1. 概览

大纲


预备知识

  • 希望你不是一个初学者
  • 线程安全问题,需要你接触过 Java Web 开发、Jdbc 开发、Web 服务器、分布式框架时才会遇到
  • 基于 JDK 8,最好对函数式编程、lambda 有一定了解
  • 采用了 slf4j 打印日志,这是好的实践
  • 采用了 lombok 简化 java bean 编写
  • 给每个线程好名字,这也是一项好的实践

2.进程与线程

本章内容

  • 进程和线程的概念
  • 并行和并发的概念
  • 线程基本应用

2.1 进程与线程

进程

  • 程序由指令和数据组成,但这些指令要运行,数据要读写,就必须将指令加载至 CPU,数据加载至内存。在 指令运行过程中还需要用到磁盘、网络等设备。进程就是用来加载指令、管理内存、管理 IO 的 。
  • 当一个程序被运行,从磁盘加载这个程序的代码至内存,这时就开启了一个进程。
  • 进程就可以视为程序的一个实例。大部分程序可以同时运行多个实例进程(例如记事本、画图、浏览器 等),也有的程序只能启动一个实例进程(例如网易云音乐、360 安全卫士等)

线程

  • 一个进程之内可以分为一到多个线程。
  • 一个线程就是一个指令流,将指令流中的一条条指令以一定的顺序交给 CPU 执行
  • Java 中,线程作为最小调度单位进程作为资源分配的最小单位。 在 windows 中进程是不活动的,只是作 为线程的容器

二者对比

  • 进程基本上相互独立的,而线程存在于进程内,是进程的一个子集
  • 进程拥有共享的资源,如内存空间等,供其内部的线程共享
  • 进程间通信较为复杂
    • 同一台计算机的进程通信称为 IPC(Inter-process communication)
    • 不同计算机之间的进程通信,需要通过网络,并遵守共同的协议,例如 HTTP
  • 线程通信相对简单,因为它们共享进程内的内存,一个例子是多个线程可以访问同一个共享变量
  • 线程更轻量,线程上下文切换成本一般上要比进程上下文切换低

2.2 并行与并发

单核cpu下,线程实际还是串行执行的。操作系统中有一个组件叫做任务调度器,将 cpu 的时间片(windows 下时间片最小约为 15 毫秒)分给不同的程序使用,只是由于 cpu 在线程间(时间片很短)的切换非常快,人类感觉是同时运行的 。总结为一句话就是: 微观串行,宏观并行

一般会将这种线程轮流使用 CPU 的做法称为并发, concurrent

CPU 时间片 1 时间片 2 时间片 3 时间片 4
core 线程 1 线程 2 线程 3 线程 4
![ 375](https://picgo-1314714026.cos.ap-nanjing.myqcloud.com/img/image-20251105130448543.png)
多核 cpu下,每个 核(core) 都可以调度运行线程,这时候线程可以是并行的。
CPU 时间片 1 时间片 2 时间片 3 时间片 4
core1 线程 1 线程 2 线程 3 线程 4
core2 线程 4 线程 4 线程 2 线程 2
![ 413](https://picgo-1314714026.cos.ap-nanjing.myqcloud.com/img/image-20251105130710922.png)
引用 Rob Pike 的一段描述:
  • 并发(concurrent)是同一时间应对(dealing with)多件事情的能力 。
  • 并行(parallel)是同一时间动手做(doing)多件事情的能力。

2.3 应用

应用之异步调用(案例1)

以调用方角度来讲,如果

  • 需要等待结果返回,才能继续运行就是同步
  • 不需要等待结果返回,就能继续运行就是异步

需要等待结果

这时既可以使用同步处理,也可以使用异步来处理
join 实现(同步)

1
2
3
4
5
6
7
8
9
10
11
12
13
static int result = 0;
private static void test1() throws InterruptedException {
log.debug("开始");
Thread t1 = new Thread(() -> {
log.debug("开始");
sleep(1);
log.debug("结束");
result = 10;
}, "t1");
t1.start();
t1.join();
log.debug("结果为:{}", result);
}

输出

1
2
3
4
20:30:40.453 [main] c.TestJoin - 开始
20:30:40.541 [Thread-0] c.TestJoin - 开始
20:30:41.543 [Thread-0] c.TestJoin - 结束
20:30:41.551 [main] c.TestJoin - 结果为:10

