The Web Mart Design Pattern.

date:2007-05-03 13:28:05
category:Technology News

The Web Mart design pattern identifies three aspects to a piece of web content:

  • The Core Concept - the “fact” that we will present.
  • The Access Dimensions - objects that identify a concept or support navigation to the concept.
  • The Details - objects that provide details in support of a concept.

There are a number of things to like about this. First, and most important, is the close parallelism between this design pattern and the Star Schema (or Dimensional Model or Data Mart ) design pattern. I’m a big fan of the dimensional model, and presented an implementation of the dimensional design pattern at PyCon 2007 . The dimensional model surrounds a fact with independent dimensions of that fact. The dimensions serve as navigation and aggregation of the fact; the fact is the most granular detail in the data structure.

What’s important is the recognition that navigation involves independent dimensions . Many people attempt to force-fit independent dimensions into a single hierarchical taxonomy. Somewhere there’s this fantasy that a single taxonomy can represent all knowledge. This fantasy dates from the early efforts to create encyclopedias, thesauri and dictionaries, and it permeates our lives through the Dewey Decimal Classification in libraries.

No matter how we slice and dice multiple dimensions, they don’t fit into a single taxonomy well. We can always reslice and redice the independent variables to create a taxonomy which has the same information in a different ordering. Mathematically speaking, if we have n independent dimensions, we can use any of the n ! orderings of the available dimensions to create a distinct taxonomy. All n ! orders contain the same information.

Which is more important, to organize the files on your hard drive by date? By subject? Or by application software? All are viable dimensions. There are six possible orderings of those three dimensions.

Web Mart Data Model.

The Web Mart design pattern provides guidance that can constrain our data model in useful and appropriate ways. This guidance prevents us from fumbling around and creating a data model that isn’t easy to for people to navigate and maintain. The relational model, in many cases, offers us too much power. Some constraints help structure the data in familiar, usable ways.

First, our Core Concepts will be objects that have interesting, often complex attributes, and will require a fairly sophisticated page template in which they are displayed. These are the central entities in the data model, with the most relationships and the most attributes. Further, all of the user’s navigation use cases have the same essential goal: to view a specific Core Concept page.

Second, our Access Dimensions will be lists or hierarchies that facilitate navigation and search, define menus of various kinds, but aren’t – themselves – the user’s goal. Access dimensions are the means to the user’s end. These will have easy-to-manage 1-to-many relationships with Core Concepts . If they form a hierarchy, they will have simple 1-to-many relationships among the levels of the hierarchy.

What’s important is that the Access Dimensions are independent of each other. There is no tangling or confusion of the access relationships. The access relationships are all ways to find a Core Concept . Since a Core Concept will have a number of access dimensions, this forms an easy-to-manage star, not a tangled knot of relationships.

Handling Details.

There are two kinds of details that might be related to a Core Concept . In some cases, the Core Concept has supporting details that are simply collections of objects. These details have a pleasant 1-to-many relationship with the Core Concept . For example, multiple phone numbers for a person’s contact information is just many detail objects collected into a composite object.

The other kind of detail is a relationship with another Core Concept . For example, an Invoice has a number of Line Items, each of which relates to a Product. A Line Item isn’t, itself, a Core Concept . However an Invoice and a Product are Core Concepts . Invoices have a many-to-many relationship with Products, and this relationship is implemented through an association that we call a Line Item Detail.

Our data model, then, is dominated by Core Concepts , which have Access Dimensions , Details , and associations with other Core Concepts . This helps us structure our models, and navigation.

The Django Implication.

Each Django server has access to a number of “applications”, defined by the “INSTALLED_APPS” setting. Each Django application is a Python package with models, urls, and views; it may also include templates, tags and media. Django seems to make use of an Application-Model-View design pattern.

By the way, an application appears to fit into the overall Django world-view as follows. We have a Django server, configured by a settings file and started on a specific IP address and port number. This Django server may be one of many servers sharing a common code base and database, or it may stand-alone. Django servers are – for production purposes – usually front-ended by Apache, which handles static content (known as “media files”), and which handles the slow-client, fast-server balancing issue.

