Build Status Test Coverage Code Climate Dependency Status

Elasticity is a model oriented approach to Elasticsearch. In simple words, a Document is represented by it's own class, similar to what ActiveRecord does for database rows.

In Elasticsearch terminology, a document is an entity stored in Elasticsearch and associated to an index. Whenever a search is performed, a collection of documents is returned.

Elasticity maps those documents into objects, providing a rich object representation of a document.

Installation

Add this line to your application's Gemfile:

gem 'es-elasticity', require: "elasticity"

And then execute:

$ bundle

Or install it yourself as:

$ gem install es-elasticity

Usage

Version Support

This gem has elasticsearch-ruby as a dependency. In order to use different versions of elasticsearch you will need to match your version of elasticsearch-ruby to the version of elasticsearch you want to use (see here. Elasticity should work across all versions of elastisearch-ruby, although they have not all been tested so there are likely edge cases.

Currently tests are run on CirlceCI against elasticsearch 7.17.3 with elasticsearch-ruby 7.17.1.

Configuration

It is recommended you use Typhoeus for HTTP connections to Elasticsearch.

# config/initializers/es-elasticity.rb
require 'typhoeus'
require 'typhoeus/adapters/faraday'

Elasticity.configure do |config|
  config.client = Elasticsearch::Client.new
  config.namespace = Rails.env.to_s.downcase
end

Document model definition

The first thing to do, is setup a model representing your documents. The class level represents the index, while the instance level represents each Document stored in the index. This is similar to how ActiveRecord maps tables vs rows.

class Search::User < Elasticity::Document
  configure do |c|
    # Defines how the index will be named, the final name
    # will depend on the stragy being used.
    c.index_base_name = "users"

    # Defines the document type that this class represents.
    c.document_type = "user"

    # Select which strategy should be used. AliasIndex uses two aliases
    # in order to support hot remapping of indexes. This is the recommended
    # strategy.
    c.strategy = Elasticity::Strategies::AliasIndex

    # Defines the mapping for this index/document_type.
    c.mapping  = {
      properties: {
        name: { type: "string" },
        birthdate: { type: "date" },
      }
    }
  end

  # Defines a search method.
  def self.adults
    date = Date.today - 21.years

    # This is the query that will be submitted to ES, same format ES would
    # expect, translated to a Ruby hash, note the pagination params.
    body = {
      from: 0,
      size: 10,
      filter: {
        range: { birthdate: { lte: date.iso8601 }},
      },
    }

    # Creates a search object from the body and return it.  The returned
    # object is a lazy evaluated search that behaves like a collection, being
    # automatically triggered when data is iterated over.
    self.search(body)
  end

  # All models automatically have the id attribute but you need to define the
  # other accessors so that they can be set and get properly.
  attr_accessor :name, :birthdate

  # to_document is the only required method that needs to be implemented so an
  # instance of this model can be indexed.
  def to_document
    {
      name: self.name,
      birthdate: self.birthdate.iso8601
    }
  end
end

Indexing

An instance of the model is an in-memory representation of a Document. The document can be stored on the index by calling the update method.

# Creates a new document on the index
u = Search::User.new(id: 1, name: "John", birthdate: Date.civil(1985, 10, 31))
u.update

# Updates the same document on the index
u.name = "Jonh Jon"
u.update

If you need to index a collection of documents, you can use bulk_index:

users = [
  Search::User.new(id: 1, name: "John", birthdate: Date.civil(1985, 10, 31)),
  Search::User.new(id: 2, name: "Mary", birthdate: Date.civil(1986, 9, 24)),
]

Search::User.bulk_index(users)

Individual Attribute Indexing

If you you'd like to update a single attribute of the document instead of updating the entire document, you can use the bulk_update api.

documents = [
  { _id: 1, attr_name: "attr_name", attr_vale: "attr_value" },
  { _id: 2, attr_name: "attr_name", attr_vale: "attr_value" }
]

Search::User.bulk_update(documents)

Searching

Class methods have access to the search method, which returns a lazy evaluated search. That means that the search will only be performed when the data is necessary, not when the search method is called.

The search object implements Enumerable, so it can be treated as a collection:

# Get the search object, which is an instance of `Elasticity::DocumentSearchProxy`.
# Search is not performed until data is accessed.
adults = Search::User.adults

# Iterating over the results will trigger the query
adults.each do |user|
  # do something with user
end

# Or get the count
adults.count

It also has some pretty interesting methods that affects the way the query is performed. Here is a list of available search types:

# Returns an array of document instances, this is the default and what the
# enumerable methods will delegate to.
adults.documents

# Returns an array of hashes representing the documents.
adults.document_hashes

# Performs the search using scan&scroll. It returns a cursor that will lazily
# fetch all the pages of the search. It can be iterated by batch/page or by
# document.
cursor = adults.scan_documents
cursor.each_batch { |batch| ... }
cursor.each { |doc| ... }

# Lastly, a search that maps back to an ActiveRecord::Relation.
adults = adults.active_records(User)

Search Args

explain: true

For search definitions we support passing { explain: true } to the search as a second argument in order to surface the reason a search result was returned.

