问题
Scenario:
I have a client and server written using boost::asio 1.63. Generally the connection and communication part works well and great.
I have written a Watchdog on both sides which send dummy packets to peers in an interval of 2 seconds each. The objective of the watchdog is that the concerned peer reports a connection error if it does not receive the dummy packet it is expecting in the next 2 seconds. This is even more important for me because it might happen the 2 peers are not transacting packets for any user purpose but each of them is required to report a connection error if any of the peer goes down. The peer can go down even because of a kernel crash in which case it would not be possible for that peer to send a message. This is a classic problem of course which exists even beyond asio and TCP.
My Watchdog works perfectly well. No issues at all.
But, recently I read about the keep_alive feature in sockets. I tried out the following code and seems like I can a property called keep_alive on the TCP socket by getting the native handle to the socket from within my code using boost::asio.
boost::asio::io_service ioService;
boost::asio::ip::tcp::socket mySocket(ioService);
int on = 1;
int delay = 120;
setsockopt(mySocket.native_handle(), SOL_SOCKET, SO_KEEPALIVE, &on, sizeof(on));
setsockopt(mySocket.native_handle(), IPPROTO_TCP, TCP_KEEPALIVE, &delay, sizeof(delay));
Question:
Above code compiles well on macOS, Linux and iOS. That looks great. But, how do I benefit from this? Does this give me a callback or event when the peer goes down? Does this free me up from writting the Watchdog that I described above?
I have used boost::asio::async_connect to connect to the peer. Can I get a callback to my connectionHandler when the perr goes down after the defined timeout interval?
Having set the keep_alive options, how do I then get to know that my peer is not responding anymore?
回答1:
If the disconnetion was detected when an async operation is pending, your socket's completion handler will be invoked with the appropriate error code.
The problem is that TCP keep_alive option doesn't not always detect disconnects.
In general, there is no reliable way to detect sudden disconnection, other than by implementing application-level ping/heartbeat.
You can also see this thread.
来源：https://stackoverflow.com/questions/58689857/how-to-use-tcp-keep-alive-property-to-get-notified-on-the-event-of-a-unresponsiv