httpd-arch - The TclHttpd Web Server System
This document is intended for developers using TclHttpd as the basis for specialized web applications. In other words, for people who have to perform heavy customizations of the basic web server.
To this end the general architecture of the web server is explained below, i.e. the packages it consists of and their relationships, together with all customization points which can be used to alter the servers behaviour.
Where necessary references into the package manuals are provided for further study.
The standard application of the tclhttpd paackages consists of two scripts and a configuration file:
The main script. It loads the basic packages and does some general initializations. Parts, especially the initialization can be customized.
A per-thread tcl script. This is used by the main script to initialize all worker threads of the web server. Note: This is used even for a non-threaded server. In that case the main thread is also the lone worker thread and initialized through this script. This script may need heavy modification, especially as this is the place where the standard domains of the server are set up.
This is the configuration file. It can be used to set the port, user ID, and other values, overriding the default values. You can configure additional features such as log file location, and more, by editing the configuration file. There is an explanation about each option, so you can make a copy of the configuration file and try out new settings.
Developers planning to extend Tcl Httpd with their own code, may need to add initialization code to "bin/httpd.tcl" and "bin/httpdthread.tcl". This code is typically a "package require" to load the required special modules and one or two calls to initialize it. For example, this code in "bin/httpdthread.tcl" enables a /debug URL implementation that lets users examine the state of the server.
package require httpd::debug
Debug_Url /debug Debug
Note: It is not strictly necessary to modify "bin/httpd.tcl" or "bin/httpdthread.tcl". Instead, you can provide a custom code directory and put additional Tcl source files there and instruct the server to load that code on startup. Use the "-library" command line option to specify the custom code directory.
The web server should have access to any Tcl package installed along with your Tcl installation. Consult the on-line HTML documentation for a more indepth discussion of programming the server.
The core of the system, the protocol engine accepting and reading http requests, and the database which dispatches requests to handlers based on where the url is in the server tree.
This is the first level where customization is possible. Create special domains and domain implementations which are then hooked into the server by modifying the application (initialization) scripts.
However there are also a number of pre-defined domains which can be used as well. These are described below, in the following items.
A standard domain implementation. These two packages implement the serving of urls from the filesystem. Actually it can manage multiple domains, each associated with a different directory in the filesystem.
This is the most important domain for an unmodified webserver as this is the part to serve all static content of a website. Note that the standard initialization uses this domain to associate a directory with the root of the url tree for the server. If a special server does not serve any static content at all deactivation of this module is not enough. It is also necessary to associate some other domain with the root.
The customization feature in this part of the system are type specific handlers. This means that the package determines the mime-type of the file associated with the requested url and then calls a type specific tcl command if such can be found. This allows other packages to modify the way the contents of a file are returned. Like for example running some substitution process to insert dynamic parts of the page. Note that if there is no type-specific command for a mime-type then the domain will simply return the contents of the requested file as they are.
Comment: It is unclear why doc and dirlist are separate packages. They are interdependent (i.e. each requires the other) and actually handle only two different aspects of the same thing (files and directories).
Answer: In theory your application could provide a different directory listing module. But, really, Brent just likes smaller packages.
A standard domain. Handles the serving of urls through external applications using the CGI interface. Internally this domain interacts with doc (s.a.). It has no customization features beyond the ability of the configurator to place the domain prefix at will (default is "/cgi-bin/"). If a webserver does not provide cgi applications this domain can be deactivated at leisure.
A standard domain. Actually a meta domain implementation. Its sole purpose is to provide a simpler interface for the implementation of special url trees than what is provided by the core package http::url. In other words, this package does not register a domain with very specific functionality, like httpd::cgi (and httpd::doc), but provides a framework for the implementation of domains directly through tcl commands.
Domains implemented based on the interface provided here are called application direct. Note that they are a restricted to a finite set of urls. This is because each url to be served needs a matching tcl command generating the contents. In other words, an unbounded number of urls requires an unbounded number of tcl commands. This means that for unbounded url trees an application direct domain is not the correct choice. An individual domain implementation is required.
The system comes with a number of packages which implement standard application direct domains. An example is the status package which delivers server status information.
http, tclhttpd, web server
Copyright © 2003 Andreas Kupries <[email protected]>