A Fast and Flexible Document Database by Combining Features of Ruby and PostgreSQL with PLV8
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


Welcome to PostJson

PostJson is everything you expect from ActiveRecord and PostgreSQL with the added power and dynamic nature of a document database (Free as a bird! No schemas!).

PostJson combines features of Ruby, ActiveRecord and PostgreSQL to provide a great document database by taking advantage of PostgreSQL 9.2+ support for JavaScript (Google's V8 engine). We started the work on PostJson, because we love document databases and PostgreSQL.

README contents


Add the gem to your Gemfile:

gem 'post_json'


$ bundle install

Run the generator and migrate the db:

$ rails g post_json:install
$ rake db:migrate

That's it!

(See POSTGRESQL_INSTALL.md if you need the install instructions for PostgreSQL with PLV8)


PostJson also tries hard to respect the ActiveRecord API, so, if you have experience with ActiveRecord, the model methods work as you would expect.


All PostJson models represent a collection.

class Person < PostJson::Collection["people"]
me = Person.create(name: "Jacob")

Notice you don't have to define model attributes anywhere!

As you can see, this is very similar to a standard ActiveRecord model. PostJson::Collection["people"] inherits from PostJson::Base, which, in turn, inherits from ActiveRecord::Base. This is part of the reason the Person model will seem so familiar.

You can also skip the creation of a class:

people = PostJson::Collection["people"]
me = people.create(name: "Jacob")


Use standard ActiveRecord validations in your models:

class Person < PostJson::Collection["people"]
  validates :name, presence: true

Read the Rails guide about validation if you need more information.


me = Person.create(name: "Jacob", details: {age: 33})

# PostJson supports filtering on nested attributes
also_me_1 = Person.where(details: {age: 33}).first
also_me_2 = Person.where("details.age" => 33).first

# It is possible to use a pure JavaScript function for selecting documents
also_me_3 = Person.where("function(doc) { return doc.details.age == 33; }").first

# It is also possible to write real SQL queries. Just prefix the JSON attributes with `json_`
also_me_4 = Person.where("json_details.age = ?", 33).first

Accessing attributes

person = Person.create(name: "Jacob")
puts person.name            # => "Jacob"
puts person.name_was        # => "Jacob"
puts person.name_changed?   # => false
puts person.name_change     # => nil

person.name = "Martin"
puts person.name_was        # => "Jacob"
puts person.name            # => "Martin"
puts person.name_changed?   # => true
puts person.name_change     # => ["Jacob", "Martin"]

puts person.name            # => "Martin"
puts person.name_was        # => "Martin"
puts person.name_changed?   # => false
puts person.name_change     # => nil

Transformation with select

The select method allows you to transform a collection of documents into an array of hashes that contain only the attributes you want. The hash passed to select maps keys to selectors of arbitrary depth.

In this example we only want the 'name' and 'age' attributes from the Person but 'age' is nested under 'details'.

# create a person with age nested under details
me = Person.create(name: "Jacob", details: {age: 33})

# the dot (.) signifies that the selector is looking for a nested attribute
other_me = Person.limit(1).select({name: "name", age: "details.age"}).first

puts other_me   
# => {name: "Jacob", age: 33}


JSON does not natively support dates so Time and DateTime values are serialized as strings with the format of strftime('%Y-%m-%dT%H:%M:%S.%LZ'). PostJson will parse an attribute's value to a DateTime object if the value is a string and matches the serialization format.

me = Person.create(name: "Jacob", nested: {now: Time.now})

# `now` is a regular Time instance before being saved
puts me.attributes
# => {"name"=>"Jacob", "nested"=>{"now"=>2013-10-24 16:15:05 +0200}, "id"=>"fb9ef4bb-1441-4392-a95d-6402f72829db", "version"=>1, "created_at"=>Thu, 24 Oct 2013 14:15:05 UTC +00:00, "updated_at"=>Thu, 24 Oct 2013 14:15:05 UTC +00:00}


# after save `now` is the formatted string
puts me.attributes
# => {"name"=>"Jacob", "nested"=>{"now"=>"2013-10-24T14:15:05.783Z"}, "id"=>"fb9ef4bb-1441-4392-a95d-6402f72829db", "version"=>1, "created_at"=>"2013-10-24T14:15:05.831Z", "updated_at"=>"2013-10-24T14:15:05.831Z"}

Supported methods

all, any?, blank?, count, delete, delete_all, destroy, destroy_all, each, empty?, except, exists?, find, find_by, find_by!, find_each, find_in_batches, first, first!, first_or_create, first_or_initialize, ids, last, limit, load, many?, offset, only, order, pluck, reorder, reverse_order, select, size, take, take!, to_a, to_sql, and where.

