Java 并发实践 — ConcurrentHashMap 与 CAS

转载 http://www.importnew.com/26035.html

最近在做接口限流时涉及到了一个有意思问题,牵扯出了关于concurrentHashMap的一些用法,以及CAS的一些概念。限流算法很多,我主要就以最简单的计数器法来做引。先抽象化一下需求:统计每个接口访问的次数。一个接口对应一个url,也就是一个字符串,每调用一次对其进行加一处理。可能出现的问题主要有三个:

  1. 多线程访问,需要选择合适的并发容器
  2. 分布式下多个实例统计接口流量需要共享内存
  3. 流量统计应该尽可能不损耗服务器性能

但这次的博客并不是想描述怎么去实现接口限流,而是主要想描述一下遇到的问题,所以,第二点暂时不考虑,即不使用Redis。

说到并发的字符串统计,立即让人联想到的数据结构便是ConcurrentHashpMap<String,Long> urlCounter;
如果你刚刚接触并发可能会写出如代码清单1的代码

代码清单1:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
public class CounterDemo1 {
 
    private final Map<String, Long> urlCounter = new ConcurrentHashMap<>();
 
    //接口调用次数+1
    public long increase(String url) {
        Long oldValue = urlCounter.get(url);
        Long newValue = (oldValue == null) ? 1L : oldValue + 1;
        urlCounter.put(url, newValue);
        return newValue;
    }
 
    //获取调用次数
    public Long getCount(String url){
        return urlCounter.get(url);
    }
 
    public static void main(String[] args) {
        ExecutorService executor = Executors.newFixedThreadPool(10);
        final CounterDemo1 counterDemo = new CounterDemo1();
        int callTime = 100000;
        final String url = "http://localhost:8080/hello";
        CountDownLatch countDownLatch = new CountDownLatch(callTime);
        //模拟并发情况下的接口调用统计
        for(int i=0;i<callTime;i++){
            executor.execute(new Runnable() {
                @Override
                public void run() {
                    counterDemo.increase(url);
                    countDownLatch.countDown();
                }
            });
        }
        try {
            countDownLatch.await();
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        executor.shutdown();
        //等待所有线程统计完成后输出调用次数
        System.out.println("调用次数:"+counterDemo.getCount(url));
    }
}
 
console output:
调用次数:96526

都说concurrentHashMap是个线程安全的并发容器,所以没有显示加同步,实际效果呢并不如所愿。

问题就出在increase方法,concurrentHashMap能保证的是每一个操作(put,get,delete…)本身是线程安全的,但是我们的increase方法,对concurrentHashMap的操作是一个组合,先get再put,所以多个线程的操作出现了覆盖。如果对整个increase方法加锁,那么又违背了我们使用并发容器的初衷,因为锁的开销很大。我们有没有方法改善统计方法呢?
代码清单2罗列了concurrentHashMap父接口concurrentMap的一个非常有用但是又常常被忽略的方法。

代码清单2:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
/**
 * Replaces the entry for a key only if currently mapped to a given value.
 * This is equivalent to
 *  <pre> {@code
 * if (map.containsKey(key) && Objects.equals(map.get(key), oldValue)) {
 *   map.put(key, newValue);
 *   return true;
 * } else
 *   return false;
 * }</pre>
 *
 * except that the action is performed atomically.
 */
boolean replace(K key, V oldValue, V newValue);

这其实就是一个最典型的CAS操作,except that the action is performed atomically.这句话真是帮了大忙,我们可以保证比较和设置是一个原子操作,当A线程尝试在increase时,旧值被修改的话就回导致replace失效,而我们只需要用一个循环,不断获取最新值,直到成功replace一次,即可完成统计。

改进后的increase方法如下

代码清单3:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
public long increase2(String url) {
        Long oldValue, newValue;
        while (true) {
            oldValue = urlCounter.get(url);
            if (oldValue == null) {
                newValue = 1l;
                //初始化成功,退出循环
                if (urlCounter.putIfAbsent(url, 1l) == null)
                    break;
                //如果初始化失败,说明其他线程已经初始化过了
            } else {
                newValue = oldValue + 1;
                //+1成功,退出循环
                if (urlCounter.replace(url, oldValue, newValue))
                    break;
                //如果+1失败,说明其他线程已经修改过了旧值
            }
        }
        return newValue;
    }
 
console output:
调用次数:100000

再次调用后获得了正确的结果,上述方案看上去比较繁琐,因为第一次调用时需要进行一次初始化,所以多了一个判断,也用到了另一个CAS操作putIfAbsent,他的源代码描述如下:

代码清单4:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
/**
     * If the specified key is not already associated
     * with a value, associate it with the given value.
     * This is equivalent to
     *  <pre> {@code
     * if (!map.containsKey(key))
     *   return map.put(key, value);
     * else
     *   return map.get(key);
     * }</pre>
     *
     * except that the action is performed atomically.
     *
     * @implNote This implementation intentionally re-abstracts the
     * inappropriate default provided in {@code Map}.
     *
     * @param key key with which the specified value is to be associated
     * @param value value to be associated with the specified key
     * @return the previous value associated with the specified key, or
     *         {@code null} if there was no mapping for the key.
     *         (A {@code null} return can also indicate that the map
     *         previously associated {@code null} with the key,
     *         if the implementation supports null values.)
     * @throws UnsupportedOperationException if the {@code put} operation
     *         is not supported by this map
     * @throws ClassCastException if the class of the specified key or value
     *         prevents it from being stored in this map
     * @throws NullPointerException if the specified key or value is null,
     *         and this map does not permit null keys or values
     * @throws IllegalArgumentException if some property of the specified key
     *         or value prevents it from being stored in this map
     */
     V putIfAbsent(K key, V value);

简单翻译如下:“如果(调用该方法时)key-value 已经存在,则返回那个 value 值。如果调用时 map 里没有找到 key 的 mapping,返回一个 null 值”。值得注意点的一点就是concurrentHashMap的value是不能存在null值的。实际上呢,上述的方案也可以把Long替换成AtomicLong,可以简化实现, ConcurrentHashMap

1
2
3
4
5
6
7
8
9
10
11
private AtomicLongMap<String> urlCounter3 = AtomicLongMap.create();
 
public long increase3(String url) {
    long newValue = urlCounter3.incrementAndGet(url);
    return newValue;
}
 
 
public Long getCount3(String url) {
    return urlCounter3.get(url);
}

看一下他的源码就会发现,其实和代码清单3思路差不多,只不过功能更完善了一点。

和CAS很像的操作,我之前的博客中提到过数据库的乐观锁,用version字段来进行并发控制,其实也是一种compare and swap的思想。

posted @ 2019-01-22 22:24  He_quotes  阅读(325)  评论(0编辑  收藏  举报