Tuesday, May 24, 2016

TCP flow control with non-blocking sends: EAGAIN

So, let's say you're sending data on a TCP socket faster than the receiver can unload it. The socket buffers fill up. Then what happens? The send call returns fewer bytes sent than were requested. Everybody knows that. (Interestingly, http://linux.die.net/man/2/send does not mention this behavior, but I see certainly it during testing.)

But what if the previous send exactly filled the buffer so that your next send can't put *any* bytes in? Does send return zero? Apparently not. It returns -1 with an errno of EAGAIN or EWOULDBLOCK (also verified by testing).  If I ever knew this, I forgot it till today.

Finally, something I did already know, but rarely include in my code: see this excerpt from http://linux.die.net/man/2/send:
The socket is marked nonblocking and the requested operation would block. POSIX.1-2001 allows either error to be returned for this case, and does not require these constants to have the same value, so a portable application should check for both possibilities.

No comments: