django-pgjsonb

Django Postgres JSONB Fields support with lookups

Originaly inspired by django-postgres

Change Logs

2017-09-13: 0.0.29 Fix JsonAdapter Python2 incompatible

2017-09-11: 0.0.28 Fix contained_by contains empty {} Fix error of has_any/ has_all Fix lookup filter(meta={})

2017-08-31: 0.0.27 Fix as_{} lookup for python3

2017-08-31: 0.0.26 Fix has lookup after Django 1.10

2017-05-18：0.0.25 Supress exception when drop index and the index already removed.

2017-03-14: 0.0.24 Add support for __near lookup with postgres earthdistance plugin, Thanks to @steinliber

2016-06-01: 0.0.23 Fix value from select_json not been decode from json introduce by 0.0.18

2016-03-24: 0.0.22 Fix error #11 remove the unexpect decode float to Decimal

2016-03-19: 0.0.21 Fix error #10

2016-03-09: 0.0.20 Add the array length for select_json

2016-03-08: 0.0.19 fix when add a json field with db_index=True and it's fail to generate the create index sql

2016-03-01: 0.0.18 we want to be able to use customize decoder to load json, so get avoid the psycopg2's decode json, just return raw text then we deserilize by the field from_db_value

2016-03-01: 0.0.17 patch the django serilizer to not return the stringifyed result

2015-07-23: 0.0.16 Add support for ./manage.py inspectdb

2015-06-10: 0.0.15 Add support for db_index to add GIN index

Install

pip install django-pgjsonb

Definition

from django_pgjsonb import JSONField

class Article(models.Model):
	meta=JSONField([null=True,default={},decode_kwargs={},encode_kwargs={},db_index=False,db_index_options={}])

Encoder and Decoder Options

by define decode_kwargs and encode_kwargs you can use your customize json dump and load behaveior, basicly these parameters will just pass to json.loads(**decode_kwargs) and json.dumps(**encode_kwargs)

here is an example for use EJSON to store native datetime object

import ejson

class Article(models.Model):
	meta=JSONField(encode_kwargs={"cls":ejson.EJSONEncoder},decode_kwargs={"cls":ejson.EJSONDecoder})

Add Index

[new add in 0.0.15]

jsonb field support gin type index to accelerator filtering. Since JSON is a data structure contains hierarchy, so the index of jsonb field will be more complicate than another single value field. More information, please reference Postgres document 8.14.4

meta=JSONField(db_index=True)
or
meta=JSONField(db_index=True,db_index_options={"path":"authors__name","only_contains":True})
or
meta=JSONField(db_index=True,db_index_options=[{},{"path":"authors__name","only_contains":True}])

When set db_index as True and do not set db_index_options, it will generate default GIN index, most case it's enough.

When specify db_index_options={"only_contains":True}, the index will be as the non-default GIN operator class jsonb_path_ops that supports indexing the contains operator only, but it's consume less space and more efficient.

When specify the path parameter in db_index_options, db_index_options={"path":"authors__name"}, then index will generate to the specify path, so that Article.objects.filter(meta__authors__name__contains=["asd"]) can utilize the index.

So you can create multiple index in one JSONField, just pass the db_index_options parameter as a list that contains multiple options, it will generate multiple correspond indexes. Empty dict stand for the default GIN index.

Lookups

###Contains a wide range of lookups supported natively by postgres

1. has :if field has specific key ("?")

Article.objects.filter(meta__has="author")

2. has_any : if field has any of the specific keys ("?|")

Article.objects.filter(meta__has_any=["author","date"])

3. has_all : if field has all of the specific keys ("?&")

Article.objects.filter(meta__has_all=["author","date"])

4. contains : if field contains the specific keys and values ("@>")

Article.objects.filter(meta__contains={"author":"yjmade","date":"2014-12-13"})

5. in or contained_by : if all field key and value contain by input ("<@")

Article.objects.filter(meta__in={"author":"yjmade","date":"2014-12-13"})

6. len : the length of the array, transform to int, and can followed int lookup like gt or lt ("jsonb_array_length()")

Article.objects.filter(meta__authors__len__gte=3)
Article.objects.filter(meta__authors__len=10)

7. as_(text,int,float,bool,date,datetime) : transform json field into specific data type so that you can follow operation of this type ("CAST(FIELD as TYPE)")

Article.objects.filter(meta__date__as_datetime__year__range=(2012,2015))
Article.objects.filter(meta__view_count__as_float__gt=100)
Article.objects.filter(meta__title__as_text__iregex=r"^\d{4}")

8. path_(PATH) : get the specific path, path split by '_' ("#>")

Article.objects.filter(meta__path_author_articles__contains="show me the money")

Extend function to QuerySet

1.select_json("JSON_PATHS",field_name="JSON_PATHS")

JSON_PATHS in the format of paths separated by "__",like "meta__location__geo_info". It will use the queryset's extra method to transform a value inside json as a field. If no field_name provided, it will generate a field name with lookups separate by _ without the json field self's name, so select_json("meta__author__name") is equal to select_json(author_name="meta__author__name")

Article.objects.select_json("meta__author__name",geo="meta__location__geo_info")`

This operation will translate to sql as

SELECT "article"."meta"->'location'->'geo_info' as "geo", "article"."meta"->'author'->'name' as "author_name"

[new add in 0.0.20] You can also select the length of a json array as a field by use Length object

from django_pgjsonb.fields import Length
Article.objects.select_json(authors_len=Length("meta__authors")).values("authors_len")

After select_json, the field_name can be operate in values() and values_list() method, so that

1. select only one specific value inside json
2. to group by one value inside json

is possible.

Demo:

Article.objects.all().select_json(tags="meta__tags").values_list("tags")
# select only "meta"->'tags'

Article.objects.all().select_json(author_name="meta__author__name")\
	.values("author_name").annotate(count=models.Count("author_name"))
# GROUP BY "meta"->'author'->'name'

support geo search in jsonb

require: postgresql plugin:

1. cube

2. earthdistance

3. to install these two plugin, run command below in psql

   CREATE EXTENSION cube;  
   CREATE EXTENSION earthdistance; 
   

how to save location json record

{"location": [30.2, 199.4]}  # just keep a latitude, longitude list

Demo

Article.objects.filter(data__location__near=[39.9, 116.4,5000]) # latitude，longitude，search range

or

Article.objects.filter(data__location__near='39.9,116.4,5000') # latitude，longitude, search range

Alert: if you don't pass exact number of params, this filter will not be used

for more earthdistance, see Postgresql Earthdistance Documentation

---

#####For more information about raw jsonb operation, please see PostgreSQL Documentation