What does it mean when GetQueuedCompletionStatus return ERROR_SEM_TIMEOUT?

A customer asked for assistance interpreting a failure of the Get­Queued­Completion­Status function.

We are observing that Get­Queued­Completion­Status is intermittently behaving as follows:

• The handle is a SOCKET.
• The function returns FALSE.
• lpOverlapped != NULL.
• Get­Last­Error reports ERROR_SEM_TIMEOUT: "The semaphore timeout period has expired."

That's all the information we have in our log files. We don't know the value of number­Of­Bytes or completion­Key, sorry.

We realize that this is a rather vague question, but when this problem hits our machines, it causes our internal logic to go into a reset state since it doesn't know what the error means or how to recover. Resetting is expensive, and we would prefer to handle this error in a less drastic manner, if only we knew what it meant.

The error code ERROR_SEM_TIMEOUT is a rather bad translation of the underlying status code STATUS_IO_TIMEOUT, which is much more meaningful. It means that the I/O operation timed out.

Colleagues of mine from the networking team chimed in with additional information:

A common source of this error with TCP sockets is that the maximum retransmission count and timeout have been reached on a bad (or broken) link.

If you know that the handle is a socket, then you can use WSA­Get­Overlapped­Result on the lpOverlapped that got returned. Winsock will convert the status code to something more Winsocky. In this case, it would have given you WSA­ETIMED­OUT, which makes it clearer what happened.