庆祝lv4升级成功，搬一旧博客的笔记

Java NIO的空轮询问题描绘

若Selector的轮询成果为空，也没有调用wakeup或新消息处理，Selector.select()被唤醒而产生空轮询，CPU使用率100%。

官方的BUG描绘

BUG只在Linux下产生，由于java的epoll完成存在bug，而linux下NIO底层使用的是epoll来完成的而windows不是。

selector = Selector.open();
serverSocketChannel = ServerSocketChannel.open();
serverSocketChannel.configureBlocking(false);
serverSocketChannel.socket().setReuseAddress(true);
serverSocketChannel.socket().bind(new InetSocketAddress(port));
log.info("服务器启动，等候衔接...");
for(;;){
  //int num = selector.select(): 一般情况下是堵塞形式，但确被唤醒，num==0即selectionKey为空，则 while (it.hasNext()) { ..}不会履行。  
   int num = selector.select();
   if (num==0){
       log.error("select wakes up with zero!!!");
   }
    //或者 while(true){ selector.select();
    Set<SelectionKey> readyKeys = selector.selectedKeys();
    Iterator<SelectionKey> it = readyKeys.iterator();
    while (it.hasNext()) {
        SelectionKey key = null;
        try {
            key = it.next();
            if (key.isReadable()){
                receive(key);
            }
           //...
      }
   }
}

Netty中处理该bug的方法

  1. 对Selector的select操作周期进行统计，每完成一次空的select操作进行一次计数，若在某个周期内接连产生N次空轮询，则触发了epoll死循环bug。
  2. 重建Selector，判断是否是其他线程建议的重建请求，若不是则将原SocketChannel从旧的Selector上去除注册，从头注册到新的Selector上，并将本来的Selector封闭。

详解代码

  • 1、selector.select(timeout)
    设置了一个超时时间，selector有以下4种情况跳出堵塞

    • 1.有事情产生
    • 2.wakeup
    • 3.超时
    • 4.空轮询bug

当时两种返回值不为0，可以跳出循环，超时有时间戳记录。所以当每次空轮询产生时会有专门的计数器+1，假如空轮询的次数超过了512次，就以为其触发了空轮询bug。

  • 2、触发bug后，netty直接重建一个selector，将本来的channel从头注册到新的selector上，将旧的 selector关掉，代码如下
private void select(boolean oldWakenUp) throws IOException {
for(;;)
    int selectedKeys = selector.select(timeoutMillis);
    selectCnt ++;
    if (selectedKeys != 0 || oldWakenUp || wakenUp.get() || hasTasks() || hasScheduledTasks()) {
            // - Selected something,
            // - waken up by user, or the task queue has a pending task.
            // - a scheduled task is ready for processing
            break;
        }
        long time = System.nanoTime();
        if (time - TimeUnit.MILLISECONDS.toNanos(timeoutMillis) >= currentTimeNanos) {
            // timeoutMillis elapsed without anything selected.
            // 超时
            selectCnt = 1;
        } else if (SELECTOR_AUTO_REBUILD_THRESHOLD > 0 &&
            selectCnt >= SELECTOR_AUTO_REBUILD_THRESHOLD) {//默认值512
            // The code exists in an extra method to ensure the method is not too big to inline as this
            // branch is not very likely to get hit very frequently.
            // 空轮询一次 cnt+1  假如一个周期内次数超过512，则假定产生了空轮询bug，重建selector
            selector = selectRebuildSelector(selectCnt);
            selectCnt = 1;
            break;
        }
    }
}
/**
 * Replaces the current {@link Selector} of this event loop with newly created {@link Selector}s to work
 * around the infamous epoll 100% CPU bug.
 * 新建一个selector来处理空轮询bug
 */
public void rebuildSelector() {
    if (!inEventLoop()) {
        execute(new Runnable() {
            @Override
            public void run() {
                rebuildSelector0();
            }
        });
        return;
    }
    rebuildSelector0();
}
private void rebuildSelector0() {
    final Selector oldSelector = selector;
    final SelectorTuple newSelectorTuple;
    //新建一个selector
    newSelectorTuple = openSelector();
    // 将旧的selector的channel悉数拿出来注册到新的selector上
    int nChannels = 0;
    for (SelectionKey key: oldSelector.keys()) {
        Object a = key.attachment();
        if (!key.isValid() || key.channel().keyFor(newSelectorTuple.unwrappedSelector) != null) {
            continue;
        }
        int interestOps = key.interestOps();
        key.cancel();
        SelectionKey newKey = key.channel().register(newSelectorTuple.unwrappedSelector, interestOps, a);
        if (a instanceof AbstractNioChannel) {
            // Update SelectionKey
            ((AbstractNioChannel) a).selectionKey = newKey;
        }
        nChannels ++;
    }
    selector = newSelectorTuple.selector;
    unwrappedSelector = newSelectorTuple.unwrappedSelector;
    // time to close the old selector as everything else is registered to the new one
    //关掉旧的selector
     oldSelector.close();
}

[参阅]

  • 李林峰的《Netty威望指南》