
一名程序员要了解到这里算基本合格,如果想达到更高的水平,继续往下看。这种方式有什么缺点呢?
- 上面的代码在Java中不能正常工作。这是因为Java的Memory Model实现和.NET不一样,并不保证一定在构造函数执行完成后才返回对象的引用。虽然Java 1.5版本重构了Memory Model,但是双检锁机制在不给instance field加volatile关键字时,依然不能正常工作。
- Microsoft的.net memory model并不是按照标准的ECMA CLI规范实现,而是在标准上做了一些“增强”工作。MS .net CLR memory model中所有的写操作都是VolatileWrite(参考《CLR via C#》第二版的第24章)。所以我们的代码中不加volatile也能在IA64CPU 架构的机器上正常执行。但是如Jeffrey建议,最好还是遵循ECMA标准。
- 实现复杂。
实现4:非懒加载,无锁实现线程安全
.NET中的static变量在class被第一次实例化的时候创建,且保证仅执行一次创建。利用这个特点,可以像如下实现:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading.Tasks;
6:
7: namespace SingletonNotUsingLock
8: {
9: public class Singleton
10: {
11: private volatile static Singleton instance = new Singleton();
12:
13: // Explicit static constructor to tell C# compiler
14: // not to mark type as beforefieldinit
15: static Singleton()
16: {
17: Console.WriteLine("execute static constructor");
18: }
19:
20: private Singleton()
21: {
22: Console.WriteLine("execute private constructor");
23: }
24:
25: public static Singleton Instance
26: {
27: get
28: {
29: Console.WriteLine("instance get");
30: return instance;
31: }
32: }
33: }
34: }
上面的代码可以更简化一些,去掉Instance属性,将私有的instance变量改成public的:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading.Tasks;
6:
7: namespace SingletonNotUsingLock
8: {
9: public class Singleton2
10: {
11: public volatile static Singleton2 instance = new Singleton2();
12:
13: // Explicit static constructor to tell C# compiler
14: // not to mark type as beforefieldinit
15: static Singleton2()
16: {
17: Console.WriteLine("execute static constructor");
18: }
19:
20: private Singleton2()
21: {
22: Console.WriteLine("execute private constructor");
23: }
24: }
25: }
代码非常简洁。但是为什么有个静态构造函数呢,我们看看下面的测试代码:
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading.Tasks;
6:
7: namespace SingletonNotUsingLock
8: {
9: class Program
10: {
11: static void Main(string[] args)
12: {
13: Console.WriteLine("begin create singleton");
14:
15: Singleton s1 = Singleton.Instance;
16:
17: Console.WriteLine("after create singleton");
18:
19: Singleton2 s2 = Singleton2.instance;
20:
21: Console.WriteLine("after create singleton2");
22: }
23: }
24: }
执行结果如下:
把静态构造函数去掉后执行结果如下:
这是因为没有静态构造函数的类,编译时会被标记称beforefieldinit,那么,beforefieldinit究竟表示什么样的语义呢?Scott Allen对此进行了详细的解释:beforefieldinit为CLR提供了在任何时候执行.cctor的授权,只要该方法在第一次访问类型的静态字段之前执行即可。
实现5:无锁懒加载
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading.Tasks;
6:
7: namespace SingletonNotUsingLockAndLazyLoad
8: {
9: public class Singleton
10: {
11: private Singleton()
12: {
13: Console.WriteLine("execute Singleton private constructor");
14: }
15:
16: public static Singleton Instance
17: {
18:
19: get
20: {
21: Console.WriteLine("execute Singleton.Instance get");
22: return Nested.instance;
23: }
24: }
25:
26: private class Nested
27: {
28: // Explicit static constructor to tell C# compiler
29: // not to mark type as beforefieldinit
30: static Nested()
31: {
32: Console.WriteLine("execute Nested static constructor");
33: }
34:
35: internal static readonly Singleton instance = new Singleton();
36: }
37: }
38: }
实现6:使用.NET 4.0中的Lazy<T>
1: using System;
2: using System.Collections.Generic;
3: using System.Linq;
4: using System.Text;
5: using System.Threading.Tasks;
6:
7: namespace SingletonUsingLazyType
8: {
9: public sealed class Singleton
10: {
11: private static readonly Lazy<Singleton> lazy =
12: new Lazy<Singleton>(() => new Singleton());
13:
14: public static Singleton Instance { get { return lazy.Value; } }
15:
16: private Singleton()
17: {
18: }
19: }
20: }