Linux Zone Feature: The Twisted Framework

Part Three, Stateful Web Servers and Templating


David Mertz, Ph.D.
Selector, Gnosis Software, Inc.
June, 2003

In the previous installment of this series, David looked at some higher-level techniques for writing Web services, including serving dynamic pages using the .rpy extension. In this article, he moves on to look at dynamic Web serving, and how to generate dynamic Web pages using the Woven application for templating pages.

Interacting With A Web Browser

In the last installment of this series, we looked at dynamic web pages served by Twisted using the .rpy extension. But those initial versions of a weblog server were only minimally dynamic. HTML tags were used to force a page to refresh periodically, and upon each refresh a bit of calculation was done to determine the relevant recent hits. But there was no user configuration aspect to the server.

The first thing we will look at in this installment, therefore, is how to configure user interaction into the same basic dynamic page framework we looked at before. But even before that, let me include a quick reminder of how to launch a Twisted web server for readers who may not have seen the prior installments.

Creating an "pickled application" is usually the best approach, and it can be done purely with command line options. You do not have to do it this way. If you like, you are free to include some extra capabilities inside the basic webserver (such as maintaining persistent data across users and sessions), but it is not necessary to write any custom code. Creating the pickled application looks something like:

mktap web --path ~/twisted/www --port 8080

Launching it consists of:

twistd -f web.tap

That's it. Any HTML or .rpy files that happen to be in the ~/twisted/www base directory (or subdirectories) will be served to clients on port 8080. Actually, you can serve whatever file type you like, but the .rpy files will be treated as special dynamic scripts.

The dynamic page config_refresher.rpy is a bit longer than any presented in the prior installment; much of that is simply the fact that it includes HTML templates in its body rather than importing them. Let us first look at the setup code

Dynamic script config_refresher.py (setup)

from twisted.web import resource, server
from persist import Records
from webloglib import log_fields, COLOR
from urllib import unquote_plus as uqp

fieldnames = """ip timestamp request status
                bytes referrer agent""".split()
field_dict = dict(zip(fieldnames, range(len(fieldnames)))

Other than a few imports that we have seen in prior installments, I map field names to their positions in the tuple returned by log_fields(). Do notice also the use of a custom persist module that will hold the weblog in memory within the Twisted web server, so that the whole log file does not need to be read each time any client requests some records. Next the HTML templates:

config_refresher.py script (templates)

TOP = '''<html><head><title>Weblog Refresher</title>
  <META HTTP-EQUIV="Refresh" CONTENT="30"/></head>
  <body>
  <table border="1" width="100%%">
  <tr bgcolor="yellow">
  <form action="http://gnosis.cx:8080/config_refresher.rpy"
        method="GET">
    <td> IP  <input type="checkbox" name="ip" %s/> </td>
    <td> Timestamp <input type="checkbox" name="timestamp" %s/></td>
    <td> Request <input type="checkbox" name="request" %s/></td>
    <td> Status <input type="checkbox" name="status" %s/></td>
    <td> Bytes  <input type="checkbox" name="bytes" %s/></td>
    <td> Referrer <input type="checkbox" name="referrer" %s/></td>
    <td> Agent <input type="checkbox" name="agent" %s/></td>
    <td> <input type="submit" value="Change Fields"></td>
  </form>
  </td></tr>
  <table border="0" cellspacing="0" width="100%%">'''
ROW = '<tr bgcolor=" %s">%s</tr>\n'
END = '</table></body></html>'
COLOR = ['white','lightgray']
END = '''</table></body></html>'''

Setting up an HTML form is not too mysterious, but one little trick in the example is to interpolate the string "checked" into those checkboxes in the HTML that have been checked.

config_refresher.py script (persistence)

records = registry.getComponent(Records)
if not records:
   records = Records()
   registry.setComponent(Records, records)

The Twisted registry works as described in the last installment--it is simply the place where the latest records in the web log file are held. Finally, we create a Resource, with a corresponding .render() method--this does this actual page creation:

config_refresher.py script (rendering)

class Resource(resource.Resource):
    def render(self, request):
        showlist = []
        for field in request.args.keys():
            showlist.append(field_dict[field])
        showlist.sort()
        checked = [""] * len(fieldnames)
        for n in showlist:
            checked[n] = 'checked'
        request.write(TOP % tuple(checked))
        odd = 0
        for rec in records.getNew():
            hit = [field.strip('"') for field in log_fields(rec)]
            flds='\n'.join(['<td>%s</td>'%hit[n] for n in showlist])
            request.write(ROW % (COLOR[odd],
                                 uqp(flds).replace('&amp;',' &')))
            odd = not odd
        request.write(END)
        request.finish()
        return server.NOT_DONE_YET
resource = Resource()

The main new thing in this Resource is the access to the request.args attribute. Generally, this attribute is similar to the Python cgi modules FieldStorage class--it collects any information passed with the page request, both GET and PUT data. Twisted's request data is a dictionary of passed values--in our case we are just interested in which checkboxes fields are passed in and which not. But it would follow the same pattern if we wanted to check some values stored in request.args--perhaps you might add options to filter based on field values, for example (and choose this with a text entry or an HTML listbox).

