Dynamoid is an ORM for Amazon's DynamoDB for Ruby applications. It provides similar functionality to ActiveRecord and improves on Amazon's existing HashModel by providing better searching tools and native association support.
DynamoDB is not like other document-based databases you might know, and is very different indeed from relational databases. It sacrifices anything beyond the simplest relational queries and transactional support to provide a fast, cost-efficient, and highly durable storage solution. If your database requires complicated relational queries and transaction support, then this modest Gem cannot provide them for you, and neither can DynamoDB. In those cases you would do better to look elsewhere for your database needs.
But if you want a fast, scalable, simple, easy-to-use database (and a Gem that supports it) then look no further!
Installing Dynamoid is pretty simple. First include the Gem in your Gemfile:
gem 'dynamoid'
Dynamoid depends on the aws-sdk, and this is tested on the current version of aws-sdk (1.6.9), rails 3.2.8. Hence the configuration as needed for aws to work will be dealt with by aws setup.
Here are the steps to setup aws-sdk.
gem 'aws-sdk'
(or) include the aws-sdk in your Gemfile.
- Just like the config/database.yml this file requires an entry for each environment, create config/aws.yml as follows:
Fill in your AWS Access Key ID and Secret Access Key
development:
access_key_id: REPLACE_WITH_ACCESS_KEY_ID
secret_access_key: REPLACE_WITH_SECRET_ACCESS_KEY
dynamo_db_endpoint: dynamodb.ap-southeast-1.amazonaws.com
test:
<<: *development
production:
<<: *development
(or)
- Create config/initializers/aws.rb as follows:
#Additionally include any of the dynamodb paramters as needed.
#(eg: if you would like to change the dynamodb endpoint, then add the parameter in
# in the file aws.yml or aws.rb
AWS.config({
:access_key_id => 'REPLACE_WITH_ACCESS_KEY_ID',
:secret_access_key => 'REPLACE_WITH_SECRET_ACCESS_KEY',
:dynamo_db_endpoint => 'dynamodb.ap-southeast-1.amazonaws.com'
})
For a full list of the DDB regions, you can go here.
Refer to the documentation of the AWS module at the below link for all of the configuration options supported by DynamoDB.
Then you need to initialize Dynamoid config to get it going. Put code similar to this somewhere (a Rails initializer would be a great place for this if you're using Rails):
Dynamoid.configure do |config|
config.adapter = 'aws_sdk' # This adapter establishes a connection to the DynamoDB servers using Amazon's own AWS gem.
config.namespace = "dynamoid_app_development" # To namespace tables created by Dynamoid from other tables you might have.
config.warn_on_scan = true # Output a warning to the logger when you perform a scan rather than a query on a table.
config.partitioning = true # Spread writes randomly across the database. See "partitioning" below for more.
config.partition_size = 200 # Determine the key space size that writes are randomly spread across.
config.read_capacity = 100 # Read capacity for your tables
config.write_capacity = 20 # Write capacity for your tables
end
Once you have the configuration set up, you need to move on to making models.
You must include Dynamoid::Document
in every Dynamoid model.
class User
include Dynamoid::Document
end
Dynamoid has some sensible defaults for you when you create a new table, including the table name and the primary key column. But you can change those if you like on table creation.
class User
include Dynamoid::Document
table :name => :awesome_users, :key => :user_id, :read_capacity => 400, :write_capacity => 400
end
These fields will not change an existing table: so specifying a new read_capacity and write_capacity here only works correctly for entirely new tables. Similarly, while Dynamoid will look for a table named awesome_users
in your namespace, it won't change any existing tables to use that name; and if it does find a table with the correct name, it won't change its hash key, which it expects will be user_id. If this table doesn't exist yet, however, Dynamoid will create it with these options.
You'll have to define all the fields on the model and the data type of each field. Every field on the object must be included here; if you miss any they'll be completely bypassed during DynamoDB's initialization and will not appear on the model objects.
By default, fields are assumed to be of type :string
. But you can also use :integer
, :float
, :set
, :array
, :datetime
, :boolean
, and :serialized
. You get magic columns of id (string), created_at (datetime), and updated_at (datetime) for free.
class User
include Dynamoid::Document
field :name
field :email
field :rank, :integer
field :number, :float
field :joined_at, :datetime
field :hash, :serialized
end
You can also define indexes on fields, combinations of fields, and one range field. Yes, only one range field: in DynamoDB tables can have at most one range index, so make good use of it! To make an index, just specify the fields you want it on, either single or in an array. If the entire index is a range, pass :range => true
. Otherwise, pass the attribute that will become the range key. The only range attributes you can use right now are integers, floats, and datetimes. If you pass a string as a range key likely DynamoDB will complain a lot.
class User
include Dynamoid::Document
...
index :name
index :email
index [:name, :email]
index :created_at, :range => true
index :name, :range_key => :joined_at
end
Just like in ActiveRecord (or your other favorite ORM), Dynamoid uses associations to create links between models.
The only supported associations (so far) are has_many
, has_one
, has_and_belongs_to_many
, and belongs_to
. Associations are very simple to create: just specify the type, the name, and then any options you'd like to pass to the association. If there's an inverse association either inferred or specified directly, Dynamoid will update both objects to point at each other.
class User
include Dynamoid::Document
...
has_many :addresses
has_many :students, :class => User
belongs_to :teacher, :class_name => :user
belongs_to :group
has_one :role
has_and_belongs_to_many :friends, :inverse_of => :friending_users
end
class Address
include Dynamoid::Document
...
belongs_to :address # Automatically links up with the user model
end
Contrary to what you'd expect, association information is always contained on the object specifying the association, even if it seems like the association has a foreign key. This is a side effect of DynamoDB's structure: it's very difficult to find foreign keys without an index. Usually you won't find this to be a problem, but it does mean that association methods that build new models will not work correctly -- for example, user.addresses.new
returns an address that is not associated to the user. We'll be correcting this soon.
Dynamoid bakes in ActiveModel validations, just like ActiveRecord does.
class User
include Dynamoid::Document
...
validates_presence_of :name
validates_format_of :email, :with => /@/
end
To see more usage and examples of ActiveModel validations, check out the ActiveModel validation documentation.
Dynamoid also employs ActiveModel callbacks. Right now, callbacks are defined on save
, update
, destroy
, which allows you to do before_
or after_
any of those.
class User
include Dynamoid::Document
...
before_save :set_default_password
after_create :notify_friends
after_destroy :delete_addresses
end
Dynamoid's syntax is generally very similar to ActiveRecord's. Making new objects is simple:
u = User.new(:name => 'Josh')
u.email = 'josh@joshsymonds.com'
u.save
Save forces persistence to the datastore: a unique ID is also assigned, but it is a string and not an auto-incrementing number.
u.id # => "3a9f7216-4726-4aea-9fbc-8554ae9292cb"
Along with persisting the model's attributes, indexes are automatically updated on save. To use associations, you use association methods very similar to ActiveRecord's:
address = u.addresses.create
address.city = 'Chicago'
address.save
Querying can be done in one of three ways:
Address.find(address.id) # Find directly by ID.
Address.where(:city => 'Chicago').all # Find by any number of matching criteria... though presently only "where" is supported.
Address.find_by_city('Chicago') # The same as above, but using ActiveRecord's older syntax.
And you can also query on associations:
u.addresses.where(:city => 'Chicago').all
But keep in mind Dynamoid -- and document-based storage systems in general -- are not drop-in replacements for existing relational databases. The above query does not efficiently perform a conditional join, but instead finds all the user's addresses and naively filters them in Ruby. For large associations this is a performance hit compared to relational database engines.
You can also limit returned results, or select a record from which to start, to support pagination:
Address.limit(5).start(address) # Only 5 addresses.
