在Java中,它的内存管理包括两方面:内存分配(创建Java对象的时候)和内存回收,这两方面工作都是由JVM自动完成的,降低了Java程序员的学习难度,避免了像C/C++直接操作内存的危险。但是,也正因为内存管理完全由JVM负责,所以也使Java很多程序员不再关心内存分配,导致很多程序低效,耗内存。因此就有了Java程序员到最后应该去了解JVM,才能写出更高效,充分利用有限的内存的程序。

1.Java在内存中的状态

首先我们先写一个代码为例子:
Person.java

package test;
import java.io.Serializable;
public class Person implements Serializable {
    static final long serialVersionUID = 1L;
    String name; // 姓名
    Person friend;    //朋友
    public Person() {}
    public Person(String name.html) {
        super();
        this.name = name;
    }
}

Test.java

package test;

public class Test{
    public static void main(String[] args.html) {
        Person p1 = new Person("Kevin".html);
        Person p2 = new Person("Rain".html);
        Person p3 = new Person("Sunny".html);
        p1.friend = p2;
        p3 = p2;
        p2 = null;
    }
}

把上面Test.java中main方面里面的对象引用画成一个从main方法开始的对象引用图的话就是这样的(顶点是对象和引用,有向边是引用关系):

当程序运行起来之后,把它在内存中的状态看成是有向图后,可以分为三种:
1)可达状态:在一个对象创建后,有一个以上的引用变量引用它。在有向图中可以从起始顶点导航到该对象,那它就处于可达状态。
2)可恢复状态:如果程序中某个对象不再有任何的引用变量引用它,它将先进入可恢复状态,此时从有向图的起始顶点不能再导航到该对象。在这个状态下,系统的垃圾回收机制准备回收该对象的所占用的内存,在回收之前,系统会调用finalize()方法进行资源清理,如果资源整理后重新让一个以上引用变量引用该对象,则这个对象会再次变为可达状态;否则就会进入不可达状态。
3)不可达状态:当对象的所有关联都被切断,且系统调用finalize()方法进行资源清理后依旧没有使该对象变为可达状态,则这个对象将永久性失去引用并且变成不可达状态,系统才会真正的去回收该对象所占用的资源。
上述三种状态的转换图如下:

2.Java对对象的4种引用

1)强引用 :创建一个对象并把这个对象直接赋给一个变量,eg :Person person = new Person("sunny".html); 不管系统资源有么的紧张,强引用的对象都绝对不会被回收,即使他以后不会再用到
2)软引用 :通过SoftReference类实现,eg : SoftReference p = new SoftReference(new Person("Rain".html));,内存非常紧张的时候会被回收,其他时候不会被回收,所以在使用之前要判断是否为null从而判断他是否已经被回收了。
3)弱引用 :通过WeakReference类实现,eg : WeakReference p = new WeakReference(new Person("Rain".html));不管内存是否足够,系统垃圾回收时必定会回收。
4)虚引用 :不能单独使用,主要是用于追踪对象被垃圾回收的状态。通过PhantomReference类和引用队列ReferenceQueue类联合使用实现,eg :

package test;
import java.lang.ref.PhantomReference;
import java.lang.ref.ReferenceQueue;

public class Test{
    public static void main(String[] args.html) {
        //创建一个对象
        Person person = new Person("Sunny".html);
        //创建一个引用队列
        ReferenceQueue<Person> rq = new ReferenceQueue<Person>();
        //创建一个虚引用,让此虚引用引用到person对象
        PhantomReference<Person> pr = new PhantomReference<Person>(person, rq.html);
        //切断person引用变量和对象的引用
        person = null;
        //试图取出虚引用所引用的对象
        //发现程序并不能通过虚引用访问被引用对象,所以此处输出为null
        System.out.println(pr.get(.html));
        //强制垃圾回收
        System.gc();
        System.runFinalization();
        //因为一旦虚引用中的对象被回收后,该虚引用就会进入引用队列中
        //所以用队列中最先进入队列中引用与pr进行比较,输出true
        System.out.println(rq.poll(.html) == pr);
    }
}

运行结果:

3.Java垃圾回收机制

其实Java垃圾回收主要做的是两件事:1)内存回收 2)碎片整理

3.1垃圾回收算法

1)串行回收(只用一个CPU)和并行回收(多个CPU才有用):串行回收是不管系统有多少个CPU,始终只用一个CPU来执行垃圾回收操作,而并行回收就是把整个回收工作拆分成多个部分,每个部分由一个CPU负责,从而让多个CPU并行回收。并行回收的执行效率很高,但复杂度增加,另外也有一些副作用,如内存碎片增加。
2)并发执行和应用程序停止 :应用程序停止(Stop-the-world)顾名思义,其垃圾回收方式在执行垃圾回收的同时会导致应用程序的暂停。并发执行的垃圾回收虽然不会导致应用程序的暂停,但由于并发执行垃圾需要解决和应用程序的执行冲突(应用程序可能在垃圾回收的过程修改对象),因此并发执行垃圾回收的系统开销比Stop-the-world高,而且执行时需要更多的堆内存。
3)压缩和不压缩和复制 :
①支持压缩的垃圾回收器(标记-压缩 = 标记清除+压缩)会把所有的可达对象搬迁到一端,然后直接清理掉端边界以外的内存,减少了内存碎片。
②不压缩的垃圾回收器(标记-清除)要遍历两次,第一次先从跟开始访问所有可达对象,并将他们标记为可达状态,第二次便利整个内存区域,对未标记可达状态的对象进行回收处理。这种回收方式不压缩,不需要额外内存,但要两次遍历,会产生碎片
复制式的垃圾回收器:将堆内存分成两个相同空间,从根(类似于前面的有向图起始顶点)开始访问每一个关联的可达对象,将空间A的全部可达对象复制到空间B,然后一次性回收空间A。对于该算法而言,因为只需访问所有的可达对象,将所有的可达对象复制走之后就直接回收整个空间,完全不用理会不可达对象,所以遍历空间的成本较小,但需要巨大的复制成本和较多的内存。

