Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


Cequel is a CQL (Doc: Datastax or Apache) query builder and object-row mapper for Cassandra.

Build Status

The library consists of two layers. The lower Cequel layer is a lightweight CQL query builder, which uses chained scopes to construct CQL queries, execute them against your Cassandra instance, and return results in friendly form. The Cequel::Model layer implements an object-row mapper on top of Cequel, with full ActiveModel integration and an interface that conforms to established patterns for Ruby persistence layers (e.g. ActiveRecord).

The lower Cequel layer is heavily inspired by the excellent Sequel library; Cequel::Model more closely follows the form of ActiveRecord.


To use only the lower-level Cequel query builder, add the following to your Gemfile:

gem 'cequel'

For Cequel::Model, instead add:

gem 'cequel', :require => 'cequel/model'

Rails integration

Cequel and Cequel::Model do not require Rails, but if you are using Rails, you will need version 3.2+. Cequel::Model will read from the configuration file config/cequel.yml if it is present. A simple example configuration would look like this:

  host: ''
  keyspace: myapp_development

    - ''
    - ''
    - ''
  keyspace: myapp_production
    retries: 10
    timeout: 15
    connect_timeout: 15

Cequel Query Builder

To connect to a keyspace, use Cequel.connect:

cassandra = Cequel.connect(
  :host => '',
  :keyspace => 'myapp_development'

Column family handles are referenced as follows:

posts = cassandra[:posts]

Reading Data

To select data, you can form a query using the familiar chained scope pattern:

posts = cassandra[:posts].select(:title).
  where(:id => 1).

titles = { |post| post[:title] }

When working with wide rows, you often want to select a range of columns rather than a predefined set:

# Select columns 1-5

# Select columns 5 and up
cassandra[:posts].select(:from => 5)

# Select columns up to 5
cassandra[:posts].select(:to => 5)

# Select the first 8 columns (in natural order of column type)
cassandra[:posts].select(:first => 8)

# Select the last 6 columns
cassandra[:posts].select(:last => 6)

# Combine ranges and limits
cassandra[:posts].select(1..100, :first => 5)

# Or open-ended ranges and limits
cassandra[:posts].select(:first => 5, :from => 20)

Data set scopes also support the first and count methods.


Cequel scopes support a subquery-like syntax, which can be used to populate the scope of an outer query with the results of an inner query:

cassandra[:blogs].where(:id => cassandra[:posts].select(:blog_id))

This actually performs two queries to Cassandra, since CQL itself does not support subqueries.

Writing data

To insert data, use insert:

cassandra[:posts].insert(:id => 1, :title => 'My Post', :body => 'Some wisdom')

You can control consistency, timestamp, and time to live by passing a second options hash to insert:

  {:id => 1, :title => 'My Post', :body => 'Some wisdom'},
  :consistency => :quorum, :ttl => 10.minutes, :timestamp =>

To update data, construct a scope and then call update with the columns to write:

cassandra[:posts].where(:id => [1, 2]).update(:title => 'My Post')

To update a counter column, use the increment or decrement method:

cassandra[:comment_counts].where(:id => 1).increment(post_id => 1)
cassandra[:comment_counts].where(:id => 1).decrement(post2_id => 4)

To delete entire rows, call the delete method; to delete certain columns from a row, pass those columns to delete:

# delete rows 1 and 2 entirely
cassandra[:posts].where(:id => [1, 2]).delete
# delete title column from rows 1 and 2
cassandra[:posts].where(:id => [1, 2]).delete(:title)


Cequel::Model is a higher-level object-row mapper built on top of the low-level functionality described above. Cequel models are ActiveModel-compliant and generally follow ActiveRecord-like patterns.

Keyspace setup and migrations

The current version of Cequel does not provide built-in functionality for schema creation and modification, but ActiveRecord-like migrations for Cequel are available via the cequel-migrations-rails library.

The forthcoming release of Cequel will support full schema introspection and modification, and will also provide auto-migration functionality for models.

Defining a model

Cequel models include the Cequel::Model module; the example below demonstrates most of what's available for defining a model:

class Post

  include Cequel::Model
  include Cequel::Model::Timestamps

  key :id, :uuid
  column :title, :text
  column :body, :text

  belongs_to :blog
  has_many :comments

  attr_accessible :title, :body

  validates :title, :body, :blog_id, :presence => true

  after_create :post_to_twitter

  default_scope limit(100)


  def generate_key


Working with models: The non-surprising parts

Model behavior will be largely familiar to anyone who has worked with ActiveRecord or another ActiveRecord-inspired object mapper. All of these operations work pretty much exactly as you'd expect:

