It's normal in the Unix world for server processes to be invoked by harness programs[109] such as inetd(8) in such a way that the server sees commands on standard input and ships responses to standard output; the harness program then takes care of ensuring that the server's stdin and stdout are connected to a specified TCP/IP service port. One benefit of this division of labor is that the harness program can act as a single security gatekeeper for all of the servers it launches.
One of the classic interface patterns is therefore a CLI server. This is a program which, when invoked in a foreground mode, has a simple CLI interface reading from standard input and writing to standard output. When backgrounded, the server detects this and connects its standard input and standard output to a specified TCP/IP service port.
In some variants of this pattern, the server backgrounds itself by default, and has to be told with a command-line switch when it should stay in foreground. This is a detail; the essential point is that most of the code neither knows nor cares whether it is running in foreground or a TCP/IP harness.
POP, IMAP, SMTP, and HTTP servers normally obey this pattern. It can be combined with any of the server/client patterns described earlier in this chapter. An HTTP server can also act as a harness program; the CGI scripts that supply most live content on the Web run in a special environment provided by the server where they can take input (form arguments) from standard input, and write the generated HTML that is their result to standard output.
Though this pattern is quite traditional, the term “CLI server” is my invention.