We also added page(page, per_page), which translate into offset((page-1)*per_page).limit(per_page).

Configuration Options

PostJson.setup "people" do |collection|
  collection.record_timestamps = true                           # default is 'true'
  collection.created_at_attribute_name = "created_at"           # default is 'created_at'
  collection.updated_at_attribute_name = "updated_at"           # default is 'updated_at'
  collection.include_version_number = true                      # default is 'true'
  collection.version_attribute_name = "version"                 # default is 'version'
  collection.use_dynamic_index = true                           # default is 'true'
  collection.create_dynamic_index_milliseconds_threshold = 50   # default is '50'

For a Rails project this configuration could go in an initializer (config/initializers/post_json.rb).


As an example of PostJson performance we created 100,000 documents on a virtual machine running on a 3 year old laptop:

test_model = PostJson::Collection["test"]
100000.times { test_model.create(content: SecureRandom.uuid) }
content = test_model.last.content
result = test_model.where(content: content).count
# Rails debug duration was 975.5ms

Not bad but we know it can get better with some very specific indexing. After an index is created on test.content the performance is greatly improved:

result = test_model.where(content: content).count
# Rails debug duration was 1.5ms

You could create this index manually but PostJson has a feature called Dynamic Index that is enabled by default to work behind the scenes automatically creating indexes on slow queries.

Dynamic Indexes

Dynamic Index is a feature of PostJson that will measure the duration of each SELECT query and instruct PostgreSQL to create an index if the query duration is above a specified threshold. Since most applications perform the same queries over and over again we think you'll find this useful.

Each collection (like PostJson::Collection["people"] above) has two index attributes:

  • use_dynamic_index (default: true)
  • create_dynamic_index_milliseconds_threshold (default: 50)


PostJson::Collection["people"].where(name: "Jacob").count

# => query duration > 50ms

PostJson will check for an index on name and create it if it doesn't exist.

Index configuration

class Person < PostJson::Collection["people"]
  self.create_dynamic_index_milliseconds_threshold = 75


PostJson::Collection["people"].create_dynamic_index_milliseconds_threshold = 75

Do not set the dynamic index threshold too low as PostJson will try to create an index for every query. A threshold of 1 millisecond would be less than the duration of almost all queries.

Manual creation of index

class Person < PostJson::Collection["people"]
  self.ensure_dynamic_index("name", "details.age")


PostJson::Collection["people"].ensure_dynamic_index("name", "details.age")

List existing indexes

puts Person.existing_dynamic_indexes
# => ["name", "details.age"]


puts PostJson::Collection["people"].existing_dynamic_indexes
# => ["name", "details.age"]

Destroying an index




Primary Keys

PostJson assigns UUID as primary key (id):

me = Person.create(name: "Jacob")

puts me.id    
# => "297a2500-a456-459b-b3e9-e876f59602c2"

or you can set it directly:

john_doe = Person.create(id: "John Doe")

The primary key is downcased when doing a query or finding records:

found = Person.where(id: "JOhN DoE").first

puts found.attributes
# => {"id"=>"John Doe", "version"=>1, "created_at"=>"2013-10-22T10:42:26.190Z", "updated_at"=>"2013-10-22T10:42:26.190Z"}
found_again = Person.find("JOhN DoE")

puts found_again.attributes
# => {"id"=>"John Doe", "version"=>1, "created_at"=>"2013-10-22T10:42:26.190Z", "updated_at"=>"2013-10-22T10:42:26.190Z"}

Migrating to PostJson

Given a User model:

class User < ActiveRecord::Base

Add each user to the new users collection:

PostJson::Collection["users"].transaction do
  User.all.find_each do |user|

Then replace ActiveRecord::Base with the PostJson::Collection class:

class User < PostJson::Collection["users"]

Users will have the exact same content, including their primary keys (id).


Please note the roadmap might change as we move forward.

Version 2.0: Reboot – Close To The Metal


PostJson should be an add-on to your Rails project and not a replacement for ActiveRecord models. To accomplish this, we plan to split the implementation into components and concerns. This will make it possible to include as much (or as little) of PostJson into your ActiveRecord models as you like.

