完成端口，如何判断客户端异常断开？

[djfu]

有些情况通过GetQueuedCompletionStatus是无法得知客户端连接已经异常断开的，
比如：客户端网线突然被拔了。

这时如何判断客户端连接已经断开？
这时send还是可以成功的，尽管连接已经断了。

[vagrantisme]

BOOL bIORet = GetQueuedCompletionStatus(
hCompletionPort,
&dwIoSize,
(LPDWORD) &lpClientContext,
&lpOverlapped, INFINITE);
// Simulate workload (for debugging, to find possible reordering)
//Sleep(20);



// If Something whent wrong..
if (!bIORet)
{
DWORD dwIOError = GetLastError();
if(dwIOError != WAIT_TIMEOUT) // It was not an Time out event we wait for ever (INFINITE)
{
TRACE(_T("GetQueuedCompletionStatus errorCode: %i, %s\n"),WSAGetLastError(),pThis->ErrorCode2Text(dwIOError));
// if we have a pointer & This is not an shut down..
//if (lpClientContext!=NULL && pThis->m_bShutDown == false)
if (lpClientContext!=NULL)
{
/*
* ERROR_NETNAME_DELETED Happens when the communication socket
* is cancelled and you have pendling WSASend/WSARead that are not finished.
* Then the Pendling I/O (WSASend/WSARead etc..) is cancelled and we return with
* ERROR_NETNAME_DELETED..
*/
if(dwIOError==ERROR_NETNAME_DELETED)
{

//TRACE("ERROR_NETNAME_DELETED\r\n");
pThis->DisconnectClient(lpClientContext);
TRACE(">IOWORKER1 (%x)\r\n",lpClientContext);
pThis->ReleaseClientContext(lpClientContext); //Later Implementati

}else
{ // Should not get here if we do: disconnect the client and cleanup & report.

pThis->AppendLog(pThis->ErrorCode2Text(dwIOError));
pThis->DisconnectClient(lpClientContext);
TRACE(">IOWORKER2 (%x)\r\n",lpClientContext);
pThis->ReleaseClientContext(lpClientContext); //Should we do this ?
}
// Clear the buffer if returned.
pOverlapBuff=NULL;
if(lpOverlapped!=NULL)
pOverlapBuff=CONTAINING_RECORD(lpOverlapped, CIOCPBuffer, m_ol);
if(pOverlapBuff!=NULL)
pThis->ReleaseBuffer(pOverlapBuff);
continue;
}
// We shall never come here
// anyway this was an error and we should exit the worker thread
bError = true;
pThis->AppendLog(pThis->ErrorCode2Text(dwIOError));
pThis->AppendLog("IOWORKER KILLED BECAUSE OF ERROR IN GetQueuedCompletionStatus");

pOverlapBuff=NULL;
if(lpOverlapped!=NULL)
pOverlapBuff=CONTAINING_RECORD(lpOverlapped, CIOCPBuffer, m_ol);
if(pOverlapBuff!=NULL)
pThis->ReleaseBuffer(pOverlapBuff);
continue;
}
}

[vagrantisme]

			BOOL bIORet = GetQueuedCompletionStatus(
				hCompletionPort,
				&dwIoSize,
				(LPDWORD) &lpClientContext,
				&lpOverlapped, INFINITE);
			// Simulate workload (for debugging, to find possible reordering)
			//Sleep(20);



			// If Something whent wrong..
			if (!bIORet)  
			{
				DWORD dwIOError = GetLastError();
				if(dwIOError != WAIT_TIMEOUT) // It was not an Time out event we wait for ever (INFINITE) 
				{	
					TRACE(_T("GetQueuedCompletionStatus errorCode: %i, %s\n"),WSAGetLastError(),pThis->ErrorCode2Text(dwIOError));			
					// if we have a pointer & This is not an shut down.. 
					//if (lpClientContext!=NULL && pThis->m_bShutDown == false)
					if (lpClientContext!=NULL)
					{
						/*
						* ERROR_NETNAME_DELETED Happens when the communication socket
						* is cancelled and you have pendling WSASend/WSARead that are not finished. 
						* Then the Pendling I/O (WSASend/WSARead etc..) is cancelled and we return with 
						* ERROR_NETNAME_DELETED.. 
						*/
						if(dwIOError==ERROR_NETNAME_DELETED) 
						{

							//TRACE("ERROR_NETNAME_DELETED\r\n");
							pThis->DisconnectClient(lpClientContext); 
							TRACE(">IOWORKER1 (%x)\r\n",lpClientContext);
							pThis->ReleaseClientContext(lpClientContext); //Later Implementati

						}else
						{ // Should not get here if we do: disconnect the client and cleanup & report. 

							pThis->AppendLog(pThis->ErrorCode2Text(dwIOError)); 
							pThis->DisconnectClient(lpClientContext); 
							TRACE(">IOWORKER2 (%x)\r\n",lpClientContext);
							pThis->ReleaseClientContext(lpClientContext); //Should we do this ? 
						}
						// Clear the buffer if returned. 
						pOverlapBuff=NULL;
						if(lpOverlapped!=NULL)
							pOverlapBuff=CONTAINING_RECORD(lpOverlapped, CIOCPBuffer, m_ol);
						if(pOverlapBuff!=NULL)
							pThis->ReleaseBuffer(pOverlapBuff);			
						continue;
					}
					// We shall never come here  
					// anyway this was an error and we should exit the worker thread
					bError = true;
					pThis->AppendLog(pThis->ErrorCode2Text(dwIOError)); 
					pThis->AppendLog("IOWORKER KILLED BECAUSE OF ERROR IN GetQueuedCompletionStatus");

					pOverlapBuff=NULL;
					if(lpOverlapped!=NULL)
						pOverlapBuff=CONTAINING_RECORD(lpOverlapped, CIOCPBuffer, m_ol);
					if(pOverlapBuff!=NULL)
						pThis->ReleaseBuffer(pOverlapBuff);
					continue;
				}
			}

[djfu]

楼上的, 你这个不行，
我说了, 根据GetQueuedCompletionStatus不能得出网络是否异常掉线的结果。

我把客户端网线拔了，服务端无法通过GetQueuedCompletionStatus得出是否客户端已经掉线的结论。

[reality]

自己检测

[caohongtai]

ping 一下（用程序ping）

[wanglovec]

心跳检测

[百事烟]

心跳

[总哈哈]

心跳和定时检测一个意思，两者都行！

[coast1]

心跳检测

[stivenjia]

GetQueuedCompletionStatus之后再select一下

[WinEggDrop]

加入超时检测机制.某一段时间内(例如5分钟),连接都没有数据交换,定义为超时,不论是不是异常断开还是挂着没数据,一律断开.

[saiyaman5]

学习下

[路人乙2019]

用心跳是好,不过客户端太多了势必会加重服务端的工作.

[无间兔]

设置WSAIoctl的保活项SIO_KEEPALIVE_VALS，注意SIO_KEEPALIVE_VALS项的输入参数是个结构体struct tcp_keepalive {
u_long onoff; 非零数启用保活并使后续参数有效
u_long keepalivetime; 首次保活探测包发送前的空闭时间(ms)
u_long keepaliveinterval; 发送保活探测包的间隔时间(ms)
};使用该结构要包含mstcpip.h文件。
WSAIoctl还有个好处就是只针对当前socket进行设置。