A ruby library for saving an object's state as a series of events.
"Capture all changes to an application state as a sequence of events." Event Sourcing
When state changes made to an object is important a common technique is to store the changes in a separate history log where the log is generated in parallel with the object internal state. With event sourcing the history log is now integrated within the object and generated based on the actions made to the object. The entries in log is the facts the object is built upon.
If you have been following Uncle Bob you know what he thinks of the "Rails way" and how we get bound to the Rails framework. We have created Sandthorn to decouple our models from Active Record and restore them to what they should be, i.e., Plain Old Ruby Objects (PORO) with a twist of Sandthorn magic.
Check out examples of Sandthorn:
- Examples including a product shop and TicTacToe game.
- Live demo comparing Active Record and Sandthorn.
Add this line to your application's Gemfile:
gem 'sandthorn'
And then execute:
$ bundle
Or install it yourself as:
$ gem install sandthorn
Sandthorn can be setup with one or more drivers. A driver is bound to a specific data store where events are saved and loaded from. The current implemented drivers are sandthorn_driver_sequel for SQL via Sequel and sandthorn_driver_event_store that uses Get Event Store.
This means Sandthorn can be used with any data store given that a driver exists.
Here's an example of setting up Sandthorn with the Sequel driver and a sqlite3 database.
url = "sqlite://sql.sqlite3"
driver = SandthornDriverSequel.driver_from_url(url: url)
Sandthorn.configure do |conf|
conf.event_stores = { default: driver }
end
Its possible to save events from different classes into different stores. Below the events from class FooAggregate are stored into the sql_foo.sqlite3 database and events from class BarAggregate are stored in sql_bar.sqlite3.
driver_foo = SandthornDriverSequel.driver_from_url(url: "sqlite://sql_foo.sqlite3")
driver_bar = SandthornDriverSequel.driver_from_url(url: "sqlite://sql_bar.sqlite3")
class FooAggregate
Include Sandthorn::AggregateRoot
end
class BarAggregate
Include Sandthorn::AggregateRoot
end
Sandthorn.configure do |conf|
conf.event_stores = { foo: driver_foo, bar: driver_bar }
conf.map_types = { foo: [FooAggregate], bar: [BarAggregate] }
end
Any object that should have event sourcing capability must include the methods provided by Sandthorn::AggregateRoot
. These make it possible to commit
events and save
changes to an aggregate. Use the include
directive as follows:
require 'sandthorn'
class Board
include Sandthorn::AggregateRoot
end
All objects that include Sandthorn::AggregateRoot
is provided with an aggregate_id
which is a UUID.
An abstraction over commit
that creates events methods that can be used from within a command method.
In this exampel the events
method will generate a method called marked
, this method take an block that will be executed before the event is commited and is used to groups the state changes to the event. The block is optional and the state changes could have been made outside the marked
method.
class Board
include Sandthorn::AggregateRoot
events :marked
def mark player, pos_x, pos_y
# change some state
marked() do
@pos_x = pos_x
@pos_y = pos_y
end
end
end
With constructor_events
its possible to be more specific on how an aggregate came to be. The first event will now have the name board_created
instead of the default new
.
class Board
include Sandthorn::AggregateRoot
# creates a private class method `board_created`
constructor_events :board_created
def self.create name
board_created(name) do
@name = name
end
end
end
Calling stateless_events
creates public class methods. The first argument is an aggregate_id
and the second argument is optional but has to be a hash and is stored in the event_data of the event.
When creating a stateless event, the corresponding aggregate is never loaded and the event is saved without calling the save method.
class Board
include Sandthorn::AggregateRoot
stateless_events :player_went_to_toilet
end
Board.player_went_to_toilet "board_aggregate_id", {player_id: "1", time: "10:12"}
Its possible to add a default_attributes method on an aggregate and set default values to new and already created aggregates.
The default_attributes
method will be run before initialize on Class.new and before the events when an aggregate is rebuilt. This will make is possible to add default attributes to an aggregate during its hole life cycle.
def default_attributes
@new_array = []
end
To generate an event the commit method has to be called within the aggregate. commit
extracts the object's delta and locally caches the state changes that has been applied to the aggregate.
def mark player, pos_x, pos_y
# change some state
marked
end
def marked
commit
end
commit
determines the state changes by monitoring the object's readable fields.
