django-trackstats 0.7.0

Use Case

• You need an elegant solution for storing statistics in a generic and structural fashion.

• You need to denormalize the results of various aggregated queries.

• You require access to the stored statistics within your application layer.

So, the focus is purely on storing statistics for use within your application later on. Other features, such as charting, reports, OLAP, query builders, slicing & dicing, integration with Datadog and the likes are all beyond scope.

Concepts

The following concepts are used:

Metric

A piece of information to keep track of. For example, “Order count”, or “Number of users signed up”.

Domain

Metrics are organized in groups, each group is called a domain. For example you can have a “shopping” domain with metrics such as “Order count”, “Items sold”, “Products viewed”, and a “users” domain with “Login count”, “Signup count”. Or, in case you are tracking external statistics from social networks, you may introduce a “Twitter” domain, and metrics “Followers count”.

Statistic

Used to store the actual values by date, for a specific metric.

Period

The time period for which the stored value holds. For example, you can keep track of cumulative, all-time, numbers (Period.LIFETIME), store incremental values on a daily basis (Period.DAY), or keep track of a rolling count for the last 7 days (Period.WEEK).

Reference IDs

Domains and metrics must be assigned unique reference IDs (of type string). Rationale: Having a human readable, non PK based, reference is esential as soon as you are going to export statistics.

Usage

First, setup your domains:

from trackstats.models import Domain

Domain.objects.SHOPPING = Domain.objects.register(
    ref='shopping',
    name='Shopping')
Domain.objects.USERS = Domain.objects.register(
    ref='users',
    name='Users')
Domain.objects.TWITTER = Domain.objects.register(
    ref='twitter',
    name='Twitter')

Define a few metrics:

from trackstats.models import Domain, Metric

Metric.objects.SHOPPING_ORDER_COUNT = Metric.objects.register(
    domain=Domain.objects.SHOPPING,
    ref='order_count',
    name='Number of orders sold')
Metric.objects.USERS_USER_COUNT = Metric.objects.register(
    domain=Domain.objects.USERS,
    ref='user_count',
    name='Number of users signed up')
Metric.objects.TWITTER_FOLLOWER_COUNT = Metric.objects.register(
    # Matches Twitter API
    ref='follower_count',
    domain=Domain.objects.TWITTER)

Now, let’s store some one-off statistics:

from trackstats.models import StatisticByDate, Domain, Metric, Period

# All-time, cumulative, statistic
n = Order.objects.all().count()
StatisticByDate.objects.record(
    metric=Metric.objects.SHOPPING_ORDER_COUNT,
    value=n,
    period=Period.LIFETIME)

# Users signed up, at a specific date
dt = date.today()
n = User.objects.filter(
    date_joined__day=dt.day,
    date_joined__month=dt.month,
    date_joined__year=dt.year).count()
StatisticByDate.objects.record(
    metric=Metric.objects.USERS_USER_COUNT,
    value=n,
    period=Period.DAY)

Creating code to store statistics yourself can be a tedious job. Luckily, a few shortcuts are available to track statistics without having to write any code yourself.

Consider you want to keep track of the number of comments created on a daily basis:

from trackstats.trackers import CountObjectsByDateTracker

CountObjectsByDateTracker(
    period=Period.DAY,
    metric=Metric.objects.COMMENT_COUNT,
    date_field='timestamp').track(Comment.objects.all())

Or, in case you want to track the number of comments, per user, on a daily basis:

CountObjectsByDateAndObjectTracker(
    period=Period.DAY,
    metric=Metric.objects.COMMENT_COUNT,
    # comment.user points to a User
    object_model=User,
    object_field='user',
    # Comment.timestamp is used for grouping
    date_field='timestamp').track(Comment.objects.all())

Models

The StatisticByDate model represents statistics grouped by date – the most common use case.

Another common use case is to group by both date and some other object (e.g. a user, category, site). For this, use StatisticByDateAndObject. It uses a generic foreign key.

If you need to group in a different manner, e.g. by country, province and date, you can use the AbstractStatistic base class to build just that.

Cross-Selling

If you like this, you may also like:

• django-allauth: https://github.com/pennersr/django-allauth

• netwell: https://github.com/pennersr/netwell