# example in single search
search_results_with_explanation = SearchDoc::A.search(query_body, { explain: true }).search_results

# In multisearch
search_a = SearchDoc::A.search(query_body, { explain: true })
search_b = SearchDoc::B.search(query_body, { explain: true })
search_c = SearchDoc::C.search(query_body, { explain: true })

multi = Elasticity::MultiSearch.new do |m|
  m.add(:a, search_a, documents: ::SearchDoc::A)
  m.add(:b, search_b, documents: ::SearchDoc::B)
  m.add(:c, search_c, documents: ::SearchDoc::C)
end

For more information about the active_records method, read ActiveRecord integration.

Segmented Documents

The idea of segmented documents is that documents of the same type/class can be distributed over separate segments, which are backed by separated indexes. This is good for manually sharding documents into their own indexes. For example, an application that supports multiple clients/organizations might want to separate the documents for each organization under separate indexes, making it easier to delete the data and isolate the documents.

Using this feature is very easy and very similar to traditional documents. The only difference is that your document class should inherit from Elasticity::SegmentedDocument. Adjusting the definition that we had before:

class Search::User < Elasticity::SegmentedDocument
  configure do |c|
    # Defines how the index will be named, the final name
    # will depend on the strategy being used.
    c.index_base_name = "users"

    # Defines the document type that this class represents.
    c.document_type = "user"

    # Select which strategy should be used. AliasIndex uses two aliases
    # in order to support hot remapping of indexes. This is the recommended
    # strategy.
    c.strategy = Elasticity::Strategies::AliasIndex

    # Defines the mapping for this index/document_type.
    c.mapping  = {
      properties: {
        name: { type: "string" },
        birthdate: { type: "date" },
      }
    }
  end

  # Defines a search method.
  def self.adults
    date = Date.today - 21.years

    # This is the query that will be submitted to ES, same format ES would
    # expect, translated to a Ruby hash, note the pagination params.
    body = {
      from: 0,
      size: 10,
      filter: {
        range: { birthdate: { lte: date.iso8601 }},
      },
    }

    # Creates a search object from the body and return it.  The returned
    # object is a lazy evaluated search that behaves like a collection, being
    # automatically triggered when data is iterated over.
    self.search(body)
  end

  # All models automatically have the id attribute but you need to define the
  # other accessors so that they can be set and get properly.
  attr_accessor :name, :birthdate

  # to_document is the only required method that needs to be implemented so an
  # instance of this model can be indexed.
  def to_document
    {
      name: self.name,
      birthdate: self.birthdate.iso8601
    }
  end
end

This class on itself can't be queried or manipulated directly. Trying to call the adults search method, one would get an error: NoMethodError: undefined method 'search' for #<Class:0x007fd582933460>. To be able to call any method you first need to define the segment that you want to use, which can easily be done by calling the method segment, which will return a class derived from the base class that contains all the available methods:

users = Search::User.segment("doximity.com")
users.create_index

# users is a dynamically defined class that inherits from the Search::User class,
# therefore having all the necessary methods defined just properly.
users           # => Search::User{"doximity.com"}
users.class     # => Class
users.ancestors # => [Search::User{"doximity.com"}, User, Elasticity::SegmentedDocument, ...]

john = users.new(name: "John", birthdate: Date.civil(1985, 10, 31))
john # => #<Search::User{"doximity.com"}:0x81a3ab6a0dea @name="John" @birthdate=Thu, 31 Oct 1985>
john.update

users.adults.to_a # => [#<Search::User{"doximity.com"}:0x819cc5a50cd5 @_id="AVCHLz5JyttLSz7M-tRI" @name="John" @birthdate="1985-10-31" @highlighted=nil>]

Strategies and Hot-remapping

Strategies define how index creation and index operation happens on the lower level. Basically it defines the structure that backs the document model. Currently, there are two strategies available: single-index and alias-index.

The single-index strategy is the most straightforward one. It causes one index to be created and any operation will be performed directly on that index. It's very simple but it has the downside of being a lot harder to update existing mapping since you'll have to drop the index and recreate from scratch.

The alias-index strategy is a bit more complex but it allows for seamless hot remapping. It works by creating an index and two aliases pointing to that index. Any operation is performed on the aliases rather than the index, which allows hot swapping due atomic aliases updates.

Here is what it looks like:

|¯¯¯¯¯¯¯¯¯¯¯¯¯|
|  MainAlias  |---------|
|_____________|         |------------> |¯¯¯¯¯¯¯¯¯¯¯¯¯|
                                       |    Index    |
|¯¯¯¯¯¯¯¯¯¯¯¯¯|         |------------> |_____________|
| UpdateAlias |---------|
|_____________|

Every time a search operation is performed, it is performed against the main alias; when an update operation is performed, it is performed against the update alias; and, when a delete operation is performed, it is performed against the indexes pointed by both aliases.