The Web Mart design pattern seems to fit nicely with the Django’s Application-Model-View design pattern. Each Core Concept is a Django application. The model contains the Core Concept, the Access Dimensions and the Details. We design the URLs to provide navigation aids through the Access Dimensions. We have two kinds of view: core concept detail views and access dimension list views. Each kind of view, in turn, relies on either a core concept template or an access list template.

Building The Web Mart.

Let’s say we’re building a web site to describe a church summer camp. We have a number of Core Concepts in our camping ministries web site: we have “About” statements, “Volunteer” information, “Employment” information “Privacy and Safety” information, “Camps”, “Retreats”, “Profiles” and “Forms and Paperwork”. Each of these is a separate core concept, and can be implemented as a separate, small application.

Let’s look at Camps, specifically. The Core Concept is Camp, which has relatively few attributes. It has a name, a description, and a few administrative details like the number of campers and the number of counsellors.

The access dimensions for a Camp include the schedule dates, the appropriate age group, and perhaps some other classification scheme that the camping program uses, like “outdoor” or “adventure” or “creativity” or “special needs”.

The detail dimensions for a Camp might include additional descriptions, photos from previous years, a list of things to bring.

Django Managers.

In Django, a Manager can be used to simplify queries against the database. In this case, each access dimension may have a manager. The camp schedule date dimension, for example, should have a manager which uses the current date to filter only camps which are scheduled to begin in the future. This trivially filters past camps from web queries, making the views and templates much simpler.

Here’s something that looks like it might be a reasonable model. This has only a single access dimension, Schedule.

class CurrentCamp( models.Manager ):
    """Manager for currently scheduled camps only."""
    def imageScheduleList( self, aDate=None ):
        """Creates a nested-list structure of [ ( camp, image, ( sched, ... ) ), ... ]"""
        now= aDate or
        campSchedList= []
        qs= super(CurrentCamp, self).get_query_set()
        qs= qs.filter( site=settings.SITE_ID )
        qs= qs.filter( schedule__startDate__gte=now )
        for c in qs.distinct():
            imgList= c.campimage_set.filter( startDate__lte=now, expireDate__gt=now )
            if imgList:
                img= random.choice( imgList )
                img= None
            sch= c.schedule_set.filter( startDate__gte=now )
            campSchedList.append( ( c, img, sch ) )
        return campSchedList
    def get_query_set(self):
        qs= super(CurrentCamp, self).get_query_set()
        qs= qs.filter( site=settings.SITE_ID )
        qs= qs.filter(schedule__startDate__gte=now )
        return qs.distinct()

class Camp( models.Model ):
    """A camping program.

    This is the generic description.  Only descriptions which have a schedule
    will be shown.  This allows you to have descriptions for programs that aren't
    scheduled in the current year.
    site= models.ForeignKey( Site )
    name= models.CharField( maxlength=64 )
    description= models.TextField()
    staff= models.IntegerField( null=True )
    campers= models.IntegerField( null=True )
    duration= models.IntegerField( null=True, default=5, help_text='Days.' )
    objects = models.Manager() # default manager
    current= CurrentCamp() # currently scheduled camps only
    class Admin:
        list_display= ( 'name', 'site', 'duration', )
    def __str__( self ):
    def image( self, aDate ):
        img_set= self.campimage_set.filter( startDate__lte=aDate, expireDate__gt=aDate )
        if img_set:
            return random.choice( img_set )
    def __repr__( self ):
        return "Camp( name=%(name)r, description=%(description)r, \
staff=%(staff)r, campers=%(campers)r, duration=%(duration)r )" % ( self.__dict__ )

class Schedule( models.Model ):
    """A schedule for a Camping program.

    This is the actual schedule.  A camp which is scheduled in the future
    is shown to visitors.
    camp= models.ForeignKey(Camp)
    startDate= models.DateField( help_text='Starting date for this camp' )
    class Admin:
        list_display= ( 'camp', 'startDate', )
    def __str__( self ):
        return "%s on %s" % (, self.startDate.strftime( "%Y-%m-%d" ) )
    def __repr__( self ):
        return "Schedule( startDate=%(startDate)r )" % ( self.__dict__ )

class CampImage( models.Model ):
    """An image that decorates a specific camp listing.

