JMH在基准测试之间徘徊(无法杀死分叉的JVM？)

Question

我有一大套宏基准，用于测量使用不同执行引擎（其中一些是严重多线程的）的各种模拟的执行时间。这些作业中的大多数使用我在抽象基类中指定的jmh设置，如下所示：

@ContextConfiguration(value = AbstractJobExecutorBenchmarker.CONTEXT_LOCATION)
@State(Scope.Benchmark)
@Fork(1)
@BenchmarkMode(Mode.SingleShotTime)
@Warmup(iterations = 1, timeUnit = TimeUnit.MILLISECONDS)
@Measurement(iterations = 5, timeUnit = TimeUnit.MILLISECONDS)
public abstract class AbstractJobExecutorBenchmarker extends AbstractTestNGSpringContextTests

（我还将 -gc -si 错误添加到命令行中）

当我启动这些基准中的任何一个或一些小的子集时，它们都可以正常工作。然而，如果我尝试一次运行所有的测试（即没有过滤选项），jmh成功地完成了其中的一些测试，然后在一个特定的基准测试后突然出现异常……重复完成，最终结果打印出来，但看起来并不是分叉JVM被终止，新的JVM没有启动，似乎整个过程都被卡住了。最后打印的报表示例：

Iteration   3: 03:59:52.059 [pool-8-thread-1] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.059 [pool-8-thread-3] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.060 [pool-8-thread-5] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.060 [pool-8-thread-6] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.061 [pool-8-thread-7] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.061 [pool-8-thread-8] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.061 [pool-8-thread-9] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.061 [pool-8-thread-10] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.062 [pool-8-thread-13] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.062 [pool-8-thread-14] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.063 [pool-8-thread-15] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.063 [pool-8-thread-18] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.063 [pool-8-thread-19] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.064 [pool-8-thread-20] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.064 [pool-8-thread-22] INFO  MessageTrafficController: Starting MessageTrafficController
03:59:52.065 [pool-8-thread-25] INFO  MessageTrafficController: Starting MessageTrafficController
04:00:57.600 [pool-8-thread-6] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.601 [pool-8-thread-7] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.604 [pool-8-thread-13] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.604 [pool-8-thread-10] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.603 [pool-8-thread-8] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.604 [pool-8-thread-9] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.606 [pool-8-thread-22] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.608 [pool-8-thread-25] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.606 [pool-8-thread-20] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.605 [pool-8-thread-15] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.606 [pool-8-thread-19] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.606 [pool-8-thread-18] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
04:00:57.604 [pool-8-thread-14] WARN  MessageTrafficController: Stopped due to class java.lang.InterruptedException
65598.920 ms

Result : 76749.529 Âą(99.9%) 804946.698 ms
  Statistics: (min, avg, max) = (39272.712, 76749.529, 125376.954), stdev = 44121.845
  Confidence interval (99.9%): [-728197.169, 881696.226]

如果我手动终止forked JVM，整个进程就会解锁。然而，在完成下一个基准测试之后，它又被卡住了……你知道为什么会这样吗？为什么当我分别启动相同的基准测试时，它不会发生？

此外，JMH中是否有一些超时选项，因此如果正常关闭时间太长，它会强行终止分叉的进程？

Answer 1

这已经在JMH开发邮件列表中得到回答:你需要调试你的工作负载，以禁止杂散线程，钩子等。阻止分叉的虚拟机退出。