For example, perhaps you want PostJson::Attributes:

class Person < ActiveRecord::Base
  # PostJson::Attributes hides the column assigned to `document_hash_column` and includes the dynamic attributes in results like `to_json`.  

  include PostJson::Attributes
  self.document_hash_column = "__dynamic_attributes"

We're excited to see how far we can take it to make everything re-usable in existing applications!

Collection tables

Version 1.x stores all documents in a table called post_json_documents. In Version 2, each collection will have its own table with the same name as the collection. Each collection's metadata (title, settings etc.) will be stored as a record in the dedicated table.

Version 2.1: JavaScript Bindings

PostJson should support JavaScript bindings to its collection methods for query and find. We plan to integrate the therubyracer gem to translate JavaScript queries to ActiveRecord queries allowing seamless integration with rich JavaScript clients.

Imagine a Rails controller's index method:

def index
  js_query = params[:query]
  puts js_query
  # => function(people) { return people.limit(10).where({'gender': 'male'}); }

  result = PostJson::Collection['people'].eval_js_query(js_query)
  render json: result

Version 2.2: Relations

PostJson should support relations between collections like has_many, has_one, and belongs_to as persistable queries being able to work as dependent associations. Relations should not be tied to class definitions. This will allow the creation of relations from client software and enable relations to be copied / included in backups and migrations.

Imagine we have two collection: customers and orders.


class Customer < PostJson::Collection["customers"]
  has_one :order  

The order relation will use the order_id attribute from the customer, look-up the orders collection, and find the correct document using that id.

has_one will support override of the foreign_key ('order_id'), collection_name ('orders'), and mark dependent as :delete or :destroy. It will also support alias with :as, :readonly and a -> { ... } block for specializing.


class Customer < PostJson::Collection["customers"]
  has_many :orders

The order relation will use the order_ids attribute from the customer, look-up the orders collection, and find all related order documents using where(id: order_ids).

has_many will support override of foreign_keys ('order_ids'), collection_name ('orders') and mark dependent as :delete_all or :destroy_all. It will also support alias with :as, :readonly and a -> { ... } block for specializing.

class Order < PostJson::Collection["orders"]
  belongs_to :customer

The customer relation will look-up a collection named customers and find a customer with the order_id attribute set to the current order primary key value.

belongs_to will support override of foreign_key ('order_id'), collection_name ('customers') and mark dependent as :delete or :destroy. It will also support alias with :as, :readonly, and a -> { ... } block for specializing.

PostJson will use ActiveSupport::Inflector to implement the naming conventions.

PostJson will serialize lambda (-> { ... } block for specializing) as part of the collection definitions.

Version 2.3: Versioning

The history of data has great potential. It should be as easy as possible to get a view of the past. PostJson should be able to store the history of each document, including the possibility of restoring a document's previous state as a new document, roll back a document to its previous state, and view the changes for each version of a document.

This should also work for an entire collection. PostJson should be able to create a new collection from an existing collection's previous state. It should also be possible to query and view the previous state of a collection without restoring it to a new collection first.

Version 2.4: Bulk Import

Importing data can often be boring and troublesome. PostJson should make it easy to setup a data transformation and use bulk inserts to improve performance.

Version 2.5: Export CSV and HTML

PostJson already supports transformation with select. PostJson should be able to return results as CSV and HTML.


CSV stores tabular data and is not compatible with JSON. PostJson will flatten the data:

{ parent: {child: 123} }

will be converted to

{ "parent.child" => 123 }


HTML will be rendered by templates. PostJson will support Mustache and store templates in the database. This will allow PostgreSQL to do the rendering and return results as strings.

Version 2.6: File Attachments

PostJson should be able to store files in a specialized files collection when it makes sense to store files in the database. Files can be attached to other collections by using a has_one or has_many relation.

Version 2.7: Unused Index Deletion

PostJson should provide automatic deletion of unused dynamic indexes as an optional feature.

version 3.x: Full Text Search

PostgreSQL has many great features to support Full Text Search and PostJson should take advantage of these.

The future

We would love to hear new ideas or random thoughts about PostJson.


  • PostgreSQL 9.2 or 9.3
  • PostgreSQL PLV8 extension.


PostJson is released under the MIT License. See the MIT-LICENSE file.

Want to contribute?

That's awesome, thank you!

Do you have an idea or suggestion? Please create an issue or send us an e-mail (hello@webnuts.com). We would be happy to implement right away.

You can also send us a pull request with your contribution.

Webnuts.com ##### Sponsored by Webnuts.com