Massive Record

Massive Record is a Ruby client for HBase. It provides a basic API through Thrift and an ORM with advanced features.

See introduction to HBase model architecture:
http://wiki.apache.org/hadoop/Hbase/HbaseArchitecture
Understanding terminology of Table / Row / Column family / Column / Cell:
http://jimbojw.com/wiki/index.php?title=Understanding_Hbase_and_BigTable

HBase requirement

MassiveRecord is following the Cloudera packages of HBase: http://www.cloudera.com

Currently, MassiveRecord is tested against HBase 0.90.3, which can be found at the following address: https://ccp.cloudera.com/display/SUPPORT/CDH3+Downloadable+Tarballs

Install HBase (OSX):
Download the package 'HBase 0.90.3+15.3' and extract it.
Start HBase using the following command:

path_to_hbase/bin/start-hbase.sh

Start Thrift (HBase service interface):

path_to_hbase/bin/hbase thrift -b 127.0.0.1 start

Installation

First of all: Please make sure you are using Ruby 1.9.2. For now, we are only ensuring that Massive Record works on that Ruby version, and we know it has some problems with 1.8.7.

gem install massive_record

Ruby on Rails

Add the following Gems in your Gemfile:

gem 'massive_record'

Create an config/hbase.yml file with the following content:

defaults: &defaults
  host: somewhere.compute.amazonaws.com # No 'http', it's a Thrift connection
  port: 9090

development:
  <<: *defaults

test:
  <<: *defaults

production:
  <<: *defaults

Usage

There are two ways for using the Massive Record library. At the highest level we have ORM. This is Active Model compliant and makes it easy to use. The second way of doing things is working directly against the adapter.

ORM

Both MassiveRecord::ORM::Table and MassiveRecord::ORM::Embedded do now have some functionality which you can expect from an ORM. This includes:

  • An initializer which takes attribute hash and assigns them to your object.
  • Write and read methods for the attributes
  • Validations, as you expect from an ActiveRecord.
  • Callbacks, as you expect from an ActiveRecord.
  • Information about changes on attributes.
  • Casting of attributes
  • Serialization of array / hashes
  • Timestamps like created_at and updated_at. Updated at will always be available, created_at must be defined. See example down:
  • Finder scopes. Like: Person.select(:only_columns_from_this_family).limit(10).collect(&:name)
  • Ability to set a default scope.
  • Time zone aware time attributes.
  • Basic instrumentation and logging of query times.
  • Attribute mass assignment security.

Tables also have:

  • Persistencey method calls like create, save and destroy (but they do not actually save things to hbase)
  • Easy access to adapter's connection via Person.connection
  • Easy access to adapter's hbase table via Person.table
  • Finder method, like Person.find("an_id"), Person.find("id1", "id2"), Person.all etc
  • Save / update methods
  • Auto-creation of table and column families on save if table does not exists.
  • Destroy records
  • Relations: Both references to other tables and simple embedded records. See MassiveRecord::ORM::Relations::Interface ClassMethods for documentation
  • Observable. See MassiveRecord::ORM::Observer. If you know how to use ActiveRecord's observer you know how to use this one.
  • IdentityMap (when enabled)

Here are some examples setting up models:

class Person < MassiveRecord::ORM::Table
  references_one :boss, :class_name => "Person", :store_in => :info
  references_one :attachment, :polymorphic => true
  references_many :friends, :store_in => :info
  references_many :blog_posts, :records_starts_from => :posts_start_id

  embeds_many :addresses

  default_scope select(:info)

  column_family :info do
    field :name
    field :email
    field :phone_number
    field :points, :integer, :default => 0
    field :date_of_birth, :date, :allow_nil => false # Defaults to today
    field :newsletter, :boolean, :default => false
    field :type # Used for single table inheritance
    field :in_the_future, :time, :default => Proc.new { 2.hours.from_now }
    field :hobbies, :array, :allow_nil => false # Default to empty array

    timestamps # ..or field :created_at, :time
  end

  column_family :misc do
    field :with_a_lot_of_uninteresting_data
  end

  attr_accessible :name, :email, :phone_number, :date_of_birth

  validates_presence_of :name, :email
  validates_format_of :email, :with => /\A([^@\s]+)@((?:[-a-z0-9]+\.)+[a-z]{2,})\Z/i

  # Returns the id the scanner should start from in the BlogPost table
  # to fetch blog posts related to this person
  def posts_start_id
    id+'-'
  end
end

class Friend < Person
  # This one will be stored in Person's table with it's type set to Friend.
  # Calling Person.all will return object back as a Friend.
end

class PersonObserver < MassiveRecord::ORM::Observer
  def after_create(person_created)
    # Do something smart with that person
  end
end

class Address < MassiveRecord::ORM::Embedded
  embedded_in :person

  field :street
  field :number, :integer
  field :nice_place, :boolean, :default => true
end

class BlogPost < MassiveRecord::ORM::Embedded
  references_one :author, :class_name => "Person", :store_in => :info

  field :title
  field :content

  private

  # Set yourself an ID to your model
  def default_id
    "#{author_id}|#{Time.now.strftime("%Y-%m-%d-%k-%M")}"
  end
end

Perform requests:

# Fetch an object
u = User.find("45")

# Blog posts associated
u.blog_posts

# Blog posts associated during May 2011
u.blog_posts(:starts_with => "45-2011-05") # user_id - year - month

# Blog posts from May 2011
u.blog_posts(:offset => "45-2011-05")

# Only five blog posts
u.blog_posts(:limit => 5)

You can find a small example application here: https://github.com/thhermansen/massive_record_test_app

We have developed some gems which adds support for MassiveRecord. These are:

ORM Adapter

https://github.com/CompanyBook/orm_adapter Used by Devise. I guess we'll might release the code used to get Devise support in MR.

