CompletableFuture.allOf().get()引起的线程池死锁

CompletableFuture<Void> voidCompletableFuture = CompletableFuture.allOf(completableFutures);可以同步等待一组异步请求都返回结果后，再往下执行。voidCompletableFuture.get()会同步等待所有结果返回，并且不会超时。就因为没有设置超时时间，造成了死锁的发生。

这是发生死锁的代码：

import java.util.concurrent.*;
import java.util.concurrent.atomic.AtomicInteger;

/**
 * 线程池死锁案例
 */
public class ThreadPoolTest {
    /**
     * 线程池
     */
    public static ThreadPoolExecutor THREAD_POOL = new ThreadPoolExecutor(5, 5, 60, TimeUnit.SECONDS, new LinkedBlockingQueue<>(), r -> new Thread(r, "THREAD_POOL_TEST"));

    public static void main(String[] args) throws InterruptedException {
        AtomicInteger atomicInteger = new AtomicInteger();
        for (int i = 0; i < 10; i++) {
            THREAD_POOL.submit(() -> {
                System.out.println("run task" + atomicInteger.getAndIncrement());
                executeSubTask();
            });
        }
        Thread.sleep(1000);

        System.out.println("task count: " + THREAD_POOL.getTaskCount());
        System.out.println("task in queue count: " + THREAD_POOL.getQueue().size());
    }

    private static void executeSubTask() {
        CompletableFuture[] completableFutures = new CompletableFuture[6];
        AtomicInteger atomicInteger = new AtomicInteger();
        for (int i = 0; i < 6; i++) {
            CompletableFuture<String> task = CompletableFuture.supplyAsync(() -> {
                System.out.println("ok-" + atomicInteger.getAndIncrement());
                try {
                    Thread.sleep(3000);
                } catch (InterruptedException e) {
                    throw new RuntimeException(e);
                }
                return "ok:" + atomicInteger.get();
            }, THREAD_POOL);
            completableFutures[i] = task;
        }
        System.out.println("submit subTask size:" + completableFutures.length);
        try {
            CompletableFuture.allOf(completableFutures).get();
        } catch (InterruptedException | ExecutionException e) {
            throw new RuntimeException(e);
        }
    }
}

处理逻辑是：

异步执行父任务
父任务里包含多个子任务

为了异步执行任务，所以放到线程池THREAD_POOL里。然后，任务里包含多个子任务，为了降低执行时间，把子任务做成了并发执行，使用了CompletableFuture.allOf(completableFutures).get()方法。

get()一般都会设置超时时间，但是因为异步任务设置了超时时间，所以就没再单独设置get()的超时时间，就引发了线程池死锁发生。

原因分析

子任务执行较慢，线程池的核心线程马上被父任务占满（5个）。父任务里提交的子任务都跑到阻塞队里里，等待核心线程来调度任务。核心线程又阻塞在CompletableFuture.allOf(completableFutures).get()这里，任务无法结束。

线程池状态

5个线程都处于WAITING状态，阻塞在 at java.util.concurrent.CompletableFuture.get(java.base@17/CompletableFuture.java:2072)

"THREAD_POOL_TEST" #12 prio=5 os_prio=0 cpu=4.14ms elapsed=485.93s tid=0x00007f64a448fcf0 nid=0x23b2 waiting on condition  [0x00007f648c1fa000]
   java.lang.Thread.State: WAITING (parking)
        at jdk.internal.misc.Unsafe.park(java.base@17/Native Method)
        - parking to wait for  <0x00000000e37436b0> (a java.util.concurrent.CompletableFuture$Signaller)
        at java.util.concurrent.locks.LockSupport.park(java.base@17/LockSupport.java:211)
        at java.util.concurrent.CompletableFuture$Signaller.block(java.base@17/CompletableFuture.java:1864)
        at java.util.concurrent.ForkJoinPool.unmanagedBlock(java.base@17/ForkJoinPool.java:3463)
        at java.util.concurrent.ForkJoinPool.managedBlock(java.base@17/ForkJoinPool.java:3434)
        at java.util.concurrent.CompletableFuture.waitingGet(java.base@17/CompletableFuture.java:1898)
        at java.util.concurrent.CompletableFuture.get(java.base@17/CompletableFuture.java:2072)
        at ThreadPoolTest.executeSubTask(ThreadPoolTest.java:45)
        at ThreadPoolTest.lambda$main$1(ThreadPoolTest.java:19)
        at ThreadPoolTest$$Lambda$237/0x0000000800d54428.run(Unknown Source)
        at java.util.concurrent.Executors$RunnableAdapter.call(java.base@17/Executors.java:539)
        at java.util.concurrent.FutureTask.run(java.base@17/FutureTask.java:264)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(java.base@17/ThreadPoolExecutor.java:1136)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(java.base@17/ThreadPoolExecutor.java:635)
        at java.lang.Thread.run(java.base@17/Thread.java:833)