For large queries that return many rows, Dynamoid can use AWS' support for requesting documents in batches:
#Do some maintenance on the entire table without flooding DynamoDB
Address.all(batch_size: 100).each { |address| address.do_some_work; sleep(0.01) }
Address.limit(10_000).batch(100). each { … } #batch specified as part of a chain
Querying supports consistent reading. By default, DynamoDB reads are eventually consistent: if you do a write and then a read immediately afterwards, the results of the previous write may not be reflected. If you need to do a consistent read (that is, you need to read the results of a write immediately) you can do so, but keep in mind that consistent reads are twice as expensive as regular reads for DynamoDB.
Address.find(address.id, :consistent_read => true) # Find an address, ensure the read is consistent.
Address.where(:city => 'Chicago').consistent.all # Find all addresses where the city is Chicago, with a consistent read.
If you have a range index, Dynamoid provides a number of additional other convenience methods to make your life a little easier:
User.where("created_at.gt" => DateTime.now - 1.day).all
User.where("created_at.lt" => DateTime.now - 1.day).all
It also supports .gte and .lte. Turning those into symbols and allowing a Rails SQL-style string syntax is in the works. You can only have one range argument per query, because of DynamoDB's inherent limitations, so use it sensibly!
DynamoDB achieves much of its speed by relying on a random pattern of writes and reads: internally, hash keys are distributed across servers, and reading from two consecutive servers is much faster than reading from the same server twice. Of course, many of our applications request one key (like a commonly used role, a superuser, or a very popular product) much more frequently than other keys. In DynamoDB, this will result in lowered throughput and slower response times, and is a design pattern we should try to avoid.
Dynamoid attempts to obviate this problem transparently by employing a partitioning strategy to divide up keys randomly across DynamoDB's servers. Each ID is assigned an additional number (by default 0 to 199, but you can increase the partition size in Dynamoid's configuration) upon save; when read, all 200 hashes are retrieved simultaneously and the most recently updated one is returned to the application. This results in a significant net performance increase, and is usually invisible to the application itself. It does, however, bring up the important issue of provisioning your DynamoDB tables correctly.
When your read or write throughput exceed your table's allowed provisioning, DynamoDB will wait on connections until throughput is available again. This will appear as very, very slow requests and can be somewhat frustrating. Partitioning significantly increases the amount of throughput tables will experience; though DynamoDB will ignore keys that don't exist, if you have 20 partitioned keys representing one object, all will be retrieved every time the object is requested. Ensure that your tables are set up for this kind of throughput, or turn provisioning off, to make sure that DynamoDB doesn't throttle your requests.
Dynamoid supports basic, ActiveRecord-like optimistic locking on save operations. Simply add a lock_version
column to your table like so:
class MyTable
...
field :lock_version, :integer
...
end
In this example, all saves to MyTable
will raise an AWS::DynamoDB::Errors::ConditionalCheckFailedException
if a concurrent process loaded, edited, and saved the same row. Your code should trap this exception, reload the row (so that it will pick up the newest values), and try the save again.
Calls to update
and update!
also increment the lock_version
, however they do not check the existing value. This guarantees that a update operation will raise an exception in a concurrent save operation, however a save operation will never cause an update to fail. Thus, update
is useful & safe only for doing atomic operations (e.g. increment a value, add/remove from a set, etc), but should not be used in a read-modify-write pattern.
Dynamoid borrows code, structure, and even its name very liberally from the truly amazing Mongoid. Without Mongoid to crib from none of this would have been possible, and I hope they don't mind me reusing their very awesome ideas to make DynamoDB just as accessible to the Ruby world as MongoDB.
Also, without contributors the project wouldn't be nearly as awesome. So many thanks to:
Running the tests is fairly simple. In one window, run fake_dynamo --port 4567
, and in the other, use rake
.
Copyright (c) 2012 Josh Symonds.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.