debian-quixote3/doc/web-server.txt

252 lines
9.3 KiB
Plaintext

Web Server Configuration for Quixote
====================================
For a simple Quixote installation, there are two things you have to get
right:
* installation of the Quixote modules to Python's library (the
trick here is that the ``quixote`` package must be visible to the user
that CGI scripts run as, not necessarily to you as an interactive
command-line user)
* configuration of your web server to run Quixote driver scripts
This document is concerned with the second of these.
Which web servers?
------------------
We are only familiar with Apache, and we develop Quixote for use under
Apache. However, Quixote doesn't rely on any Apache-specific tricks;
if you can execute CGI scripts, then you can run Quixote applications
(although they'll run a lot faster with mod_scgi or FastCGI). If you
can redirect arbitrary URLs to a CGI script and preserve parts of the
URL as an add-on to the script name (with ``PATH_INFO``), then you can
run Quixote applications in the ideal manner, ie. with superfluous
implementation details hidden from the user.
Which operating systems?
------------------------
We are mainly familiar with Unix, and develop and deploy Quixote under
Linux. However, we've had several reports of people using Quixote under
Windows, more-or-less successfully. There are still a few Unix-isms in
the code, but they are being rooted out in favor of portability.
Remember that your system is only as secure as its weakest link.
Quixote can't help you write secure web applications on an inherently
insecure operating system.
Basic CGI configuration
-----------------------
Throughout this document, I'm going to assume that:
* CGI scripts live in the ``/www/cgi-bin`` directory of your web server,
and have the extension ``.cgi``
* HTTP requests for ``/cgi-bin/foo.cgi`` will result in the execution
of ``/www/cgi-bin/foo.cgi`` (for various values of ``foo``)
* if the web server is instructed to serve an executable file
``bar.cgi``, the file is treated as a CGI script
With Apache, these configuration directives will do the trick::
AddHandler cgi-script .cgi
ScriptAlias /cgi-bin/ /www/cgi-bin/
Consult the Apache documentation for other ways of configuring CGI
script execution.
For other web servers, consult your server's documentation.
Installing driver scripts
-------------------------
Given the above configuration, installing a Quixote driver script is the
same as installing any other CGI script: copy it to ``/www/cgi-bin`` (or
whatever). To install the Quixote demo's cgi driver script::
cp -p server/cgi_server.py /www/cgi-bin/demo.cgi
(The ``-p`` option ensures that ``cp`` preserves the file mode, so that
it remains executable.)
URL rewriting
-------------
With the above configuration, users need to use URLs like ::
http://www.example.com/cgi-bin/demo.cgi
to access the Quixote demo (or other Quixote applications installed in
the same way). This works, but it's ugly and unnecessarily exposes
implementation details.
In our view, it's preferable to give each Quixote application its own
chunk of URL-space -- a "virtual directory" if you like. For example,
you might want ::
http://www.example.com/qdemo
to handle the Quixote demo.
With Apache, this is quite easy, as long as mod_rewrite is compiled,
loaded, and enabled. (Building and loading Apache modules is beyond the
scope of this document; consult the Apache documentation.)
To enable the rewrite engine, use the ::
RewriteEngine on
directive. If you have virtual hosts, make sure to repeat this for each
``<VirtualHost>`` section of your config file.
The rewrite rule to use in this case is ::
RewriteRule ^/qdemo(/.*) /www/cgi-bin/demo.cgi$1 [last]
This is *not* a redirect; this is all handled with one HTTP
request/response cycle, and the user never sees ``/cgi-bin/demo.cgi`` in
a URL.
Note that requests for ``/qdemo/`` and ``/qdemo`` are *not* the same; in
particular, with the above rewrite rule, the former will succeed and the
latter will not. (Look at the regex again if you don't believe me:
``/qdemo`` doesn't match the regex, so ``demo.cgi`` is never invoked.)
The solution for ``/qdemo`` is the same as if it corresponded to a
directory in your document tree: redirect it to ``/qdemo/``. Apache
(and, presumably, other web servers) does this automatically for "real"
directories; however, ``/qdemo/`` is just a directory-like chunk of
URL-space, so either you or Quixote have to take care of the redirect.
It's almost certainly faster for you to take care of it in the web
server's configuration. With Apache, simply insert this directive
*before* the above rewrite rule::
RewriteRule ^/qdemo$ /qdemo/ [redirect=permanent]
If, for some reason, you are unwilling or unable to instruct your web
server to perform this redirection, Quixote will do it for you.