Templating With Woven

The dynamic pages we have looked at so far have all been conceptually fairly similar to a CGI approach. Twisted asynchronous server is faster--and it especially saves time to avoid the overhead of opening a new process for each script request. But fastcgi or mod_python achieve a similar speedup. There is nothing all that special about Twisted in this regard.

One way to move web application development to a higher level is to use Woven. In concept, Woven is somewhat similar to PHP, ASP (Active Server Pages), or JSP (Java Server Pages). That is, Woven XHTML pages are not simply the pages delivered to browsers, but rather templates or skeletons of pages that are filled in programmatically. However, the separation between code and HTML is a bit different in Woven than in those page-embedding technologies. You do not write Python code directly inside a Woven template. Instead you define a series of custom XHTML attributes on normal tags that let external code enhance and manipulate the page in preparation for delivery to the browser client.

The model attribute determines the data that is used for expanding the XHTML element it is attached to. The idea is that a Model represents the "business logic" of an application--how the data content of a page is determined. The view attribute, in contrast determines the particular presentation of the generated data. There is also the concept of a Controller in Woven which is the code that combines the Model with the View of a node (i.e. an XHTML element). This last part is usually handled by a Page object, which is a class that can be specialized.

The nomenclature behind Woven is admittedly a bit difficult; and unfortunately, the HOWTO documents at the Twisted Matrix website do almost as much to obscure matters as they do to illuminate them. It is hard going to figure out exactly how to use Woven. I do not yet claim to wholly understand Woven concepts myself, but Twisted user Alex Levy (<mesozoic@polynode.com>) was kind enough to help me through developing the example I present below. Still, there is quite a bit you can do with Woven, so it is worth working through.

The first step for a Woven application is to develop a template file (or files). These are simply XHTML files with special attributes, e.g.:

WeblogViewer.xhtml template

<html>
<head>
  <title>Weblog Viewer</title>
  <meta HTTP-EQUIV="Refresh" CONTENT="30" />
  <style type="text/css"><!--
    div.info {
      background-color: lightblue;
      padding: 2px dotted; }
    table th, table td {
      text-align: left;
      cellspacing: 0px;
      cellpadding: 0px; }
    table.log {
      border: 0px;
      width: 100%; }
    table.log tr.even { background-color: white; }
    table.log tr.odd  { background-color: lightgray; }
  --></style>
</head>
<body>
  <div class="info">
  You are displaying the contents of
  <code model="filename" view="Text">filename</code>.
  </div>
  <table border="0" cellspacing="0" width="100%"
         class="log" model="entries" view="List">
    <tr bgcolor="yellow" pattern="listHeader">
      <th>Referrer</th><th/>
      <th>Resource</th>
    </tr>
    <tr pattern="listItem" view="alternateColor">
      <td model="referrer" view="Text">
          Referrer</td>
      <td>-&gt;</td>
      <td model="request_resource" view="Text">
          Resource</td>
    </tr>
    <tr pattern="emptyList">
      <td colspan="2">There is nothing to display.</td>
    </tr>
  </table>
</body>
</html>

Alex Levy developed this template, and showing better style than I have in my examples, used CSS2 to control the exact presentation of elements. Obviously, the basic layout of the page is the same with or without the style sheet.

Notice that the <table> element is assigned the view "List", which is a basic Woven View, as is "Text"--on the other hand, "alternateColor" is a customized view that we define in code below. Some elements have a pattern attribute which is used by the controlling View to locate matching children. In particular a List view is composed of an optional listHeader, some listItem children (one template tag, but expanded during generation), and an emptyList child in case the Model does not locate any data. These patterns are standard attributes that a List view uses, other views would utilize other patterns for expansion.