    Any number of images can be associated with a given camp.
    However, one is selected arbitrarily to show with a camp entry.
    camp= models.ForeignKey( Camp, edit_inline=True, )
    caption= models.CharField( maxlength=128, core=True, )
    startDate= models.DateField( help_text='First date to display this image' )
    expireDate= models.DateField(,12,31),
        help_text='Date on which this image is removed.', validator_list=[campValidators.checkExpireDate] )
    image= models.ImageField( upload_to="photos/%Y%m" )
    class Admin:
        list_display= ('caption', 'camp', 'startDate', 'expireDate', )
    def __str__( self ):
        return "%s: %s ( %s to %s )" % (, self.caption, self.startDate, self.expireDate )
    def __repr__( self ):
        return "CampImage( caption=%(caption)r, startDate=%(startDate)r, \
expireDate=%(expireDate)r, image=%(image)r )" % ( self.__dict__ )

The URLs and Views.

We have two overall kinds of templates and views. We have the Core Concept detail view, which locates a specific Core Concept and associated details; this uses a template that shows all of the relevant details. This detailed view could be located in several places in the URL scheme because there may be several access dimensions that lead us to the resulting Core Concept.

The other kind of templates and views are the access dimensions. Each access dimension defines one or more list views, or menus. When there are multiple dimensions, a menu may be used to select which dimension is used for access. Each dimension has URL’s for traversing the dimension, views for locating relevant rows in that dimension, and a template for displaying the access dimension rows, and possibly Core Concept rows.

In our Camp example, we only have one access dimension defined. However, we have to define our URL’s to permit additional access dimensions. Many Django examples imply that a single dimension is somehow “primary” for accessing a Core Concept. This is rarely true, and a slightly different URL naming scheme makes it possible to add and change access dimensions without breaking an application.

Here’s a portion of the URL definitions. Note that we use a /camp/byDate/11 URL to use the schedule access dimension.  We can then add ``/camp/byAgeGroup/ to implement another access dimension.

from django.conf.urls.defaults import *

urlpatterns = patterns('campministry.apps.public.views',

    # The stmt_page matches the PAGE_CHOICE in the models.
    # The title should match the menu provided in the template.

    (r'^$', 'index', {'stmt_page':'Home',} ),
    (r'^home.*$', 'index', {'stmt_page':'Home',} ),
    (r'^index.*$', 'index', {'stmt_page':'Home',} ),
    ... other stuff ...
    (r'^camp/byDate/$', 'campByDate', {'stmt_page':'Camps', 'title':'Summer Camps'} ),
    (r'^camp/(?P\d+)/$', 'camp', {'stmt_page':'Camps', 'title':'Summer Camps'} ),

Here’s a portion of the view definitions. We have a generic view function (indexView) that provides the common information used by all Core Concept views. The campByDate and camp views expand on this view with either a list of Camps, based on one of the access dimensions, or a specific Camp.

def indexView( request, stmt_page, title=None ):
    """ Get Statements, Images and Profiles to fill this page.
    These items are Site-related.  The site qualifies Statements, Images and Profiles.
    pageDict= baseView(request)
    pageDict['title'] = title or pageDict['site_name']

    now= today( request, pageDict )
    pageDict['stmt_list']= Statement.current.asof( now ).filter( page=stmt_page ).order_by('startDate')

    img_list = Image.current.asof( now ).filter( page=stmt_page )
    if img_list:
        pageDict['image'] = random.choice( img_list )

    pageDict['staff_list'] = Profile.activeProfile.filter( contact=True )
    return pageDict

def campByDate( request, stmt_page, title ):
    """List of Camps, organized by the schedule access dimension."""
    pageDict= indexView( request, stmt_page, title )
    now= today( request, pageDict )

    pageDict['camp_list']= Camp.current.imageScheduleList( now )
    return render_to_response('camp.html', pageDict )

def camp( request, object_id, stmt_page, title ):
    """A specific Camp."""
    pageDict= indexView( request, stmt_page, title )
    now= today( request, pageDict )

    pageDict['camp']= Camp.objects.get(pk=object_id)
    return render_to_response('camp.html', pageDict )

The important value of the Web Mart design pattern is to prevent thinking of a single taxonomy of camps. We can organize the list of Camps by any of the available access dimensions. In this case, we’ve only defined Schedule, but the design pattern helps us recognize that we are unlikely to have a single access dimension for a Core Concept.

Previous topic

Dynamic Languages In The News

Next topic

Dynamic Languages and Corporate IT’s Objections

This Page