However, you have to make sure that the ``/qdemo`` URL is handled by
Quixote. Change the rewrite rule to::
RewriteRule ^/qdemo(/.*)?$ /www/cgi-bin/demo.cgi$1 [last]
Now a request for ``/qdemo`` will be handled by Quixote, and it will
generate a redirect to ``/qdemo/``. If you're using a CGI driver
script, this will be painfully slow, but it will work.
For redirecting and rewriting URLs with other web servers, consult your
server's documentation.
Long-running processes
----------------------
For serious web applications, CGI is unacceptably slow. For a CGI-based
Quixote application, you have to start a Python interpreter, load the
Quixote modules, and load your application's modules before you can
start working. For sophisticated, database-backed applications, you'll
probably have to open a new database connection as well for every hit.
Small wonder so many high-performance alternatives to CGI exist. (The
main advantages of CGI are that it is widely supported and easy to
develop with. Even for large Quixote applications, running in CGI mode
is nice in development because you don't have to kill a long-running
driver script every time the code changes.) Quixote includes support
for mod_scgi and FastCGI.
mod_scgi configuration
----------------------
SCGI is a CGI replacement written by Neil Schemenauer, one of
Quixote's developers, and is similar to FastCGI but is designed to be
easier to implement. mod_scgi simply forwards requests to an
already-running SCGI server on a different TCP port, and doesn't try
to start or stop processes, leaving that up to the SCGI server.
The SCGI code is available from <http://python.ca/scgi/>.
The quixote.server.scgi_server module is a script that
publishes the demo quixote application via SCGI. You can use
it for your application by importing it and calling the ``run()``
function with arguments to run your application, on the port
you choose. Here is an example::
#!/usr/bin/python
from quixote.server.scgi_server import run
from quixote.publish import Publisher
from mymodule import MyRootDirectory
def create_my_publisher():
return Publisher(MyRootDirectory())
run(create_my_publisher, port=3001)
The following Apache directive will direct requests to an SCGI server
running on port 3001::
SCGIMount / 127.0.0.1:3001
SCGI through CGI
----------------
Recent releases of the scgi package include cgi2scgi.c, a small program
that offers an extremely convenient way to take advantage of SCGI using
Apache or any web server that supports CGI. To use it, compile the
cgi2scgi.c and install the compiled program as usual for your
webserver. The default SCGI port is 3000, but you can change that
by adding ``-DPORT=3001`` (for example) to your compile command.
Although this method requires a new process to be launched for each
request, the process is small and fast, so the performance is
acceptable for many applications.
FastCGI configuration
---------------------
If your web server supports FastCGI, you can significantly speed up your
Quixote applications with a simple change to your configuration. You
don't have to change your code at all (unless it makes assumptions about
how many requests are handled by each process). (See
http://www.fastcgi.com/ for more information on FastCGI.)
To use FastCGI with Apache, you'll need to download mod_fastcgi from
http://www.fastcgi.com/ and add it to your Apache installation.
Configuring a FastCGI driver script is best done after reading the fine
documentation for mod_fastcgi at
http://www.fastcgi.com/mod_fastcgi/docs/mod_fastcgi.html
However, if you just want to try it with the Quixote demo to see if it
works, add this directive to your Apache configuration::
AddHandler fastcgi-script .fcgi
and copy server/fastcgi_server.py to demo.fcgi. If you're using a URL
rewrite to map requests for (eg.) ``/qdemo`` to
``/www/cgi-bin/demo.cgi``, be sure to change the rewrite -- it should
now point to ``/www/cgi-bin/demo.fcgi``.
After the first access to ``demo.fcgi`` (or ``/qdemo/`` with the
modified rewrite rule), the demo should be noticeably faster. You
should also see a ``demo.fcgi`` process running if you do ``ps -le``
(``ps -aux`` on BSD-ish systems, or maybe ``ps aux``). (On my 800 MHz
Athlon machine, there are slight but perceptible delays navigating the
Quixote demo in CGI mode. In FastCGI mode, the delay between pages is
no longer perceptible -- navigation is instantaneous.) The larger your
application is, the more code it loads, and the more work it does at
startup, the bigger a win FastCGI will be for you (in comparison to CGI).