esengine - The Elasticsearch Object Document Mapper

PyPI versions downloads Travis CI Coverage Status Code Health

esengine is an ODM (Object Document Mapper) it maps Python classes in to Elasticsearch index/doc_type and object instances() in to Elasticsearch documents.


Out of the box ESengine takes care only of the Modeling and CRUD operations including:


ESengine does not communicate directly with ElasticSearch, it only creates the basic structure, To communicate it relies on an ES client providing the transport methods (index, delete, update etc).

ES client

ESengine does not enforce the use of the official ElasticSearch client, but you are encouraged to use it because it is well maintained and has the support to bulk operations. But you are free to use another client or create your own (useful for tests).

Querying the data

ESengine does not enforce or encourage you to use a DSL language for queries, out of the box you have to write the elasticsearch payload representation as a raw Python dictionary. However ESEngine comes with utils.payload helper module to help you building payloads in a less verbose and Pythonic way.

Why not elasticsearch_dsl?

ElasticSearch DSL is an excellent tool, a very nice effort by the maintainers of the official ES library, it is handy in most of the cases, but because it is built on top of operator overiding, sometimes leads to a confuse query building, sometimes it is better to write raw_queries or use a simpler payload builder having more control and visibility of what os being generated.

ElasticSearch_DSL as a high level abstraction promotes Think only of Python objects, dont't worry about Elastic queries while ESengine promotes Know well the Elastic queries and then write them as Python objects.

ElasticSearch_DSL is more powerful and more complete, tight more with ES specifications while ESEngine is simpler, lightweight shipping only the basics.

Project Stage

It is in beta-Release, working in production, but missing a lot of features, you can help using, testing,, discussing or coding!

Getting started


ESengine needs a client to communicate with E.S, you can use one of the following:

Because of bulk operations you are recommendded to use elasticsearch-py (Official E.S Python library) so the instalation depends on the version of elasticsearch you are using.

in short

Install the client and then install ESEngine

For the latest use:

$ pip install elasticsearch
$ pip install esengine

Or install them together

Elasticsearch 2.x

pip install esengine[es2]

Elasticsearch 1.x

pip install esengine[es1]

Elasticsearch 0.90.x

pip install esengine[es0]

The above command will install esengine and the elasticsearch library specific for you ES version.


from elasticsearch import ElasticSearch
from esengine import Document, StringField

Defining a document

class Person(Document):
    # define _meta attributes
    _doctype = "person"  # optional, it can be set after using "having" method
    _index = "universe"  # optional, it can be set after using "having" method
    _es = ElasticSearch(host='host', port=port)  # optional, it can be explicit passed to methods

    # define fields
    name = StringField()

If you do not specify an "id" field, ESEngine will automatically add "id" as StringField. It is recommended that when specifying you use StringField for ids.


Base Fields

name = StringField()
age = IntegerField()
weight = FloatField()
factor = LongField()
active = BooleanField()
birthday = DateField()

Special Fields


A field to hold GeoPoint with modes dict|array|string and its mappings

class Obj(Document):
    location = GeoPointField(mode='dict')  # default
    # An object representation with lat and lon explicitly named

Obj.init() # important to put the proper mapping for geo location

obj = Obj()

obj.location = {"lat": 40.722, "lon": -73.989}}

class Obj(Document):
    location = GeoPointField(mode='string')
    # A string representation, with "lat,lon"

obj.location = "40.715, -74.011"

class Obj(Document):
    location = GeoPointField(mode='array')
    # An array representation with [lon,lat].

obj.location = [-73.983, 40.719]


A field to hold nested one-dimension objects, schema-less or with properties validation.

# accepts only dictionaries having strct "street" and "number" keys
address = ObjectField(properties={"street": "string", "number": "integer"})

# Accepts any Python dictionary
extravalues = ObjectField() 


A Field to hold arrays (python lists)

In the base, any field can accept multi parameter

colors = StringField(multi=True)   # accepts ["blue", "green", "yellow", ....]