评价

  • 需要外部共享变量,不符合面向对象封装的思想
  • 必须等待线程结束,不能配合线程池使用
Future 实现(同步)
1
2
3
4
5
6
7
8
9
10
11
private static void test2() throws InterruptedException, ExecutionException {
log.debug("开始");
FutureTask<Integer> result = new FutureTask<>(() -> {
log.debug("开始");
sleep(1);
log.debug("结束");
return 10;
});
new Thread(result, "t1").start();
log.debug("结果为:{}", result.get());
}

输出

1
2
3
4
10:11:57.880 c.TestSync [main] - 开始
10:11:57.942 c.TestSync [t1] - 开始
10:11:58.943 c.TestSync [t1] - 结束
10:11:58.943 c.TestSync [main] - 结果为:10

评价

  • 规避了使用 join 之前的缺点
  • 可以方便配合线程池使用
1
2
3
4
5
6
7
8
9
10
11
12
private static void test3() throws InterruptedException, ExecutionException {
ExecutorService service = Executors.newFixedThreadPool(1);
log.debug("开始");
Future<Integer> result = service.submit(() -> {
log.debug("开始");
sleep(1);
log.debug("结束");
return 10;
});
log.debug("结果为:{}, result 的类型:{}", result.get(), result.getClass());
service.shutdown();
}

输出

1
2
3
4
10:17:40.090 c.TestSync [main] - 开始
10:17:40.150 c.TestSync [pool-1-thread-1] - 开始
10:17:41.151 c.TestSync [pool-1-thread-1] - 结束
10:17:41.151 c.TestSync [main] - 结果为:10, result 的类型:class java.util.concurrent.FutureTask

评价

  • 仍然是 main 线程接收结果
  • get 方法是让调用线程同步等待
自定义实现(同步)

见模式篇:保护性暂停模式

CompletableFuture 实现(异步)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
private static void test4() {
// 进行计算的线程池
ExecutorService computeService = Executors.newFixedThreadPool(1);
// 接收结果的线程池
ExecutorService resultService = Executors.newFixedThreadPool(1);
log.debug("开始");
CompletableFuture.supplyAsync(() -> {
log.debug("开始");
sleep(1);
log.debug("结束");
return 10;
}, computeService).thenAcceptAsync((result) -> {
log.debug("结果为:{}", result);
}, resultService);
}

输出

1
2
3
4
10:36:28.114 c.TestSync [main] - 开始
10:36:28.164 c.TestSync [pool-1-thread-1] - 开始
10:36:29.165 c.TestSync [pool-1-thread-1] - 结束
10:36:29.165 c.TestSync [pool-2-thread-1] - 结果为:10

评价

  • 可以让调用线程异步处理结果,实际是其他线程去同步等待
  • 可以方便地分离不同职责的线程池
  • 以任务为中心,而不是以线程为中心
BlockingQueue 实现(异步)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
private static void test6() {
ExecutorService consumer = Executors.newFixedThreadPool(1);
ExecutorService producer = Executors.newFixedThreadPool(1);
BlockingQueue<Integer> queue = new SynchronousQueue<>();
log.debug("开始");
producer.submit(() -> {
log.debug("开始");
sleep(1);
log.debug("结束");
try {
queue.put(10);
} catch (InterruptedException e) {
e.printStackTrace();
}
});
consumer.submit(() -> {
try {
Integer result = queue.take();
log.debug("结果为:{}", result);
} catch (InterruptedException e) {
e.printStackTrace();
}
});
}

不需等待结果

这时最好是使用异步来处理

普通线程实现
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
@Slf4j(topic = "c.FileReader")
public class FileReader {
public static void read(String filename) {
int idx = filename.lastIndexOf(File.separator);
String shortName = filename.substring(idx + 1);
try (FileInputStream in = new FileInputStream(filename)) {
long start = System.currentTimeMillis();
log.debug("read [{}] start ...", shortName);
byte[] buf = new byte[1024];
int n = -1;
do {
n = in.read(buf);
} while (n != -1);
long end = System.currentTimeMillis();
log.debug("read [{}] end ... cost: {} ms", shortName, end - start);
} catch (IOException e) {
e.printStackTrace();
}
}
}

没有用线程时,方法的调用是同步的:

1
2
3
4
5
6
7
8
@Slf4j(topic = "c.Sync")
public class Sync {
public static void main(String[] args) {
String fullPath = "E:\\1.mp4";
FileReader.read(fullPath);
log.debug("do other things ...");
}
}

输出

1
2
3
18:39:15 [main] c.FileReader - read [1.mp4] start ...
18:39:19 [main] c.FileReader - read [1.mp4] end ... cost: 4090 ms
18:39:19 [main] c.Sync - do other things ...

