package async_unix

Sockets have a phantom type parameter that tracks the state of the socket in order to eliminate certain errors in which socket functions are called in the wrong order. Initially, a socket is `Unconnected. As various socket functions are called, they return a socket with a new phantom state. Here is a chart of the allowed state transitions.

        Unconnected ---connect--> Active
        |
        | ---bind--> Bound ---listen--> Passive ---accept---> Active
                     |
                     | ---connect--> Active

bind socket addr sets close_on_exec for the fd of socket.

bind_inet socket addr is just like bind but is restricted to Inet.t addresses and is therefore guaranteed not to block.

accept_at_most is like accept, but will return up to limit connections before yielding, where limit >= 1. accept_at_most first waits for one connection and then attempts to retrieve up to limit connections through non-blocking Unix.accept calls. If a call to Unix.accept would block before limit is reached, accept_at_most returns the connections retrieved thus far.

Servers that must service a large number of connections tend to observe a stall in connection accept rates when under heavy load. Increasing limit will ameliorate this effect, and increase accept rates and overall throughput of the server at the cost of increased contention for resources amongst connections.

For details, see:

         Acceptable strategies for improving web server performance
         Brecht, Pariag, and Gammo.  USENIX ATEC '04

bind_to_interface_exn t (`Interface_name "eth0") restricts messages from being received or sent on interfaces other than eth0. See Linux_ext.bind_to_interface.

Typically, one would use this function for very specific non-multicast requirements. For similar functionality when using multicast, see Core_unix.mcast_set_ifname.