But sometimes (specially for nested objects) it is better to be explicit, and also it generates a better mapping

# accepts an array of strings ["blue", "green", "yellow", ....]
colors = ArrayField(StringField()) 

It is available for any other field

locations = ArrayField(GeoPointField())
numbers = ArrayField(IntegerField())
fractions = ArrayField(FloatField())
addresses = ArrayField(ObjectField(properties={"street": "string", "number": "integer"}))
list_of_lists_of_strings = ArrayField(ArrayField(StringField()))


person = Person(id=1234, name="Gonzo")  # or pass .save(es=es_client_instance) if not specified in model 

Getting by id


filtering by IDS

ids = [1234, 5678, 9101]
power_trio = Person.filter(ids=ids)

filtering by fields



ESengine does not try to create abstraction for query building, by default ESengine only implements search transport receiving a raw ES query in form of a Python dictionary.

query = {
    "query": {
        "filtered": {
            "query": {
                "match_all": {}
            "filter": {
                "ids": {
                    "values": [1, 2]
}, size=10)

Getting all documents (match_all)


# with more arguments





A single document

A single document can be updated simply using the .save() method

person = Person.get(id=1234) = "Another Name"

Updating a Resultset

The Document methods .get, .filter and .search will return an instance of ResultSet object. This object is an Iterator containing the hits reached by the filtering or search process and exposes some CRUD methods[ update, delete and reload ] to deal with its results.

people = Person.filter(field='value')

When updating documents sometimes you need the changes done in the E.S index reflected in the objects of the ResultSet iterator, so you can use .reload method to perform that action.

The use of reload method

people = Person.filter(field='value')
print people
... <Resultset: [{'field': 'value', 'another_field': None}, 
                 {'field': 'value', 'another_field': None}]>

# Updating another field on both instances
print people
... <Resultset: [{'field': 'value', 'another_field': None}, {'field': 'value', 'another_field': None}]>

# Note that in E.S index the values weres changed but the current ResultSet is not updated by defaul
# you have to fire an update

print people
... <Resultset: [{'field': 'value', 'another_field': 'another_value'},
                 {'field': 'value', 'another_field': 'another_value'}]>

Deleting documents

A ResultSet

people = Person.all()

A single document


Bulk operations

ESEngine takes advantage of elasticsearch-py helpers for bulk actions, the ResultSet object uses bulk melhod to update and delete documents.

But you can use it in a explicit way using Document's update_all, save__all and delete_all methods.

Lets create a bunch of document instances

top_5_racing_bikers = []

for name in ['Eddy Merckx', 
             'Bernard Hinault', 
             'Jacques Anquetil', 
             'Sean Kelly', 
             'Lance Armstrong']:

Save it all


Using the create shortcut

The above could be achieved using create shortcut

A single
Person.create(name='Eddy Merckx', active=False)

Create will return the instance of the indexed Document

All using list comprehension
top_5_racing_bikers = [
    Person.create(name=name, active=False)
    for name in ['Eddy Merckx', 
                 'Bernard Hinault', 
                 'Jacques Anquetil', 
                 'Sean Kelly', 
                 'Lance Armstrong']

NOTE: .create method will automatically save the document to the index, and will not raise an error if there is a document with the same ID (if specified), it will update it acting as upsert.

Updating all

Turning the field active to True for all documents

Person.update_all(top_5_racing_bikes, active=True)

Deleting all


Chunck size

chunk_size is number of docs in one chunk sent to ES (default: 500) you can change using meta argument.

    top_5_racing_bikes, # the documents
    active=True,  # values to be changed
    meta={'chunk_size': 200}  # meta data passed to **bulk** operation    


Mapping and Mapping migrations

ESEngine does not saves mappings automatically, but it offers an utility to generate and save mappings on demand You can create a cron job to refresh mappings once a day or run it every time your model changes

