scgi: Web Server HTTP SCGI and CGI

3:3

top ← prev up next →

scgi: Web Server HTTP SCGI and CGI

Neil Van Dyke

License: LGPLv3 Web: http://www.neilvandyke.org/racket/scgi/

package: scgi

1 Introduction

The scgi library implements fast Web CGI using the SCGI protocol. This library is used in conjunction with an HTTP server supporting SCGI, such as Apache HTTP Server with the mod_scgi module.

The scgi library also supports running as normal Web CGI without any change to the source code of the app, such as during development of an application intended to be deployed using SCGI. This also gives flexibility in deployment, allowing a system administrator to switch between either mode just by editing the HTTP server configuration.

The SCGI protocol was specified by Neil Schemenauer in “SCGI: A Simple Common Gateway Interface alternative,” dated 2008-06-23.

An example use of this library:

(require scgi)

(cgi #:startup  (lambda ()
                  (printf "Startup for cgi-type ~S..."
                          (cgi-type)))

     #:request  (lambda ()
                  (display "Content-type: text/html\r\n\r\n")
                  (display "<p>Hello, world!</p>")
                  (printf "<p>Your IP address is: ~S</p>"
                          (cgi-remote-addr)))

     #:shutdown (lambda ()
                  (printf "Shutdown for cgi-type ~S..."
                          (cgi-type))))

The procedure provided to the #:request is the one that gets called for each request. Within that procedure, is the CGI request context, in which procedures like cgi-remote-addr may be called to get information about that particular request. All three procedures are called within CGI context, in which procedures concerning the CGI mechanism separate from individual requests, such as cgi-type may be called. Many programs will need to provide only the #:request procedure.

2 Interface

procedure
(cgi [ #:startup startup-proc]
#:request request-proc
[ #:shutdown shutdown-proc
#:scgi-hostname scgi-hostname
#:scgi-max-allow-wait scgi-max-allow-wait
#:scgi-portnum scgi-portnum
#:reuse-scgi-port? reuse-scgi-port?]) → void?
  startup-proc : (-> any) = void
  request-proc : (-> any)
  shutdown-proc : (-> any) = void
  scgi-hostname : (or/c string? #f) = "127.0.0.1"
  scgi-max-allow-wait : exact-nonnegative-integer? = 4
   scgi-portnum :
(and/c exact-nonnegative-integer?
       (integer-in 0 65535))
= 4000
  reuse-scgi-port? : boolean? = #t

Implement CGI. Normal CGI is used if the REQUEST_URI environment variable is defined (which suggests that the code is being called in a CGI context); otherwise, SCGI is used.

startup-proc is a thunk that is evaluated once (before listener starts). request-proc is evaluated once for each request (which, in normal CGI, is once). shutdown-proc is evaluated once, as processing of all CGI requests has finished.

For evaluation of request-proc, the default input and output ports are as with normal CGI, regardless of whether normal CGI or SCGI is in use.

This procedure also accepts a few optional keyword arguments, all of which apply to SCGI mode only. scgi-hostname is the hostname or IP address (as a string) for the interface on which to listen. scgi-portnum is the TCP port number on that interface. scgi-max-allow-wait is the maximum number of unaccepted connections to permit waiting. reuse-scgi-port? is whether or not to reuse the TCP port number, such as if a previous server exited and the port is in a TIME_WAIT state.

procedure
(cgi-content-length) → exact-nonnegative-integer?

In a CGI request context, returns the CGI content length – the number of bytes that can be read from the default input port – as integer.

procedure
(make-cgi-variable-proc proc-name-sym
name-bytes) → (-> string?)
proc-name-sym : symbol?
name-bytes : bytes?

Produces a procedure for getting a CGI environment variable value as a string. Works whether in normal CGI or SCGI. This is useful for accessing non-standard variables, such as might be provided by an unusual Apache module. Argument proc-name-sym is a symbol for the name of the procedure, which is used in error reporting. Argument name-bytes is the name of the environment variable, as a byte string.

For example, were the cgi-remote-user procedure not already defined, it could be defined as:

(define cgi-remote-user
(make-cgi-variable-proc 'cgi-remote-user
#"REMOTE_USER"))

procedure
(cgi-content-type) → string?
(cgi-document-root) → string?
(cgi-http-cookie) → string?
(cgi-http-host) → string?
(cgi-http-referer) → string?
(cgi-http-user-agent) → string?
(cgi-https) → string?
(cgi-path-info) → string?
(cgi-path-translated) → string?
(cgi-query-string) → string?
(cgi-remote-addr) → string?
(cgi-remote-host) → string?
(cgi-remote-user) → string?
(cgi-request-method) → string?
(cgi-request-uri) → string?
(cgi-script-name) → string?
(cgi-server-name) → string?
(cgi-server-port) → string?

In a CGI request context, returns the corresponding CGI value as a string. Note that cgi-content-length is not in this list, and it returns a number rather than a string.

procedure
(scgi-variables) → (list/c (cons/c bytes? bytes?))

When called in SCGI mode, this procedure yields an alist of SCGI variables with both the key and value of each pair being byte strings. Calling this procedure in normal CGI mode is an error.

Note that normally you will not need to use this procedure, and will instead use procedures like cgi-request-uri, which work in both SCGI and normal CGI modes.

procedure
(cgi-type) → (or/c 'normal 'scgi)

Returns a symbol indicating the CGI type: 'normal or 'scgi. Behavior outside of the procedures of the cgi procedure is undefined.

procedure
(cgi-request-id) → any/c

In CGI request context, yields a printable identifying object for the current request that is unique at least for the current requests being handled. This identifying object is intended to be used in debugging messages.

procedure
(stop-cgi-service-immediately) → void?

