寻求关于共享Netty老板/工人池的建议(Seeking advice on sharing Netty boss/worker pools)

 上下文：Netty 3.6.3.Final，Java 1.7，Scala 2.9.x. 
 
 为了最小化（可能是空闲的）线程的数量，我想与不同的NIO套接字通道工厂（TCP）和一个NioDatagramChannelFactory共享NIO客户端/服务器和工作池。 我使用至少两个（或三个与Finagle堆栈）的服务器/客户端引导集，每个都有自己的NIO套接字通道工厂。 为每个boss和worker池使用一个新的缓存线程池会导致大量时间没有使用的线程负载。 粗略的目标是将所有bootstraps / channel工厂的工作者数量限制为2 * CPU核心数和老板数到CPU核心数。 
 
 我正在尝试切换到NioServer / ClientBossPool和NioWorkerPool，用于我自己的一组bootstrap。 但是根据底层ThreadPoolExecutor的配置，关闭引导程序会导致主线程在AbstractNioSelector关闭锁存器上永远等待。 
 
class NioClientBossPoolTest {

  @Test def shutdown() {
    val corePoolSize = 1
    val maxPoolSize = Integer.MAX_VALUE
    val keepAliveSeconds = 60
    val keepAliveUnit = TimeUnit.SECONDS

    val blocking = true
    val queue: BlockingQueue[Runnable] =
      if(blocking) new LinkedBlockingQueue[Runnable](Integer.MAX_VALUE)
      else new SynchronousQueue[Runnable]()

    val executor = new ThreadPoolExecutor(corePoolSize,
        maxPoolSize,
        keepAliveSeconds,
        keepAliveUnit,
        queue)

    val clientBossPool = new NioClientBossPool(executor, 1) // B
    new NioServerBossPool(executor, 1) // C
    val workerPool = new NioWorkerPool(executor, 1) // A

    val channelFactory = new NioClientSocketChannelFactory(clientBossPool, workerPool)
    val bootstrap = new ClientBootstrap(channelFactory)

    // hangs waiting for shutdown latch in AbstractNioSelector (NioWorker or NioClientBoss
    // depending on the order of statement A, B, C) for
    // LinkedBlockingQueue, corePoolSize = 1 and sequence of statements A, B and C other than [B, A, C]
    // LinkedBlockingQueue, corePoolSize = 2 and sequence of statements A, B and C other than
    // [A, B, C], [B, C, A] and [C, B, A]
    bootstrap.shutdown()
  }
}
 
 我很确定执行程序服务配置必须满足一些特定要求，但哪些（核心池大小，队列类型）？ bootstrap.shutdown（）将永远阻塞，除非语句A，B和C的执行顺序正好是[B，A，C]。 将核心池大小增加到2个块，用于三个语句的六个组合中的三个。 核心池大小> 2或具有SynchronousQueue时，每个组合终止。 

Context: Netty 3.6.3.Final, Java 1.7, Scala 2.9.x
 
In order to minimize the number (of possibly idling) threads, I'd like to share NIO client/server and worker pools with different NIO socket channel factories (TCP) and one NioDatagramChannelFactory. I'm using at least two (or three with the Finagle stack) sets of server/client bootstraps, each with their own NIO socket channel factories. Using a new cached thread pool for each and every boss and worker pool result in a load of threads which are not used most of the time. The rough goal would be to limit the number of workers over all bootstraps/channel factories to 2 * CPU core count and number of bosses to CPU core count.
 
I'm trying to switch over to NioServer/ClientBossPool and NioWorkerPool for one of my own set of bootstraps. But depending of the configuration of the underlying ThreadPoolExecutor, shutting down a bootstrap causes the main thread to wait forever on a AbstractNioSelector shutdown latch.
 
class NioClientBossPoolTest {

  @Test def shutdown() {
    val corePoolSize = 1
    val maxPoolSize = Integer.MAX_VALUE
    val keepAliveSeconds = 60
    val keepAliveUnit = TimeUnit.SECONDS

    val blocking = true
    val queue: BlockingQueue[Runnable] =
      if(blocking) new LinkedBlockingQueue[Runnable](Integer.MAX_VALUE)
      else new SynchronousQueue[Runnable]()

    val executor = new ThreadPoolExecutor(corePoolSize,
        maxPoolSize,
        keepAliveSeconds,
        keepAliveUnit,
        queue)

    val clientBossPool = new NioClientBossPool(executor, 1) // B
    new NioServerBossPool(executor, 1) // C
    val workerPool = new NioWorkerPool(executor, 1) // A

    val channelFactory = new NioClientSocketChannelFactory(clientBossPool, workerPool)
    val bootstrap = new ClientBootstrap(channelFactory)

    // hangs waiting for shutdown latch in AbstractNioSelector (NioWorker or NioClientBoss
    // depending on the order of statement A, B, C) for
    // LinkedBlockingQueue, corePoolSize = 1 and sequence of statements A, B and C other than [B, A, C]
    // LinkedBlockingQueue, corePoolSize = 2 and sequence of statements A, B and C other than
    // [A, B, C], [B, C, A] and [C, B, A]
    bootstrap.shutdown()
  }
}
 
I'm pretty sure that the executor service configuration has to meet some specific requirements, but which (core pool size, queue type)? bootstrap.shutdown() will block forever, except when the execution order of statement A, B and C is exactly [B, A, C]. Increasing core pool size to 2 blocks for three out of six combinations of the three statements. With a core pool size > 2 or with a SynchronousQueue each combination terminates.

原文：https://stackoverflow.com/questions/15587331