Using the document
class Person(Document):
    # define _meta attributes
    _doctype = "person"  # optional, it can be set after using "having" method
    _index = "universe"  # optional, it can be set after using "having" method
    _es = ElasticSearch(host='host', port=port)  # optional, it can be explicit passed to methods

    # define fields
    name = StringField()
You can use init class method to initialize/update mappings, settings and analyzers
Person.init()  # if not defined in model, pass an **es=es_client** here

Include above in your the last line of your model files or cron jobs or migration scripts


Field Validator

To validate each field separately you can set a list of validators, each validator is a callable receiving field_name and value as arguments and should return None to be valid. If raise or return the data will be invalidated

from esengine.exceptions import ValidationError

def category_validator(field_name, value):
    # check if value is in valid categories
    if value not in ["primary", "secondary", ...]:
        raise ValidationError("Invalid category!!!")

class Obj(Document):
    category = StringField(validators=[category_validator])

obj = Obj()
obj.category = "another"
Traceback: ValidationError(....)
Document Validator

To validate the whole document you can set a list of validators, each validator is a callable receiving the document instance and should return None to be valid. If raise or return the data will be invalidated

from esengine.exceptions import ValidationError

def if_city_state_is_required(obj):
    if and not obj.state:
        raise ValidationError("If city is defined you should define state")

class Obj(Document):
    _validators = [if_city_state_is_required]

    city = StringField()
    state = StringField()

obj = Obj() = "Sao Paulo"
Traceback: ValidationError(....)


Sometimes you need to force indices-shards refresh for testing, you can use

# Will refresh all indices

Payload builder

Sometimes queries turns in to complex and verbose data structures, to help you (use with moderation) you can use Payload utils to build queries.

Example using a raw query:
query = {
    "query": {
        "filtered": {
            "query": {
                "match_all": {}
            "filter": {
                "ids": {
                    "values": [1, 2]
}, size=10)
Same example using payload utils
from esengine import Payload, Query, Filter
payload = Payload(query=Query.filtered(query=Query.match_all(), filter=Filter.ids([1, 2]))), size=10)

Payload utils exposes Payload, Query, Filter, Aggregate, Suggesters

You can also set model on payload initialization to create a more complete payload definition

from esengine import Payload, Query, Filter
payload = Payload(
    query=Query.filtered(query=Query.match_all(), filter=Filter.ids([1, 2]))
    sort={"name": {"order": "desc"}},
More examples

You can use Payload, Query or Filter direct in search

from esengine import Payload, Query, Filter[Query.match("name", "Gonzo")]))[1, 2, 3]))

Payload object is chainable so you can do:

payload = Payload(query=query).size(10).sort("field", order="desc") 
# or the equivalent


You can paginate a payload, lets say you have indexed 500 documents under 'test' category and now you need to retrieve 50 per page.

Result will be included in pagination.items

from esengine import Payload, Filter
from models import Doc

payload = Payload(Doc, filter=Filter.term('category', 'test'))

# Total documents

# Paginate it
current_page = 1  # you have to increase it on each pagination
pagination = payload.paginate(page=current_page, per_page=50)






for item in pagination.items:
    # do something with item

# Turn the page

current_page += 1
pagination = payload.paginate(page=current_page, per_page=50)

# Another option to move pages

pagination  = pagination.next_page()

pagination = pagination.prev_page()

# Turn the page in place


Create a paginator in Jinja template

So you want to create buttons for pagination in your jinja template

{% macro render_pagination(pagination, endpoint) %}
  <div class=pagination>
  {%- for page in pagination.iter_pages() %}
    {% if page %}
      {% if page != %}
        <a href="{{ url_for(endpoint, page=page) }}">{{ page }}</a>
      {% else %}
        <strong>{{ page }}</strong>
      {% endif %}
    {% else %}
      <span class=ellipsis>…</span>
    {% endif %}
  {%- endfor %}
{% endmacro %}


ESEngine is OpenSource! join us! Small Acts Manifesto

MADE WITH #LOVE AND #PYTHON (which is the same) AT CathoLabs