The concept events
have been introduced to abstract away the usage of commit
. Commit still works as before but we think that the events
abstraction makes the aggregate more readable.
The save method store generated events, this means all commited events will be persisted via a Sandthorn driver.
board = Board.new
board.mark :o, 0, 1
board.save
Retrieve an array with all instances of a specific aggregate type.
Board.all
Since it return's an Array
you can, for example, filter on an aggregate's fields
Board.all.select { |board| board.active == true }
Loads a specific aggregate using it's uuid.
uuid = '550e8400-e29b-41d4-a716-446655440000'
board = Board.find(uuid)
If no aggregate with the specifid uuid is found, a Sandthorn::Errors::AggregateNotFound
exception is raised.
Using aggregate_trace
one can store meta data on events. The data is not aggregate specific and it can for example store who executed a specific command on the aggregate.
board.aggregate_trace {player: "Fred"} do |aggregate|
aggregate.mark :o, 0, 1
aggregate.save
end
aggregate_trace
can also be specified on a class.
Board.aggregate_trace {ip: :127.0.0.1} do
board = Board.new
board.mark :o , 0, 1
board.save
end
In this case, the resulting events from the commands new
and mark
will have the trace {ip: :127.0.0.1}
attached to them.
Check if there are unsaved events attached to the aggregate.
board = Board.new
board.mark :o, 0, 1
board.unsaved_events?
=> true
If there is a lot of events saved to an aggregate it can take some time to reload the current state of the aggregate via the .find
method. This is because all events belonging to the aggregate has to be fetched and iterated one by one to build its current state. The snapshot functionality makes it possible to store the current aggregate state and re-use it when loading the aggregate. The snapshot is used as a cache where only the events that has occurred after the snapshot has to be fetched and used to build the current state of the aggregate.
There is one global snapshot store where all snapshots are stored independent on aggregate_type. To enable snapshot on a aggregate_type the Class has to be added to the snapshot_types
Array when configuring Sandthorn. The aggregate will now be stored to the snapshot_store on every .save
and when using .find
it will look for a snapshot of the requested aggregate.
class Board
include Sandthorn::AggregateRoot
end
Sandthorn.configure do |c|
c.snapshot_types = [Board]
end
Its possible to take manual snapshots without enabling snapshots on the aggregate_type.
board = Board.new
board.save
# Save snapshot of the board aggregate
Sandthorn.save_snapshot board
# Get snapshot
snapshot = Sandthorn.find_snapshot board.aggregate_id
There is one external snapshot store available sandthorn_snapshot_memcached and it can be configured via Sandthorn.configure
require 'sandthorn_snapshot_memcached'
snapshot_store = SandthornSnapshotMemcached.from_url "memcached_url"
Sandthorn.configure do |conf|
conf.snapshot_store = snapshot_store
end
If no external snapshot store is configured snapshots will be stored in the application memory (be careful not draining your application memory space).
A bounded context is a system divider that split large systems into smaller parts. Bounded Context by Martin Fowler
A module can include Sandthorn::BoundedContext
and all aggregates within the module can be retreived via the ::aggregate_types method on the module. A use case is to use it when Sandthorn is configured and setup all aggregates in a bounded context to a driver.
require 'sandthorn/bounded_context'
module TicTacToe
include Sandthorn::BoundedContext
class Board
include Sandthorn::AggregateRoot
end
end
Sandthorn.configure do |conf|
conf.event_stores = { foo: driver_foo}
conf.map_types = { foo: TicTacToe.aggregate_types }
end
TicTacToe.aggregate_types -> [TicTacToy::Board]
Run tests: rake
Run benchmark tests: rake benchmark
Load a console: rake console
We're happy to accept pull requests that makes the code cleaner or more idiomatic, the documentation more understandable, or improves the testsuite. Even considering opening an issue for what's troubling you or writing a blog post about how you used Sandthorn is worth a lot too!
In general, the contribution process for code works like this.
- Fork this repo
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create new Pull Request