       ![](data:image/png;base64,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.html)

3.2堆内存的分代回收

1)分代回收的依据:

①对象生存时间的长短: 大部分对象在Young期间就被回收
②不同代采取不同的垃圾回收策略: 新(生存时间短)老(生存时间长)对象之间很少存在引用

2) 堆内存的分代:

①Young代 :
Ⅰ回收机制 :因为对象数量少,所以采用复制回收。
Ⅱ组成区域 :由1个Eden区和2个Survivor区构成,同一时间的两个Survivor区,一个用来保存对象,另一个是空的;每次进行Young代垃圾回收的时候,就把Eden,From中的可达对象复制到To区域中,一些生存时间长的就复制到了老年代,接着清除Eden,From空间,最后原来的To空间变为From空间,原来的From空间变为To空间。
Ⅲ对象来源 :绝大多数对象先分配到Eden区,一些大的对象会直接被分配到Old代中。
Ⅳ回收频率 :因为Young代对象大部分很快进入不可达状态,因此回收频率高且回收速度快

②Old代 :
Ⅰ回收机制 :采用标记压缩算法回收。
Ⅱ对象来源 :1.对象大直接进入老年代。
       2.Young代中生存时间长的可达对象
Ⅲ回收频率 :因为很少对象会死掉,所以执行频率不高,而且需要较长时间来完成。
③Permanent代 :
Ⅰ用 途 :用来装载Class,方法等信息,默认为64M,不会被回收
Ⅱ对象来源 :eg:对于像Hibernate,Spring这类喜欢AOP动态生成类的框架,往往会生成大量的动态代理类,因此需要更多的Permanent代内存。所以我们经常在调试Hibernate,Spring的时候经常遇到java.lang.OutOfMemoryError:PermGen space的错误,这就是Permanent代内存耗尽所导致的错误。
Ⅲ回收频率 :不会被回收

3.3常见的垃圾回收器

在此之前,我们先讲一下下面将会涉及到的并发和并行两个词的解释:
1)并行:指多条垃圾收集线程并行工作,但此时用户线程仍然处于等待状态;
2)并发:指用户线程与 垃圾收集线程同时执行(但不一定是并行的,可能会交替执行),用户程序继续执行,而垃圾收集程序运行于另一个CPU上。
好啦,继续讲垃圾回收器:
1)串行回收器(只使用一个CPU):Young代采用串行复制算法;Old代使用串行标记压缩算法(三个阶段:标记mark—清除sweep—压缩compact),回收期间程序会产生暂停,
2)并行回收器:对Young代采用的算法和串行回收器一样,只是增加了多CPU并行处理; 对Old代的处理和串行回收器完全一样,依旧是单线程。
3)并行压缩回收器:对Young代处理采用与并行回收器完全一样的算法;只是对Old代采用了不同的算法,其实就是划分不同的区域,然后进行标记压缩算法:
① 将Old代划分成几个固定区域;
② mark阶段(多线程并行),标记可达对象;
③ summary阶段(串行执行),从最左边开始检验知道找到某个达到数值(可达对象密度小)的区域时,此区域及其右边区域进行压缩回收,其左端为密集区域
④ compact阶段(多线程并行),识别出需要装填的区域,多线程并行的把数据复制到这些区域中。经此过程后,Old代一端密集存在大量活动对象,另一端则存在大块空间。
4)并发标识—清理回收(CMS):对Young代处理采用与并行回收器完全一样的算法;只是对Old代采用了不同的算法,但归根待地还是标记清理算法:
① 初始标识(程序暂停):标记被直接引用的对象(一级对象.html);
② 并发标识(程序运行):通过一级对象寻找其他可达对象;
③ 再标记(程序暂停):多线程并行的重新标记之前可能因为并发而漏掉的对象(简单的说就是防遗漏)
④ 并发清理(程序运行)

4.内存管理小技巧

1)尽量使用直接量,eg:String javaStr = "小学徒的成长历程";
2)使用StringBuilder和StringBuffer进行字符串连接等操作;
3)尽早释放无用对象;
4)尽量少使用静态变量;
5)缓存常用的对象:可以使用开源的开源缓存实现,eg:OSCache,Ehcache;
6)尽量不使用finalize()方法;
7)在必要的时候可以考虑使用软引用SoftReference。