The code for this version of a weblog server creates a custom Twisted sever. Rather than update based on requests by clients, we add a repeated callback to the update() function to the server's Reactor; this is substantially the same as with the tlogmaker.py in the prior installment. Let us look at the setup code first before we examine the customized Page resource:

WeblogViewer.py custom Twisted server

import webloglib as wll
import os, sys
from urllib import unquote_plus as uqp
from twisted.internet import reactor
from twisted.web import microdom
from twisted.web.woven import page, widgets

logfile = '../access-log'
LOG = open(logfile)
RECS = []
NUM_ROWS = 25

def update():
    global RECS
    RECS.extend(LOG.readlines())
    RECS = RECS[-NUM_ROWS*3:]
    reactor.callLater(5, update)
update()

The interesting stuff comes with our customization of the class twisted.web.woven.page.Page. Most of what we do is magic, in the sense that you need to define specially-named attributes and methods.

WeblogViewer.py Twisted server (continued)

class WeblogViewer(page.Page):
    """A Page used for viewing Apache access logs."""
    templateDirectory = '~/twisted/www'
    templateFile = "WeblogViewer.xhtml"

    # View factories and updates
    def wvupdate_alternateColor(self, request, node, data):
        """Makes our table rows alternate CSS classes"""
        # microdom.lmx is very handy; another example is located here:
        # http://twistedmatrix.com/documents/howto/picturepile#auto0
        tr = microdom.lmx(node)
        tr['class'] = ('odd','even')[data['_number']%2]

    # Model factories
    def wmfactory_filename(self, request):
        """Returns the filename of the log being examined."""
        return os.path.split(logfile)[1]

    def wmfactory_entries(self, request):
        """Return list of dict objects representing log entries"""
        entries = []
        for rec in RECS:
            hit = [field.strip('"') for field in wll.log_fields(rec)]
            if hit[wll.status] == '200' and hit[wll.referrer] != '-':
                # We add _number so our alternateColor view will work.
                d = {'_number': len(entries),
                     'ip': hit[wll.ip],
                     'timestamp': hit[wll.timestamp],
                     'request': hit[wll.request],
                     'request_resource': hit[wll.request].split()[1],
                     'status': hit[wll.status],
                     'bytes': hit[wll.bytes],
                     'referrer': uqp(hit[wll.referrer]).\
                                     replace('&amp;',' &'),
                     'agent': hit[wll.agent],
                    }
                entries.append(d)
        return entries[-NUM_ROWS:]
resource = WeblogViewer()

There are three categories of things that our custom Page does. The first sets up the template to use with this resource.

The second defines a custom View, using the magic method prefix wv (Woven view). All we really do in the custom view is set the class attribute to one of two values from the CSS stylesheet, to make alternating rows of display have a different color. But you could manipulate the node however you like, using a DOM-like API.

The third category is interesting. We define two Models using methods with wmfactory_ prefixed to the name of the Model itself. Since filename is displayed in a Text view, it is best to return a string. And likewise, entries displayed in a List view should get a list of entries as return value. But what about the Models referrer and request_resource that are used in the XHTML template? No custom methods were defined for these models. However, the listItem pattern that surrounds nodes with these Models has a dictionary made available to it--the entries dict returned by .wmfactory_entries(). And this dictionary, in turn, contains keys for request_resource and referrer; you do not need a custom method to support a Model, a dictionary with the necessary key works fine. Since the View for the referrer node is Text, so said dictionary should contain a string for a value (but if not, Woven makes good efforts to coerse it).

Creating a custom server based on the custom WeblogViewer.py resource works as we have seen before. Create a server, then launch it:

% mktap web --resource-script=WeblogViewer.py --port 8080
% twistd -f web.tap

In The Last Part

The introduction has only scratched the surface of Woven. There are whole labyrinths within that package; but I hope the example I presented gives a bit of the feel for the templating system.

In the final installment of this series on Twisted, I will pick up some odds and ends, and take a look at a few of the specialized protocols and servers contained in the Twisted package.

Resources

Twisted Matrix comes with quite a bit of documentation, and many examples. Browse around its homepage to glean a greater sense of how Twisted Matrix works, and what has been implemented with it (or wait for the next installments here):

http://twistedmatrix.com

You can read about the Woven templating framework at:

http://twistedmatrix.com/documents/howto/woven

A simple version of a weblog server was presented in the developerWorks tip, Use Simple API for XML as a long-running event processor:

http://www-106.ibm.com/developerworks/xml/library/x-tipasysax.html

About The Author

Picture of Author David Mertz believes that it is turtles all the way down. David may be reached at mertz@gnosis.cx; his life pored over at http://gnosis.cx/publish/. And buy his book: Text Processing in Python (http://gnosis.cx/TPiP/).