"THREAD_POOL_TEST" #13 prio=5 os_prio=0 cpu=0.54ms elapsed=485.93s tid=0x00007f64a4490b70 nid=0x23b3 waiting on condition  [0x00007f647dffe000]
   java.lang.Thread.State: WAITING (parking)
        at jdk.internal.misc.Unsafe.park(java.base@17/Native Method)
        - parking to wait for  <0x00000000e375c280> (a java.util.concurrent.CompletableFuture$Signaller)
        at java.util.concurrent.locks.LockSupport.park(java.base@17/LockSupport.java:211)
        at java.util.concurrent.CompletableFuture$Signaller.block(java.base@17/CompletableFuture.java:1864)
        at java.util.concurrent.ForkJoinPool.unmanagedBlock(java.base@17/ForkJoinPool.java:3463)
        at java.util.concurrent.ForkJoinPool.managedBlock(java.base@17/ForkJoinPool.java:3434)
        at java.util.concurrent.CompletableFuture.waitingGet(java.base@17/CompletableFuture.java:1898)
        at java.util.concurrent.CompletableFuture.get(java.base@17/CompletableFuture.java:2072)
        at ThreadPoolTest.executeSubTask(ThreadPoolTest.java:45)
        at ThreadPoolTest.lambda$main$1(ThreadPoolTest.java:19)
        at ThreadPoolTest$$Lambda$237/0x0000000800d54428.run(Unknown Source)
        at java.util.concurrent.Executors$RunnableAdapter.call(java.base@17/Executors.java:539)
        at java.util.concurrent.FutureTask.run(java.base@17/FutureTask.java:264)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(java.base@17/ThreadPoolExecutor.java:1136)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(java.base@17/ThreadPoolExecutor.java:635)
        at java.lang.Thread.run(java.base@17/Thread.java:833)

"THREAD_POOL_TEST" #15 prio=5 os_prio=0 cpu=1.30ms elapsed=485.92s tid=0x00007f64a4491b70 nid=0x23b4 waiting on condition  [0x00007f647defd000]
   java.lang.Thread.State: WAITING (parking)
        at jdk.internal.misc.Unsafe.park(java.base@17/Native Method)
        - parking to wait for  <0x00000000e377bef8> (a java.util.concurrent.CompletableFuture$Signaller)
        at java.util.concurrent.locks.LockSupport.park(java.base@17/LockSupport.java:211)
        at java.util.concurrent.CompletableFuture$Signaller.block(java.base@17/CompletableFuture.java:1864)
        at java.util.concurrent.ForkJoinPool.unmanagedBlock(java.base@17/ForkJoinPool.java:3463)
        at java.util.concurrent.ForkJoinPool.managedBlock(java.base@17/ForkJoinPool.java:3434)
        at java.util.concurrent.CompletableFuture.waitingGet(java.base@17/CompletableFuture.java:1898)
        at java.util.concurrent.CompletableFuture.get(java.base@17/CompletableFuture.java:2072)
        at ThreadPoolTest.executeSubTask(ThreadPoolTest.java:45)
        at ThreadPoolTest.lambda$main$1(ThreadPoolTest.java:19)
        at ThreadPoolTest$$Lambda$237/0x0000000800d54428.run(Unknown Source)
        at java.util.concurrent.Executors$RunnableAdapter.call(java.base@17/Executors.java:539)
        at java.util.concurrent.FutureTask.run(java.base@17/FutureTask.java:264)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(java.base@17/ThreadPoolExecutor.java:1136)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(java.base@17/ThreadPoolExecutor.java:635)
        at java.lang.Thread.run(java.base@17/Thread.java:833)

