Does `select` handles multiple endpoints or multiple sockets? - network-programming

Question

I am new to network programming, and I have some confusion with select function.
For a server program, we need to first create a fd to socket endpoint (server's ip and port without client's ip and port) with socket, bind and listen, then if there is a TCP connection to this socket endpoint, then accept returns the fd to the socket (server's ip, port and client's ip, port). Then we use recv on this socket's fd, and if there is no data to receive, the recv will block (for blocking socket).
I learned that select is used to handle non-blocking multiple connections. But which connection level does it handles? Does it handles multiple socket endpoints, or handles multiple sockets of one single socket endpoint?
For a normal server program, I think the socket endpoint is always single, and there are maybe hundreds of sockets connected to this endpoint. So I think handling multiple socket endpoints may be less useful to handling multiple sockets. But when talking about IO multiplexing, I find that many articles seems talking about handling multiple socket endpoints. While for handling multiple sockets to a single socket, I can't find a way to get all sockets, and put them to select's set of fds, since accept only accepts one sockets a time.

Answer 1

"Does it handles multiple socket endpoints or handles multiple sockets of one single socket endpoint?" -- There is no such thing as multiple sockets of a single socket endpoint. Every socket is an endpoint to the network communication. It just happens that the piece of code which deals with that socket might be different from the others. Consider the following socket descriptors:
int sock_acpt = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
listen(sock_acpt, 5);
int sock_cli = accept(sock_apt, ....);
Both the socket descriptors sock_acpt and sock_cli are the endpoints of different communications. After setting sock_acpt in passive mode by calling listen(), the socket just listens for TCP connection, and the server's TCP stack will manage any data that appears on that socket (most probably TCP Handshakes). While sock_cli is the end of an already established connection, and in general, the data on that socket is managed by the user-level application.
Now coming to select(), it is an IO Multiplexer, not a Network IO multiplexer. So any descriptor which can be viewed as an IO endpoint can be used with the select(). Referring to our socket descriptors sock_acpt and sock_cli, both are IO endpoints of different communications, so you can use both of them with select(). You generally do something like below
for ( ; ; ) {
fd_set rd_set;
FD_ZERO(&rd_set);
FD_SET(sock_acpt, &rd_set);
FD_SET(sock_cli, &rd_set);
if (select(sock_acpt > sock_cli ? sock_acpt + 1 : sock_cli + 1, \
&rd_set, NULL, NULL, NULL) <= 0) {
continue;
}
if (FD_ISSET(sock_acpt, &rd_set)) {
// Accept new connection
// accept(sock_acpt, ...);
}
if (FD_ISSET(sock_cli, &rd_set)) {
// Read from the socket
// read(sock_cli, ...);
}
}
But using select() is not limited to sockets, you can use with the file IO (fileno(stdin)), with the signal IO (signalfd()) and any other which can be viewed as IO endpoint.