Database Cleaner

https://github.com/CompanyBook/database_cleaner User by for instance Cucumber and ourself with Rspec.

Sunspot Rails

https://github.com/CompanyBook/sunspot_massive_record Makes it easier to make things searchable with solr.

Wrapper (adapter) API

You can, if you'd like, work directly against the adapter. It is however adviced to use the ORM as the interface to the adapter is not yet very well defined.

# Init a new connection with HBase
conn = MassiveRecord::Wrapper::Connection.new(:host => 'localhost', :port => 9090)
conn.open

# OR init a connection using the config/hbase.yml file with Rails
conn = MassiveRecord::Wrapper::Base.connection

# Fetch tables name
conn.tables # => ["companies", "news", "webpages"]

# Init a table
table = MassiveRecord::Wrapper::Table.new(conn, :people)

# Add a column family
column = MassiveRecord::Wrapper::ColumnFamily.new(:info)
table.column_families.push(column)

# Or bulk add column families
table.create_column_families([:friends, :misc])

# Create the table
table.save # will raise an exception if the table already exists

# Fetch column families from the database
table.fetch_column_families # => [ColumnFamily#RTY4424, ColumnFamily#R475424, ColumnFamily#GHJ9424]
table.column_families.collect(&:name) # => ["info", "friends", "misc"]

# Add a new row
row = MassiveRecord::Wrapper::Row.new
row.id = "my_unique_id"
row.values = { :info => { :first_name => "H", :last_name => "Base", :email => "[email protected]" } }
row.table = table
row.save

# Fetch rows
table.first # => MassiveRecord#ID1
table.all(:limit => 10) # => [MassiveRecord#ID1, MassiveRecord#ID2, ...]
table.find("ID2") # => MassiveRecord#ID2
table.find(["ID1", "ID2"]) # => [MassiveRecord#ID1, MassiveRecord#ID2]
table.all(:limit => 3, :starts_with => "ID2") # => [MassiveRecord#ID2, MassiveRecord#ID3, MassiveRecord#ID4]

# Manipulate rows
table.first.destroy # => true

# Remove the table
table.destroy

Planned work

  • Cache the decoded values of attributes, not use the value_is_already_decoded?. This will fix possible problem with YAML as coder backend.
  • Implement other Adapters, for instance using jruby and the Java API.

Contribute

If you want to contribute feel free to fork this project :-) Make a feature branch, write test, implement and make a pull request.

Getting started

git clone git://github.com/CompanyBook/massive_record.git (or the address to your fork)
cd massive_record
bundle install

Next up you need to add a config.yml file inside of spec/ which contains something like: host: url.to-a.thrift.server port: 9090 table: massive_record_test_table

You should now be able to run rspec spec/

Play with it in the console

Checkout the massive_record project and install it as a Gem :

cd massive_record/
bundle console
ruby-1.9.2-p0 > Bundler.require
 => [
      <Bundler::Dependency type=:runtime name="massive_record" requirements=">= 0">,
      <Bundler::Dependency type=:runtime name="thrift" requirements=">= 0.5.0">,
      <Bundler::Dependency type=:runtime name="activesupport" requirements=">= 0">,
      <Bundler::Dependency type=:runtime name="activemodel" requirements=">= 0">,
      <Bundler::Dependency type=:runtime name="rspec" requirements=">= 2.1.0">
    ]
ruby-1.9.2-p0 > MassiveRecord::VERSION
 => "0.0.1"

Clean HBase database between each test

We have created a helper module MassiveRecord::Rspec::SimpleDatabaseCleaner which, when included into rspec tests, will clean the database for ORM records between each test case. You can also take a look into spec/support/mock_massive_record_connection.rb for some functionality which will mock a hbase connection making it easier (faster) to test code where no real database is needed.

More Information and Resources

Thrift API

Ruby Library using the HBase Thrift API. http://wiki.apache.org/hadoop/Hbase/ThriftApi

The generated Ruby files can be found under lib/massive_record/thrift/
The whole API (CRUD and more) is present in the Client object (Apache::Hadoop::Hbase::Thrift::Hbase::Client).
The client can be easily initialized using the MassiveRecord connection :

conn = MassiveRecord::Wrapper::Connection.new(:host => 'localhost', :port => 9090)
conn.open

client = conn.client
# Do whatever you want with the client object

Q&A

How to add a new column family to an existing table?

# Connect to the HBase console on the server itself and enter the following code :
disable 'companies'
alter 'companies', { NAME => 'new_collumn_familiy' }
enable 'companies'

Copyright (c) 2011 Companybook, released under the MIT license