When the mapping needs to change, a hot remapping can be performed by doing the following:

  1. Create a new index with the new mapping;
  2. change the update alias to point to the new index, and change main alias to point to both indexes; at this point it will look something like this:
|¯¯¯¯¯¯¯¯¯¯¯¯¯|----------------------> |¯¯¯¯¯¯¯¯¯¯¯¯¯|
|  MainAlias  |                        |  Old Index  |
|_____________|----------|             |_____________|
                         |
|¯¯¯¯¯¯¯¯¯¯¯¯¯|          |-----------> |¯¯¯¯¯¯¯¯¯¯¯¯¯|
| UpdateAlias |----------------------> |  New Index  |
|_____________|                        |_____________|
  1. iterate over all documents on the old index, copying them to the new index;
  2. change aliases to point only to the new index;
  3. delete the old index.

This is a simplified version, there are other things that happen to ensure consistency and avoid race conditions. For full understanding see Elasticity::Strategies::AliasIndex#remap.

ActiveRecord integration

ActiveRecord integration is mainly a set of conventions rather than implementation, with the exception of one method that allows mapping documents back to a relation. Here is the list of conventions:

  • have a class method on the document called from_active_record that creates a document object from the active record object;
  • have a class method on the Document for rebuilding the index from the records;
  • have an after_save and an after_destroy callbacks on the ActiveRecord model;

For example:

class User < ActiveRecord::Base
  after_save    :update_index_document
  after_destroy :delete_index_document

  def update_index_document
    Search::User.from_active_record(self).update
  end

  def delete_index_document
    Search::User.delete(self.id)
  end
end

class Search::User < Elasticity::Document
  # ... configuration

  def self.from_active_record(ar)
    new(name: ar.name, birthdate: ar.birthdate)
  end

  def self.rebuild_index
    self.recreate_index

    User.find_in_batches do |batch|
      documents = batch.map { |record| from_active_record(record) }
      self.bulk_index(documents)
    end
  end
end

This makes the code very clear in intent, easier to see when and how things happen and under the developer control, keeping both parts very decoupled.

The only ActiveRecord specific utility this library have is a way to lazily map a Elasticsearch search to an ActiveRecord relation.

To extend on the previous example, imagine the Search::User class also have the following simple search method.

def self.adults
  date = Date.today - 21.years

  body = {
    filter: {
      { range: { birthdate: { gte: date.iso8601 }}},
    },
  }

  self.search(body)
end

Because the return of that method is a lazy-evaluated search, it allows specific search strategies to be used, one of them being ActiveRecord specific:

adults = Search::User.adults.active_records(User)
adults.class # => ActiveRecord::Relation
adults.all   # => [#<User: id: 1, name: "John", birthdate: 1985-10-31>, ...]

Note that the method takes a relation and not a class, so the following is also possible:

Search::User.adults.active_records(User.where(active: true))

Upgrading from 0.7.0 to 0.8.0

The default persistence strategy changed from SingleIndex to AliasIndex in version 0.8.0 Add the following to your Document configuration to maintain the legacy behaviour.

  c.strategy = Elasticity::Strategies::SingleIndex

Roadmap

  • [ ] Define from_active_record interface
  • [ ] Write more detailed documentation section for:
    • [ ] Model definition
    • [ ] Indexing, Bulk Indexing and Delete By Query
    • [ ] Search and Multi Search
    • [ ] ActiveRecord integration
  • [ ] Support for multiple document types
  • [ ] Get rid of to_document, generate automatically based on attributes
  • [ ] Add some delegations on Document to Index

Contributing

  1. See CONTRIBUTING.md
  2. Fork it ( https://github.com/[my-github-username]/elasticity/fork )
  3. Create your feature branch (git checkout -b my-new-feature)
  4. Commit your changes (git commit -am 'Add some feature')
  5. Push to the branch (git push origin my-new-feature)
  6. Create a new Pull Request

Gem documentation

You can find the documentation by going to CircleCI, looking for the build jobs, going to Artifacts and clicking on index.html. A visual guide on this can be found in our wiki at Gems Development: Where to find documentation for our gems.

Gem development

After checking out the repo, run bundle install to install dependencies. Then, run rake spec to run the tests. You can also run bundle console for an interactive prompt that will allow you to experiment.

This repository uses a gem publishing mechanism on the CI configuration, meaning most work related with cutting a new version is done automatically.

To release a new version, follow the wiki instructions.

License

MPN is licensed under an Apache 2 license. Contributors are required to sign an contributor license agreement. See LICENSE.txt and CONTRIBUTING.md for more information.