Stops processing all CGI requests. This works only within the #:request procedure of the cgi procedure.

3 Apache mod_scgi

3.1 Apache mod_scgi Configuration

Apache HTTP Server is one way to run SCGI, though not the only way. Note that your Apache installation might not have the mod_scgi module installed or enabled by default. If you happen to be running Debian GNU/Linux, this module can be installed via the Debian package libapache2-mod-scgi.

Once you’ve installed mod_scgi, you need some standard SCGI directives to end up in your Apache config files, whether you accomplish that by editing config files manually, making symbolic links in a "mods-enabled" directory, or clicking in a GUI.

For example, the following loads mod_scgi, maps URL paths under "/mypath" to the SCGI server on the local machine at the standard SCGI TCP port, and sets a 60-second timeout for the SCGI server to respond to a request before Apache drops the connection:

LoadModule scgi_module /usr/lib/apache2/modules/mod_scgi.so

SCGIMount /mypath 127.0.0.1:4000

SCGIServerTimeout 60

There are additional mod_scgi Apache config directives, including SCGIHandler and SCGIServer.

3.2 Apache mod_scgi Troubleshooting

This section has some troubleshooting tips. Currently, these come from use with mod_scgi atop Apache 2.2.9 atop Debian GNU/Linux.

Racket error “tcp-write: error writing (Broken pipe; errno=32)” or “tcp-write: error writing (Connection reset by peer; errno=104)” is likely due to the HTTP request having been dropped by the HTTP client (e.g., user stops a page load in their browser before page finishes loading) or by Apache hitting SCGIServerTimeout for the request. Note that buffered I/O means that you won’t necessarily get this error even if the request is aborted this way.
Apache error log entry “Premature end of script headers: PATH” followed by “(500)Unknown error 500: scgi: Unknown error 500: error reading response headers” can mean that SCGIServerTimeout was hit before any HTTP headers from the SCGI request handler were started or completed. Note that buffered I/O can mean that some of the Racket code of the handler wrote some text, but it was not yet flushed to the SCGI client.
Apache error log entry “(70007)The timeout specified has expired: ap_content_length_filter: apr_bucket_read() failed” followed by “(70007)The timeout specified has expired: scgi: The timeout specified has expired: ap_pass_brigade()” can mean that SCGIServerTimeout was hit after HTTP headers from the request handler had been received by the SCGI client.
Apache error log entry “(32)Broken pipe: scgi: Broken pipe: error sending request body” might be due to a request handler finished without consuming all of the HTTP POST data.
Apache error log entry “(103)Software caused connection abort: scgi: Software caused connection abort: ap_pass_brigade()” is another one that can be caused by the HTTP client dropping the connection before the handler finishes.

4 History

Version 3:3 — 2016-03-02
- Tweaked info.rkt, filenames.
Version 3:2 — 2016-02-25
- Fixed deps.
Version 3:1 — 2016-02-25
- Commented-out code in "stress1-server.rkt" as the most expedient way to get new package system build to pass.
Version 3:0 — 2016-02-21
- Moving from PLaneT to new package system.
Version 2:4 — 2015-04-02
- Documentation tweaks.
Version 2:3 — 2014-11-23
- Removed file sample-1.rkt from package, to reduce package dependencies. (Suggested by Matt Gushee.)
Version 2:2 — 2013-04-19
- Now accommodate SCGI variable names generated by Mellon authentication.
Version 2:1 — 2013-04-12
- Now doesn’t barf on headers like proxy-nokeepalive. (Thanks to Erik for bug report and diagnosis.)
Version 2:0 — 2012-06-12
- The three main procedures arguments to the cgi procedure now have keywords. What are now the #:startup and #:shutdown arguments are now also optional.
- Converted to McFly.
- Changed internal-only identifiers from having a % prefix to a %scgi: prefix, to help disambiguate when identifiers appear in error messages in large systems.
- Documentation changes.
- Improvement to the internal debugging macro.
Version 0.6 — Version 1:5 — 2011-08-22
- Small documentation fixes.
Version 0.5 — Version 1:4 — 2011-05-16
- For closer to backward compatibility with PLT Scheme 4.x, changed references to racket modules to scheme.
- More documentation.
Version 0.4 — Version 1:3 — 2011-05-16
- Added #:reuse-scgi-port? argument to procedure cgi. Added several new CGI environment variable procedures.
- Added make-cgi-variable-proc.
- The cgi procedure now uses tcp-accept/enable-break rather than tcp-accept.
- Added more log-debug.
- Added documentation about mod_scgi configuration, troubleshooting,and the keyword arguments to the cgi procedure.
- Removed documentation placeholders for end-cgi-request for now.
- Various additional quality assurance testing has been done, and more is the works.
Version 0.3 — Version 1:2 — 2010-11-14
- Added cgi-http-user-agent.
Version 0.2 — Version 1:1 — 2010-10-11
- Documentation changes to reflect that it is successfully in use in a real system, and some work remains.
Version 0.1 — Version 1:0 — 2010-05-25
- Initial release. Preliminary.

5 Legal

Copyright 2010–2013, 2015, 2016 Neil Van Dyke. This program is Free Software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but without any warranty; without even the implied warranty of merchantability or fitness for a particular purpose. See http://www.gnu.org/licenses/ for details. For other licenses and consulting, please contact the author.

top ← prev up next →