I am working on a Windows (Microsoft Visual C++ 2005) application that uses several processes
running on different hosts in an intranet.
Processes communicate with each other using TCP/IP. Different processes can be on the
same host or on different hosts (i.e. the communication can be both within the same
host or between different hosts).
We have currently a bug that appears irregularly. The communication seems to work
for a while, then it stops working. Then it works again for some time.
When the communication does not work, we get an error (apparently while a process
was trying to send data). The call looks like this:
send(socket, (char *) data, (int) data_size, 0);
By inspecting the error code we get from
WSAGetLastError()
we see that it is an error 10054. Here is what I found in the Microsoft documentation
(see here):
WSAECONNRESET
10054
Connection reset by peer.
An existing connection was forcibly closed by the remote host. This normally
results if the peer application on the remote host is suddenly stopped, the
host is rebooted, the host or remote network interface is disabled, or the
remote host uses a hard close (see setsockopt for more information on the
SO_LINGER option on the remote socket). This error may also result if a
connection was broken due to keep-alive activity detecting a failure while
one or more operations are in progress. Operations that were in progress
fail with WSAENETRESET. Subsequent operations fail with WSAECONNRESET.
So, as far as I understand, the connection was interrupted by the receiving process.
In some cases this error is (AFAIK) correct: one process has terminated and
is therefore not reachable. In other cases both the sender and receiver are running
and logging activity, but they cannot communicate due to the above error (the error
is reported in the logs).
My questions.
- What does the SO_LINGER option mean?
- What is a keep-alive activity and how can it break a connection?
- How is it possible to avoid this problem or recover from it?
Regarding the last question. The first solution we tried (actually, it is rather a
workaround) was resending the message when the error occurs. Unfortunately, the
same error occurs over and over again for a while (a few minutes). So this is not
a solution.
At the moment we do not understand if we have a software problem or a configuration
issue: maybe we should check something in the windows registry?
One hypothesis was that the OS runs out of ephemeral ports (in case connections are
closed but ports are not released because of TcpTimedWaitDelay), but by analyzing
this issue we think that there should be plenty of them: the problem occurs even
if messages are not sent too frequently between processes. However, we still are not
100% sure that we can exclude this: can ephemeral ports get lost in some way (???)
Another detail that might help is that sending and receiving occurs in each process
concurrently in separate threads: are there any shared data structures in the
TCP/IP libraries that might get corrupted?
What is also very strange is that the problem occurs irregularly: communication works
OK for a few minutes, then it does not work for a few minutes, then it works again.
Thank you for any ideas and suggestions.
EDIT
Thanks for the hints confirming that the only possible explanation was a connection closed error. By further analysis of the problem, we found out that the server-side process of the connection had crashed / had been terminated and had been restarted. So there was a new server process running and listening on the correct port, but the client had not detected this and was still trying to use the old connection. We now have a mechanism to detect such situations and reset the connection on the client side.
That error means that the connection was closed by the
remote site. So you cannot do anything on your programm except to accept that the connection is broken.