"THREAD_POOL_TEST" #14 prio=5 os_prio=0 cpu=1.08ms elapsed=485.92s tid=0x00007f6454002a90 nid=0x23b5 waiting on condition  [0x00007f647ddfc000]
   java.lang.Thread.State: WAITING (parking)
        at jdk.internal.misc.Unsafe.park(java.base@17/Native Method)
        - parking to wait for  <0x00000000e379a918> (a java.util.concurrent.CompletableFuture$Signaller)
        at java.util.concurrent.locks.LockSupport.park(java.base@17/LockSupport.java:211)
        at java.util.concurrent.CompletableFuture$Signaller.block(java.base@17/CompletableFuture.java:1864)
        at java.util.concurrent.ForkJoinPool.unmanagedBlock(java.base@17/ForkJoinPool.java:3463)
        at java.util.concurrent.ForkJoinPool.managedBlock(java.base@17/ForkJoinPool.java:3434)
        at java.util.concurrent.CompletableFuture.waitingGet(java.base@17/CompletableFuture.java:1898)
        at java.util.concurrent.CompletableFuture.get(java.base@17/CompletableFuture.java:2072)
        at ThreadPoolTest.executeSubTask(ThreadPoolTest.java:45)
        at ThreadPoolTest.lambda$main$1(ThreadPoolTest.java:19)
        at ThreadPoolTest$$Lambda$237/0x0000000800d54428.run(Unknown Source)
        at java.util.concurrent.Executors$RunnableAdapter.call(java.base@17/Executors.java:539)
        at java.util.concurrent.FutureTask.run(java.base@17/FutureTask.java:264)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(java.base@17/ThreadPoolExecutor.java:1136)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(java.base@17/ThreadPoolExecutor.java:635)
        at java.lang.Thread.run(java.base@17/Thread.java:833)

"THREAD_POOL_TEST" #16 prio=5 os_prio=0 cpu=0.33ms elapsed=485.92s tid=0x00007f645c001250 nid=0x23b6 waiting on condition  [0x00007f647dcfb000]
   java.lang.Thread.State: WAITING (parking)
        at jdk.internal.misc.Unsafe.park(java.base@17/Native Method)
        - parking to wait for  <0x00000000e37b83f8> (a java.util.concurrent.CompletableFuture$Signaller)
        at java.util.concurrent.locks.LockSupport.park(java.base@17/LockSupport.java:211)
        at java.util.concurrent.CompletableFuture$Signaller.block(java.base@17/CompletableFuture.java:1864)
        at java.util.concurrent.ForkJoinPool.unmanagedBlock(java.base@17/ForkJoinPool.java:3463)
        at java.util.concurrent.ForkJoinPool.managedBlock(java.base@17/ForkJoinPool.java:3434)
        at java.util.concurrent.CompletableFuture.waitingGet(java.base@17/CompletableFuture.java:1898)
        at java.util.concurrent.CompletableFuture.get(java.base@17/CompletableFuture.java:2072)
        at ThreadPoolTest.executeSubTask(ThreadPoolTest.java:45)
        at ThreadPoolTest.lambda$main$1(ThreadPoolTest.java:19)
        at ThreadPoolTest$$Lambda$237/0x0000000800d54428.run(Unknown Source)
        at java.util.concurrent.Executors$RunnableAdapter.call(java.base@17/Executors.java:539)
        at java.util.concurrent.FutureTask.run(java.base@17/FutureTask.java:264)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(java.base@17/ThreadPoolExecutor.java:1136)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(java.base@17/ThreadPoolExecutor.java:635)
        at java.lang.Thread.run(java.base@17/Thread.java:833)

解决方案

父任务和子任务使用单独的线程池，避免出现循环依赖
设置超时时间，CompletableFuture.allOf(completableFutures).get(3, TimeUnit.SECONDS)

另外，任务子任务数量比较多，就需要更多的线程去执行，5个核心线程数量就不够了。但加大核心线程数，对于突发类型的任务，空闲时又浪费线程。我们就可以设置更大核心线程的同时，并设置核心线程在超时时可以回收。

public static ThreadPoolExecutor THREAD_POOL = new ThreadPoolExecutor(20, 20, 60, TimeUnit.SECONDS, new LinkedBlockingQueue<>(), r -> new Thread(r, "THREAD_POOL_TEST"));
// 回收核心线程
THREAD_POOL.allowCoreThreadTimeOut(true);