# Initialize a new instance

# Initialize a new instance with some attributes => 'Hey')

# Initialize a new instance and set some properties do |post|
  post.title = 'Hey'

# Create a new instance with attributes and save it
Post.create(:title => 'Hey')

# Create a new instance with attributes and save it violently
Post.create!(:title => 'Hey')

# Update an instance
post.title = 'New title'

# Destroy an instance

# Find an instance by key

# Find an instance by magic

# Find lots of instances by magic

# Find or initialize an instance by magic
Post.find_or_initialize_by_title('My Post')

# Find or initialize an instance by magic with some extra attributes
Post.find_or_initialize_by_title(:title => 'My Post', :body => 'Read more')

# Of course, find_or_create_by works too
Post.find_or_create_by_title('My Post')

# Query by scopes => uuid).first

# Query by secondary indexes => blog_uuid).map { |post| post.title }

# This will execute three queries, because CQL secondary indexes don't play nice
# with IN restrictions. But it'll work:
  where(:blog_id => [blog_id1, blog_id2, blog_id3]).
  map { |post| post.title }

Working with models: The surprising parts

CQL is designed to be immediately familiar to those of us who are used to working with SQL, which is all of us. Cequel advances this spirit by providing an ActiveRecord-like mapping for CQL. However, Cassandra is very much not a relational database, so some behaviors can come as a surprise. Here's an overview.


CQL provides INSERT and UPDATE statements that look more or less exactly like their SQL equivalents. However, these statements do exactly the same thing, just with different syntax. What they do is to write values into columns at a key. So these two Cequel statements have identical behavior:

cassandra[:posts].insert(:id => 1, :title => 'Post')
cassandra[:posts].where(:id => 1).update(:title => 'Post')

Both of these statements instruct Cassandra to set the value of the title column in row 1 to "Post".

Cequel::Model uses the INSERT statement to persist objects that have been newly initialized in memory, and the UPDATE statement to save changes to objects that were loaded out of Cassandra. There is no particular reason for this; it just feels right. But beware: you may think you're inserting a new row when you're actually overwriting data that already exists in that row:

# I'm just creating a post here.
post1 = => 1, :title => 'My Post', :blog_id => 1)!

# And let's make another one
post2 = => 1, :title => 'Another Post')!

Living in a relational world, we'd expect the second statement to throw an error because row 1 already exists. But not Cassandra: the above code will just overwrite the title in that row. Note that the blog_id will not be touched; upserts only work on the columns that are given.

Dirty Updates

Cequel::Model includes ActiveModel's dirty tracking. When you save a persisted model, only columns that have changed in memory will be included in the UPDATE statement.

Note that updating a model may generate two CQL statements. This is because Cassandra does not have a concept of null values; a column either has data or it doesn't. So, if you change an attribute of your model from a non-nil value to nil, Cequel::Model will issue a DELETE statement just for the column(s) in question.

If you don't change anything, calling '#save' on a persisted model is a no-op.

Pondering Existence

In a relational database, there is a well-defined concept of existence; there is either a row for a given primary key or there isn't. It's possible to have a row consisting of only a primary key, and that row still "exists" in a meaningful way.

Cassandra works more like a key-value store: each key either has data, or it doesn't, but beyond that there is no explicit concept of a key or row existing. Semantically, we can think of a Cassandra row existing if it has data in any column. But that's a concept that only exists in our minds (and in Cequel), not in the database itself. Consider the following:

cassandra[:posts].where(:id => 1).first
#=> {'id' => 1}

The above behavior will hold even if no data has ever been written to key 1. It will also happen if key 1 existed at one time and then was deleted.

This behavior is complicated by "range ghosts". Range ghosts happen when you delete all the data from a row. You'll only see them when performing unlimited or key-range queries, and they go away after a while. There's a good reason for this, but it's confusing. For instance, let's say in the entire history of our database, all we've done is create post 1, and then delete it. Let's see what happens when we select all posts:

#=> [{'id' => 1}]

That's a range ghost: it's a result row consisting of only the key.

Cequel::Model makes explicit our implicit semantic idea that rows only exist if they have data in a column (not counting the key, which isn't really a column). So any time Cequel::Model sees a row that's either empty or only has a key, it drops it. You'll never get back a model instance containing data in no non-key columns.

If you perform a #find and get back no non-key data, the library will raise Cequel::Model::RecordNotFound.

This behavior can especially trip you up when you are selecting specific columns. For instance, let's say post 1 only has data in the title field:

# Gives me back a nice post object
# Raises Cequel::Model::RecordNotFound, because there was no data in the row
# Fails fast before any interaction with Cassandra: this is a meaningless query

Key and Secondary Index Selection

CQL gives you a few ways to filter the rows you want returned in a query:

  • A single key
  • A list of keys
  • A range of keys
  • A secondary index
  • A secondary index combined with one or more filters

That's it. You can't filter by:

  • A non-indexed column
  • A key/list of keys combined with a secondary index
  • A key/list of keys combined with a filter

So let's say our posts column family has a secondary index on blog_id and nothing else. These will work:

Post.find([uuid1, uuid2])
Post.where('id > ?', uuid)
Post.where(:blog_id => blog_id).where('created_at > ?',

These won't work:

Post.where('created_at > ?',
Post.where(:id => uuid, :blog_id => blog_id)
Post.where(:id => uuid).where('created_at > ?,


The functionality of the Cequel::Model class maps the "skinny row" style of column family structure: each row has a small set of predefined columns, with heterogeneous value types. However, the "wide row" structure will also play an important role in most Cassandra schemas (if this is news to you, I recommend reading this article). Cequel provides the Cequel::Model::Dictionary class, which abstracts wide rows as a dictionary object, behaving much like a Hash.

