hazelcast 3.12：本地主机上的多播自动发现停止工作

Question

昨天，我多次启动了两个嵌入式 hazelcast 节点，每次第二个节点启动时，它们都会加入到一个集群中。从今天早上开始，他们不再加入集群。这是启动节点的最少代码：

object BrokenHazelcastAutodiscovery extends App {
  val cfg = new Config()
  Hazelcast.newHazelcastInstance(cfg)
}

我在 Macbook 上运行了它。我不确定昨晚是否让节点保持运行。假设我已经并且假设他们一直在不时地使用多播发现数据包 ping 公司网络。我想，网络管理员以某种方式关闭了我笔记本电脑上的多播功能。好吧，这是一段代码，用于检查是否可以在本地主机上多播和接收多播数据包：

object WhetherMulticastEnabledCheck {
  import java.net._

  def main(args: Array[String]): Unit = {
    val port = 33766
    val addr = new InetSocketAddress(InetAddress.getByName("225.6.7.8"), port)
    val netif = NetworkInterface.getByInetAddress(InetAddress.getLocalHost)
    val sok = new MulticastSocket(addr.getPort)
    sok.joinGroup(addr, netif)
    sok.setTimeToLive(0)

    val anotherStarted =
      try { new ServerSocket(port); false }
      catch { case _: BindException => true }
    val seq = anotherStarted match {
      case true  => new Sequence(0xFFFF0000)
      case false => new Sequence(0x0000FFFF)
    }
    new Receiver(sok).start()
    new Sender(sok, addr, seq).start()
  }

  class Sequence(mask: Int) {
    private var i = 0
    def next: Int = {
      i += 1
      if (i > 0x0FFF) i = 1
      ((i << 16) | i) & mask
    }
  }

  private class Sender(sok: DatagramSocket, to: SocketAddress, seq: Sequence) {
    private val t = new Thread(() => {
      val buf = new Array[Byte](4)
      val p = new DatagramPacket(buf, buf.length, to)
      println("sending started...")
      while (true) {
        val t = seq.next
        buf.update(0, (t >> 24).toByte)
        buf.update(1, (t >> 16).toByte)
        buf.update(2, (t >>  8).toByte)
        buf.update(3, (t >>  0).toByte)
        sok.send(p)
        println(s"sent     ${t.formatted("%08X").replace('0', '.')}")
        Thread.sleep(2500)
      }
    }, "sending-thread")
    def start(): Unit = t.start()
  }

  private class Receiver(sok: DatagramSocket) {
    private val t = new Thread(() => {
      val buf = new Array[Byte](4)
      val p = new DatagramPacket(buf, buf.length)
      println("receiving started...")
      while (true) {
        sok.receive(p)
        val t =
          (buf(0) << 24) |
          (buf(1) << 16) |
          (buf(2) <<  8) |
          (buf(3) <<  0)
        println(s"received ${t.formatted("%08X").replace('0', '.')}")
      }
    }, "receiving-thread")
    def start(): Unit = t.start()
  }
}

这是输出

 proc 1              | proc 2
---------------------|----------------------
receiving started... |     
sending started...   | sending started...  
sent     .......1    | receiving started...     
received .......1    | received ...1....     
received ...1....    | sent     ...1....     
sent     .......2    | received .......2     
received .......2    | sent     ...2....     
received ...2....    | received ...2....     
sent     .......3    | received .......3     
...

我手动将其并排放置，并将第二个进程的输出移动一行，以帮助在视觉上捕捉开始时刻的变化。输出意味着多播在我的笔记本电脑上工作。

那么，你们能建议一下可能会出现什么问题，导致节点停止相互发现并加入集群吗？

Answer 1

我第二次重新启动了笔记本电脑，

BrokenHazelcastAutodiscovery 

又开始工作了。 🤷🏼