connect() to localhost non-blocking.
Quoted 14 lines
Quoted 12 lines · Patrick Mau
I wrote a little test program to do some
poll()benchmarks. I changed the host address to localhost and observed thatconnect()always returns EINPROGRESS when used with non-blocking sockets.Quoted 1 line
From the man page:
EINPROGRESS
The socket is non-blocking and the connection cannot be completed immediately. It is possible to select(2) or poll(2) for completion by selecting the socket for writing.
Quoted 8 lines
Quoted 5 lines
So my question is:
What is meant by 'cannot be completed immediately' ? I thought that connections to localhost would complete without any delay when the application listens ?
Probably the accept()ing process hasn't been scheduled yet. EINPROGRESS is a perfectly reasonable response in such a case.
You have to be prepared to handle both immediate and delayed connection, especially if you want to be portable. (Solaris behaves a bit differently than Linux in this regard.) See
http://www.kegel.com/dkftpbench/dkftpbench-0.37/ftp_client_pipe.cc
for an example of how to handle nonblocking connects more or less portably. (You have to wade through quite a bit of code, tabstops 4, to find all the connect-handling stuff -- sorry.)
- Dan
Quoted 2 lines · Dan Kegel
Quoted 2 lines
Patrick Mau <[email protected]> writes:
[Snip]
Quoted 8 lines
You have to be prepared to handle both immediate and delayed connection, especially if you want to be portable. (Solaris behaves a bit differently than Linux in this regard.) See
http://www.kegel.com/dkftpbench/dkftpbench-0.37/ftp_client_pipe.cc
for an example of how to handle nonblocking connects more or less portably. (You have to wade through quite a bit of code, tabstops 4, to find all the connect-handling stuff -- sorry.)
- Dan
Hallo Dan,
thnaks for the pointer and your remarks, but I really wondered if linux _never_ immediatly accept()'s connections.
I never saw a connect() call returning 0 (connected) on a non-blocking socket. Always EINPROGRESS. I guess I have to look at the kernel to see scheduling effects ...
thanks,
Patrick