Applications should define subclasses of the Dictionary class to interact with data in a certain column family. For instance, let's say I've got a blog_posts column family:

class BlogPosts < Cequel::Model::Dictionary

  key :blog_id, :uuid
  maps :uuid => :text


  def serialize_value(column, value)

  def deserialize_value(column, value)


In this case, your column family has a key with alias blog_id of type uuid, comparator of type uuid, and default validation of type text. The serialize_value and deserialize_value methods are optional, but aid with the common pattern of storing blobs of JSON, msgpack, etc. in wide-row values.

Reading data

To grab a handle to a dictionary, use the bracket operator:

posts = BlogPosts[blog_id]

This does not perform any queries against Cassandra; it just gives you an object pointing at a particular row. By default, reads are lazy:

post_json = posts[post_id]

This will select a single column from the blog_posts column family and return its deserialized value. The value is not retained in the dictionary itself.

If you want to work with the entire contents of the wide row in memory, use the #load method:

posts = BlogPosts[blog_id]
posts.load # loads all values into memory
posts[post_id] # doesn't do an additional query

Dictionaries expose the major read methods of the Hash interface:

posts.each_pair { |column, value| do_something(column, value) }
posts.values { |column, value| transform(column, value) }
posts.slice(uuid1, uuid2, uuid3) # returns a Hash

All of the above methods will read from Cassandra if the dictionary is unloaded, and read from memory if the dictionary is loaded. Note that for methods that read all columns out of the database, columns will be loaded in batches of 1000 by default.

Writing Data

Modifying data is, unsurprisingly, done using the []= operator. When you call #save, any keys that you have modified with the []= operator will be persisted to Cassandra. The dictionary does not use true dirty tracking, in the sense that it will write an attribute even if you set it to the same value it had previously.

Write behavior is the same regardless of loaded status.

Loading data in bulk

Dictionaries implement the ::load method, which allows you to read multiple rows at once. Unlike the #each and #load methods, ::load will not attempt to paginate over very wide rows (10K+ columns); if your rows are very wide, you will probably want to load them one at a time anyway.

post_rows = BlogPosts.load(blog1_id, blog2_id) # load rows at key blog1_id, blog2_id
post_rows = BlogPosts.load(blog1_id, blog2_id) # wider rows


Counters are a special type of Cassandra column that implements a consistent distributed counter. The only write operations that are possible on counter columns are increment and decrement (they can be deleted, technically, but the behavior is undesirable). You can create a counter dictionary using the Cequel::Model::Counter class:

class CommentCounts < Cequel::Model::Counter
  key :blog_id, :int
  columns :uuid # values are always of 'counter' type

For read operations, counters work exactly like normal dictionaries. For write operations, counters have #increment and #decrement methods available (but not #[]=):

comment_counts = CommentCounts[blog_id]
comment_counts.increment(post_id) # increment by 1
comment_counts.increment(post_id, 4) # increment by 4
comment_counts.increment([post1_id, post2_id], 3) # increment multiple at once
comment_counts.increment(post1_id => 2, post2_id => 4) # by different values
comment_counts.decrement(post_id) # accepts all the same forms as #increment

Road Map

As mentioned previously in this document, there are considerable differences between modeling data in Cassandra and modeling data in a relational database, despite their superficial similarities. In Cassandra, wide rows are an important part of schema design; "existence" is a fuzzy concept; denormalization is often a good idea; secondary indexes are of limited use. Broadly, the goal for future versions of Cequel is to provide a more robust abstraction and tool kit for modeling data in Cassandra the right way. Specifically, here are some things to look forward to in future Cequel versions:

  • Support for auto-migrations by introspecting the schema and making modifications to fit the model-defined schema.
  • One-one relationships using multiple classes per column family.
  • Additional wide-row data structures: lists and sets.
  • Tighter integration between Cequel::Model and Cequel::Model::Dictionary; references_many associations.
  • Bidirectional associations.
  • Using defined column types to ensure objects passed to CassandraCQL layer are of the correct type/encoding.

Support & Bugs

If you find a bug, feel free to open an issue on GitHub. Pull requests are most welcome.

For questions or feedback, hit up our mailing list at or find outoftime in the #cassandra IRC channel on Freenode.


Cequel is distributed under the MIT license. See the attached LICENSE for all the sordid details.


CQL query builder and object-row mapper for Cassandra







No packages published