Skip to content

jamessewell/django-timescaledb

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

82 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Django timescaledb

PyPI version fury.io

Workflow

A database backend and tooling for Timescaledb.

Based on gist from WeRiot.

Quick start

  1. Install via pip
pip install django-timescaledb
  1. Use as DATABASE engine in settings.py:

Standard PostgreSQL

DATABASES = {
    'default': {
        'ENGINE': 'timescale.db.backends.postgresql',
        ...
    },
}

PostGIS

DATABASES = {
    'default': {
        'ENGINE': 'timescale.db.backends.postgis',
        ...
    },
}

If you already make use of a custom PostgreSQL db backend you can set the path in settings.py.

TIMESCALE_DB_BACKEND_BASE = "django.contrib.gis.db.backends.postgis"
  1. Inherit from the TimescaleModel. A hypertable will automatically be created.
  class TimescaleModel(models.Model):
    """
    A helper class for using Timescale within Django, has the TimescaleManager and
    TimescaleDateTimeField already present. This is an abstract class it should
    be inheritted by another class for use.
    """
    time = TimescaleDateTimeField(interval="1 day")

    objects = TimescaleManager()

    class Meta:
        abstract = True

Implementation would look like this

from timescale.db.models.models import TimescaleModel

class Metric(TimescaleModel):
   temperature = models.FloatField()

If you already have a table, you can either add time field of type TimescaleDateTimeField to your model or rename (if not already named time) and change type of existing DateTimeField (rename first then run makemigrations and then change the type, so that makemigrations considers it as change in same field instead of removing and adding new field). This also triggers the creation of a hypertable.

from timescale.db.models.fields import TimescaleDateTimeField
from timescale.db.models.managers import TimescaleManager

class Metric(models.Model):
  time = TimescaleDateTimeField(interval="1 day")

  objects = models.Manager()
  timescale = TimescaleManager()

The name of the field is important as Timescale specific feratures require this as a property of their functions.

Reading Data

"TimescaleDB hypertables are designed to behave in the same manner as PostgreSQL database tables for reading data, using standard SQL commands."

As such the use of the Django's ORM is perfectally suited to this type of data. By leveraging a custom model manager and queryset we can extend the queryset methods to include Timescale functions.

Time Bucket More Info

  Metric.timescale.filter(time__range=date_range).time_bucket('time', '1 hour')

  # expected output

  <TimescaleQuerySet [{'bucket': datetime.datetime(2020, 12, 22, 11, 0, tzinfo=<UTC>)}, ... ]>

Time Bucket Gap Fill More Info

  from metrics.models import *
  from django.db.models import Count, Avg
  from django.utils import timezone
  from datetime import timedelta

  ranges = (timezone.now() - timedelta(days=2), timezone.now())

  (Metric.timescale
    .filter(time__range=ranges)
    .time_bucket_gapfill('time', '1 day', ranges[0], ranges[1], datapoints=240)
    .annotate(Avg('temperature')))

  # expected output

  <TimescaleQuerySet [{'bucket': datetime.datetime(2020, 12, 21, 21, 24, tzinfo=<UTC>), 'temperature__avg': None}, ...]>

Histogram More Info

  from metrics.models import *
  from django.db.models import Count
  from django.utils import timezone
  from datetime import timedelta

  ranges = (timezone.now() - timedelta(days=3), timezone.now())

  (Metric.timescale
    .filter(time__range=ranges)
    .values('device')
    .histogram(field='temperature', min_value=50.0, max_value=55.0, num_of_buckets=10)
    .annotate(Count('device')))

  # expected output

  <TimescaleQuerySet [{'histogram': [0, 0, 0, 87, 93, 125, 99, 59, 0, 0, 0, 0], 'device__count': 463}]>

Contributors