Java线程池监控小结

最近我们组杨青同学遇到一个使用线程池不当的问题：异步处理的线程池线程将主线程hang住了，分析代码发现是线程池的拒绝策略设置得不合理，设置为CallerRunsPolicy。当异步线程的执行效率降低时，阻塞队列满了，触发了拒绝策略，进而导致主线程hang死。

从这个问题中，我们学到了两点：

• 线程池的使用，需要充分分析业务场景后作出选择，必要的情况下需要自定义线程池；
• 线程池的运行状况，也需要监控

关于线程池的监控，我参考了《Java编程的艺术》中提供的思路实现的，分享下我的代码片段，如下：

public class AsyncThreadExecutor implements AutoCloseable {

    private static final int DEFAULT_QUEUE_SIZE = 1000;

    private static final int DEFAULT_POOL_SIZE = 10;

    @Setter
    private int queueSize = DEFAULT_QUEUE_SIZE;

    @Setter
    private int poolSize = DEFAULT_POOL_SIZE;

    /**
     * 用于周期性监控线程池的运行状态
     */
    private final ScheduledExecutorService scheduledExecutorService =
        Executors.newSingleThreadScheduledExecutor(new BasicThreadFactory.Builder().namingPattern("async thread executor monitor").build());

    /**
     * 自定义异步线程池
     * （1）任务队列使用有界队列
     * （2）自定义拒绝策略
     */
    private final ThreadPoolExecutor threadPoolExecutor =
        new ThreadPoolExecutor(poolSize, poolSize, 0, TimeUnit.MILLISECONDS, new ArrayBlockingQueue(queueSize),
                               new BasicThreadFactory.Builder().namingPattern("async-thread-%d").build(),
                               (r, executor) -> log.error("the async executor pool is full!!"));
    private final ExecutorService executorService = threadPoolExecutor;

    @PostConstruct
    public void init() {
        scheduledExecutorService.scheduleAtFixedRate(() -> {
            /**
             * 线程池需要执行的任务数
             */
            long taskCount = threadPoolExecutor.getTaskCount();
            /**
             * 线程池在运行过程中已完成的任务数
             */
            long completedTaskCount = threadPoolExecutor.getCompletedTaskCount();
            /**
             * 曾经创建过的最大线程数
             */
            long largestPoolSize = threadPoolExecutor.getLargestPoolSize();
            /**
             * 线程池里的线程数量
             */
            long poolSize = threadPoolExecutor.getPoolSize();
            /**
             * 线程池里活跃的线程数量
             */
            long activeCount = threadPoolExecutor.getActiveCount();

            log.info("async-executor monitor. taskCount:{}, completedTaskCount:{}, largestPoolSize:{}, poolSize:{}, activeCount:{}",
                     taskCount, completedTaskCount, largestPoolSize, poolSize, activeCount);
        }, 0, 10, TimeUnit.MINUTES);
    }

    public void execute(Runnable task) {
        executorService.execute(task);
    }

    @Override
    public void close() throws Exception {
        executorService.shutdown();
    }
}

这里的主要思路是：（1）使用有界队列的固定数量线程池；（2）拒绝策略是将任务丢弃，但是需要记录错误日志；（3）使用一个调度线程池对业务线程池进行监控。

在查看监控日志的时候，看到下图所示的监控日志：

这里我对largestPooSize的含义比较困惑，按字面理解是“最大的线程池数量”，但是按照线程池的定义，maximumPoolSize和coreSize相同的时候（在这里，都是10），一个线程池里的最大线程数是10，那么为什么largestPooSize可以是39呢？我去翻这块的源码：

    /**
     * Returns the largest number of threads that have ever
     * simultaneously been in the pool.
     *
     * @return the number of threads
     */
    public int getLargestPoolSize() {
        final ReentrantLock mainLock = this.mainLock;
        mainLock.lock();
        try {
            return largestPoolSize;
        } finally {
            mainLock.unlock();
        }
    }

注释的翻译是：返回在这个线程池里曾经同时存在过的线程数。再看这个变量largestPoolSize在ThreadExecutor中的赋值的地方，代码如下：

private boolean addWorker(Runnable firstTask, boolean core) {
        retry:
        for (;;) {
            int c = ctl.get();
            int rs = runStateOf(c);

            // Check if queue empty only if necessary.
            if (rs >= SHUTDOWN &&
                ! (rs == SHUTDOWN &&
                   firstTask == null &&
                   ! workQueue.isEmpty()))
                return false;

            for (;;) {
                int wc = workerCountOf(c);
                if (wc >= CAPACITY ||
                    wc >= (core ? corePoolSize : maximumPoolSize))
                    return false;
                if (compareAndIncrementWorkerCount(c))
                    break retry;
                c = ctl.get();  // Re-read ctl
                if (runStateOf(c) != rs)
                    continue retry;
                // else CAS failed due to workerCount change; retry inner loop
            }
        }

        boolean workerStarted = false;
        boolean workerAdded = false;
        Worker w = null;
        try {
            w = new Worker(firstTask);
            final Thread t = w.thread;
            if (t != null) {
                final ReentrantLock mainLock = this.mainLock;
                mainLock.lock();
                try {
                    // Recheck while holding lock.
                    // Back out on ThreadFactory failure or if
                    // shut down before lock acquired.
                    int rs = runStateOf(ctl.get());

                    if (rs < SHUTDOWN ||
                        (rs == SHUTDOWN && firstTask == null)) {
                        if (t.isAlive()) // precheck that t is startable
                            throw new IllegalThreadStateException();
                        workers.add(w);
                        int s = workers.size();
                        if (s > largestPoolSize)
                            largestPoolSize = s;//这里这里！
                        workerAdded = true;
                    }
                } finally {
                    mainLock.unlock();
                }
                if (workerAdded) {
                    t.start();
                    workerStarted = true;
                }
            }
        } finally {
            if (! workerStarted)
                addWorkerFailed(w);
        }
        return workerStarted;
    }

发现两点：

• largestPoolSize是worker集合的历史最大值，只增不减。largestPoolSize的大小是线程池曾创建的线程个数，跟线程池的容量无关；
• largestPoolSize<=maximumPoolSize。

PS：杨青同学是这篇文章的灵感来源，他做了很多压测。给了我很多思路，并跟我一起分析了一些代码。