使用了线程后,方法的调用时异步的:

1
2
3
4
private static void test1() {
new Thread(() -> FileReader.read(Constants.MP4_FULL_PATH)).start();
log.debug("do other things ...");
}

输出

1
2
3
18:41:53 [main] c.Async - do other things ...
18:41:53 [Thread-0] c.FileReader - read [1.mp4] start ...
18:41:57 [Thread-0] c.FileReader - read [1.mp4] end ... cost: 4197 ms
线程池实现
1
2
3
4
5
6
private static void test2() {
ExecutorService service = Executors.newFixedThreadPool(1);
service.execute(() -> FileReader.read(Constants.MP4_FULL_PATH));
log.debug("do other things ...");
service.shutdown();
}

输出

1
2
3
11:03:31.245 c.TestAsyc [main] - do other things ... 
11:03:31.245 c.FileReader [pool-1-thread-1] - read [1.mp4] start ...
11:03:33.479 c.FileReader [pool-1-thread-1] - read [1.mp4] end ... cost: 2235 ms
CompletableFuture 实现
1
2
3
4
5
private static void test3() throws IOException {
CompletableFuture.runAsync(() -> FileReader.read(Constants.MP4_FULL_PATH));
log.debug("do other things ...");
System.in.read();
}

输出

1
2
3
11:09:38.145 c.TestAsyc [main] - do other things ... 
11:09:38.145 c.FileReader [ForkJoinPool.commonPool-worker-1] - read [1.mp4] start ...
11:09:40.514 c.FileReader [ForkJoinPool.commonPool-worker-1] - read [1.mp4] end ... cost: 2369 ms

以调用方角度来讲,

  • 如果 需要等待结果返回,才能继续运行就是同步
  • 不需要等待结果返回,就能继续运行就是异步

1.设计

多线程可以让方法执行变为异步的(即不要巴巴干等着)、比如说读取磁盘文件时,假设读取操作花费了 5 秒钟,如果没有线程调度机制,这 5 秒 cpu 什么都做不了,其它代码都得暂停…

2.结论

  • 比如在项目中,视频文件需要转换格式等操作比较费时,这时开一个新线程处理视频转换,避免阻塞主线程
  • tomcat 的异步 servlet 也是类似的目的,让用户线程处理耗时较长的操作,避免阻塞 tomcat 的工作线程
  • ui 程序中,开线程进行其他操作,避免阻塞 ui 线程

应用之提高效率(案例1)

充分利用多核 cpu 的优势,提高运行效率。想象下面的场景,执行 3 个计算,最后将计算结果汇总。

1
2
3
4
计算 1 花费 10 ms
计算 2 花费 11 ms
计算 3 花费 9 ms
汇总需要 1 ms
  • 如果是串行执行,那么总共花费的时间是 10 + 11 + 9 + 1 = 31ms

  • 但如果是四核 cpu,各个核心分别使用线程 1 执行计算 1,线程 2 执行计算 2,线程 3 执行计算 3,那么 3 个 线程是并行的,花费时间只取决于最长的那个线程运行的时间,即 11ms 最后加上汇总时间只会花费 12ms

注意:

需要在多核 cpu 才能提高效率,单核仍然时是轮流执行

1.设计

代码见【应用之效率-案例1】

2.结论
  1. 单核 cpu 下,多线程不能实际提高程序运行效率,只是为了能够在不同的任务之间切换,不同线程轮流使用 cpu ,不至于一个线程总占用 cpu,别的线程没法干活
  2. 多核 cpu 可以并行跑多个线程,但能否提高程序运行效率还是要分情况的
    • 有些任务,经过精心设计,将任务拆分,并行执行,当然可以提高程序的运行效率。但不是所有计算任 务都能拆分(参考后文的【阿姆达尔定律】
    • 也不是所有任务都需要拆分,任务的目的如果不同,谈拆分和效率没啥意义
  3. IO 操作不占用 cpu,只是我们一般拷贝文件使用的是【阻塞 IO】,这时相当于线程虽然不用 cpu,但需要一 直等待 IO 结束,没能充分利用线程。所以才有后面的【非阻塞 IO】和【异步 IO】优化。