【问题标题】:Why ISN'T there a race condition when I do a read-update-write in this java code? [duplicate]当我在这个 java 代码中进行读-更新-写时,为什么没有竞争条件? [复制]
【发布时间】:2017-05-02 08:42:12
【问题描述】:

我的代码有许多线程在没有任何同步的情况下对共享哈希映射进行读取-更新-写入。我用 1000 个线程,1M 迭代运行了很多次,所以我预计 4 个键的最终值将在 250M 左右,但不完全是,因为更新/写入可能会丢失,因为多个线程读取相同的值由于读取同时。但是,经过多次试验,我发现这些值始终为 250M,因此不会丢失任何更新。有人请帮帮我。 FWIW,我正在编写这个玩具应用程序,以将它与我使用 Collections.syncronizedMap 时进行比较,但是,我需要这个应用程序失败(以显示竞争条件),但它似乎总是产生一个不更新丢失的结果。

工作量:39.496

密钥:0 值:250000000

密钥:1 值:250000000

密钥:2 值:250000000

密钥:3 值:250000000

package threadsafecollections;

import java.util.HashMap;
import java.util.Map;

public class SynchronizedCollectionTask {
    
    public static int NUM_THREADS = 1000;
    
    public static class ReadUpdateWrite implements Runnable {
        
        Map<Integer, Integer> map;
        int threadId;
        public static int NUM_ITERATIONS = 1000000;
        
        ReadUpdateWrite(Map<Integer, Integer> m, int threadId) {
            this.map = m;
            this.threadId = threadId;
        }

        @Override
        public void run() {
            for (int i = 0; i < NUM_ITERATIONS; i++) {
                int key = threadId % 4;
                Integer val = this.map.get(key);
                map.put(key, val == null ? 1 : val + 1);
            }
            
            // At this point I expect each key in map to have value of CLOSE TO
            // numThreads * NUM_ITERATIONS / 4 = 250M.  I expect some threads to have
            // both read the same value, and therefore "miss" a write.  For example,
            // thread 4 and 8 both read key = 0 and see value = 3.  Now both thread 
            // increment value to 4, instead of one thread incrementing to 4 and the
            // other incrementing to 5.
        }
        
    }
    public static void main(String[] args) throws InterruptedException {
        Map<Integer, Integer> sharedMap = new HashMap<Integer, Integer>();
        
        
        
        // Initialize threads
        Thread[] readers = new Thread[NUM_THREADS];
        for (int i = 0; i < NUM_THREADS; i++) {
            readers[i] = new Thread(new ReadUpdateWrite(sharedMap, i));
        }
        
        long start = System.currentTimeMillis();
        
        // Start threads
        for (int i = 0; i < NUM_THREADS; i++) {
            readers[i].run();
        }
        
        // Join threads
        for (int i = 0; i < NUM_THREADS; i++) {
            readers[i].join();
        }
        
        long end = System.currentTimeMillis();
        
        System.out.println("Work took: " + (end - start) / 1000D);
        
        for (int key : sharedMap.keySet()) {
            System.out.println ("Key: " + key + " val: " + sharedMap.get(key));
        }
    }
}

【问题讨论】:

  • 你需要使用Thread.start。 Thread.run 只是在当前线程上运行该方法。
  • readers[i].start();,而不是 readers[i].run(); 详情请参阅链接问题的答案。

标签: java multithreading runnable race-condition


【解决方案1】:

恭喜:您已经编写了线程安全代码,因为Thread.run() 刚刚启动内部Runnable.run,因此您按顺序调用所有操作。请改用Thread.start()

这就是Thread.run() 的样子:

@Override
public void run() {
    if (target != null) {
        target.run();
    }
}

【讨论】:

    最近更新 更多