Fully automatic database table partitioning for Django
THIS LIBRARY IS DEPRECATED IN FAVOR OF ARCHITECT. ALL FEATURES THAT ARE NOT CURRENTLY BACKPORTED TO ARCHITECT WILL BE BACKPORTED IN THE NEAR TIME. PLEASE CONSIDER SWITCHING TO ARCHITECT ASAP, AS DJANGO-DB-PARTI WILL RECEIVE NO FUTURE UPDATES AND BUG FIXES.
Django DB Parti is a package for Django which aim is to make table partitioning on the fly. Partitioning is a division of one large table into several smaller tables which represent that table. Partitioning is usually done for manageability, performance or availability reasons. If you are unsure whether you need partitioning or not, then you almost certainly don’t need it.
Django DB Parti supports multiple database backends, each database differ from each other in many ways, that means that some features may be available for one database backend, but not for the other, below you will find list of supported database backends and detailed information which database backend supports which features.
PostgreSQL’s partitioning implementation in Django DB Parti is done purely at the database level. That means that Django DB Parti creates several triggers and functions and inserts them directly into the database, so even if you issue direct insert statement from database console and not from Django, everything will work as expected and record will be inserted into the correct partition, if partition doesn’t exist, it will be created for you automatically. Also partitions may be created in any order and not only from lower to higher.
Range partitioning by date/datetime for the following periods:
Currently there are no known limitations for this backend, except that not all partitioning types are supported. New types will be added in next releases of Django DB Parti.
MySQL’s partitioning implementation in Django DB Parti is done in a mixed way, half at the python level and half at the database level. Unfortunately MySQL doesn’t support dynamic sql in triggers or functions that are called within triggers, so the only way to create partitions automatically is to calculate everything at the python level, then to create needed sql statements based on calculations and issue that statement into the database.
Range partitioning by date/datetime for the following periods:
Not all partitioning types are supported. New types will be added in next releases of Django DB Parti.
Partitioning is not available for bulk inserts (i.e. Django’s bulk_create() method) because it doesn’t call model’s save() method which this backend relies on. Currently there is no known way to remove this limitation.
New partitions can be created only from lower to higher, you can overcome this with MySQL’s special command REORGANIZE PARTITION which you have to issue from the database console. You can read more about it at the MySQL’s documentation. We plan to remove this limitation in one of the future releases of Django DB Parti.
Django >= 1.4.x
$ pip install django-db-parti
or clone from github:
$ git clone git://github.com/maxtepkeev/django-db-parti.git
Add dbparti to PYTHONPATH and installed applications:
INSTALLED_APPS = ( ... 'dbparti', )
Create the model as usual which will represent the partitioned table and run syncdb to create a table for the model, if you are using South for migrations, you can also create the model as usual via migrate. No additional steps required. After that we need to make a few changes to the model:
from dbparti.models import Partitionable
class Meta(Partitionable.Meta): partition_type = 'range' partition_subtype = 'date' partition_range = 'month' partition_column = 'added'
$ python manage.py partition app_name
That’s it! Easy right?! Now a few words about what we just did. We made our model to inherit from Partitionable, also we used “month” as partition range and “added” as partition column, that means that from now on, a new partition will be created every month and a value from “added” column will be used to determine into what partition the data should be saved. Keep in mind that if you add new partitioned models to your apps or change any settings in the existing partitioned models, you need to rerun the command from step 4, otherwise the database won’t know about your changes. You can also customize how data from that model will be displayed in the Django admin interface, for that you need to do the following:
from dbparti.admin import PartitionableAdmin
partition_show = 'all'
All model settings are done inside model’s Meta class which should inherit from Partitionable.Meta
partition_type - what partition type will be used on the model, currently accepts the following:
partition_subtype - what partition subtype will be used on the model, currently used only when “partition_type” is set to “range” and accepts the following values:
partition_range - how often a new partition will be created, currently accepts the following:
partition_column - column, which value will be used to determine which partition record belongs to
All model admin settings are done inside model admin class itself
partition_show - data from which partition will be shown in Django admin, accepts the following values:
Let’s imagine that we would like to create a table for storing log files. Without partitioning our table would have millions of rows very soon and as the table grows performance will become slower. With partitioning we can tell database that we want a new table to be created every month and that we will use a value from some column to determine to which partition every new record belongs to. To be more specific let’s call our table “logs”, it will have only 3 columns: id, content and added. Now when we insert the following record: id=’1’, content=’blah’, added=’2013-05-20’, this record will be inserted not to our “logs” table but to the “logs_y2013m05” partition, then if we insert another record like that: id=’2’, content=’yada’, added=’2013-07-16’ it will be inserted to the partition “logs_y2013m07” BUT the great thing about all of that is that you are doing your inserts/updates/selects on the table “logs”! Again, you are working with the table “logs” as usual and you don’t may even know that actually your data is stored in a lot of different partitions, everything is done for you automatically at the database level, isn’t that cool ?!
I will be glad to get your feedback, pull requests, issues, whatever. Feel free to contact me for any questions.
django-db-parti is protected by BSD licence. Check the LICENCE for details.
Fixed a bug with partition command not working for MySQL backend (Issue #11)
Added support for DateField and DateTimeField with auto_now and auto_now_add attributes set (Issue #3)
Fixed an issue with unnecessary calling of partitioning functions while reading data from database
MySQL: Fixed inability to create partitions for December when range was set to month
MySQL: Backend was completely broken in previous version, now everything should work properly (Issue #4)
Rewritten from scratch, introduced new API to add support for new backends and partition types
All default model settings which are done inside model’s Meta class are now set to None, that means that there are no more default settings. Everything should be explicitly defined inside each model class.
Introduced new model setting partition_type which currently accepts only one value range
Introduced new model setting partition_subtype which currently accepts only one value date and is used only with partition_type when it’s set to range
Better error handling, django-db-parti tries it’s best to tell you where and why an error occured
Added support for day and year partitioning for all backends in addition to week and month
PostgreSQL: new partitions are now created at the database level, that gave some speed improvement, also we don’t rely on Django’s save() method anymore, that means that there is no more limitation with Django’s bulk_create() method, you can use it freely with partitioned tables
PostgreSQL: fixed an error when last day of the week or month wasn’t inserted into partition
Python 3 compatibility
Datetime with timezone support (Issue #1)
Added mysql backend
Fixed incorrect handling of datetime object in DateTimeMixin
Fixed postgresql backend error which sometimes tried to insert the data into partitions that don’t exist
Moved all the database partition system stuff to the command partition (see readme), that gave a lot in speed improvement because we don’t need to check for trigger existance and some other things at runtime anymore
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.