adomokos / light-service

Series of Actions with an emphasis on simplicity.
MIT License
837 stars 67 forks source link
aop ruby workflow

LightService

Gem Version CI Tests Codecov License Download Count Code Climate

LightService is a powerful and flexible service skeleton framework with an emphasis on simplicity

🔥 It now comes with no external gem dependency. 🔥

Table of Contents

Why LightService?

What do you think of this code?

class TaxController < ApplicationController
  def update
    @order = Order.find(params[:id])
    tax_ranges = TaxRange.for_region(order.region)

    if tax_ranges.nil?
      render :action => :edit, :error => "The tax ranges were not found"
      return # Avoiding the double render error
    end

    tax_percentage = tax_ranges.for_total(@order.total)

    if tax_percentage.nil?
      render :action => :edit, :error => "The tax percentage  was not found"
      return # Avoiding the double render error
    end

    @order.tax = (@order.total * (tax_percentage/100)).round(2)

    if @order.total_with_tax > 200
      @order.provide_free_shipping!
    end

    redirect_to checkout_shipping_path(@order), :notice => "Tax was calculated successfully"
  end
end

This controller violates SRP all over. Also, imagine what it would take to test this beast. You could move the tax_percentage finders and calculations into the tax model, but then you'll make your model logic heavy.

This controller does 3 things in order:

The order of these tasks matters: you can't calculate the order tax without the percentage. Wouldn't it be nice to see this instead?

(
  LooksUpTaxPercentage,
  CalculatesOrderTax,
  ProvidesFreeShipping
)

This block of code should tell you the "story" of what's going on in this workflow. With the help of LightService you can write code this way. First you need an organizer object that sets up the actions in order and executes them one-by-one. Then you need to create the actions with one method (that will do only one thing).

This is how the organizer and actions interact with each other:

LightService

class CalculatesTax
  extend LightService::Organizer

  def self.call(order)
    with(:order => order).reduce(
        LooksUpTaxPercentageAction,
        CalculatesOrderTaxAction,
        ProvidesFreeShippingAction
      )
  end
end

class LooksUpTaxPercentageAction
  extend LightService::Action
  expects :order
  promises :tax_percentage

  executed do |context|
    tax_ranges = TaxRange.for_region(context.order.region)
    context.tax_percentage = 0

    next context if object_is_nil?(tax_ranges, context, 'The tax ranges were not found')

    context.tax_percentage = tax_ranges.for_total(context.order.total)

    next context if object_is_nil?(context.tax_percentage, context, 'The tax percentage was not found')
  end

  def self.object_is_nil?(object, context, message)
    if object.nil?
      context.fail!(message)
      return true
    end

    false
  end
end

class CalculatesOrderTaxAction
  extend ::LightService::Action
  expects :order, :tax_percentage

  # I am using ctx as an abbreviation for context
  executed do |ctx|
    order = ctx.order
    order.tax = (order.total * (ctx.tax_percentage/100)).round(2)
  end

end

class ProvidesFreeShippingAction
  extend LightService::Action
  expects :order

  executed do |ctx|
    if ctx.order.total_with_tax > 200
      ctx.order.provide_free_shipping!
    end
  end
end

And with all that, your controller should be super simple:

class TaxController < ApplicationContoller
  def update
    @order = Order.find(params[:id])

    service_result = CalculatesTax.for_order(@order)

    if service_result.failure?
      render :action => :edit, :error => service_result.message
    else
      redirect_to checkout_shipping_path(@order), :notice => "Tax was calculated successfully"
    end

  end
end

I gave a talk at RailsConf 2013 on simple and elegant Rails code where I told the story of how LightService was extracted from the projects I had worked on.

Getting started

Requirements

This gem requires ruby 2.x. Use of generators requires Rails 5+ (tested on Rails 5.x & 6.x only. Will probably work on Rails versions as old as 3.2)

Installation

In your Gemfile:

gem 'light-service'

And then

bundle install

Or install it yourself as:

gem install light-service

Your first action

LightService's building blocks are actions that are normally composed within an organizer, but can be run independently. Let's make a simple greeter action. Each action can take an optional list of expected inputs and promised outputs. If these are specified and missing at action start and stop respectively, an exception will be thrown.

class GreetsPerson
  extend ::LightService::Action

  expects :name
  promises :greeting

  executed do |context|
    context.greeting = "Hey there, #{name}. You enjoying LightService so far?"
  end
end

When an action is run, you have access to its returned context, and the status of the action. You can invoke an action by calling .execute on its class with key: value arguments, and inspect its status and context like so:

outcome = GreetsPerson.execute(name: "Han")

if outcome.success?
  puts outcome.greeting # which was a promised context value
elsif outcome.failure?
  puts "Rats... I can't say hello to you"
end

You will notice that actions are set up to promote simplicity, i.e. they either succeed or fail, and they have very clear inputs and outputs. Ideally, they should do exactly one thing. This makes them as easy to test as unit tests.

Your first organizer

LightService provides a facility to compose actions using organizers. This is great when you have a business process to execute that has multiple steps. By composing actions that do exactly one thing, you can sequence simple actions together to perform complex multi-step business processes in a clear manner that is very easy to reason about.

There are advanced ways to sequence actions that can be found later in the README, but we'll keep this simple for now. First, let's add a second action that we can sequence to run after the GreetsPerson action from above:

class RandomlyAwardsPrize
  extend ::LightService::Action

  expects :name, :greeting
  promises :did_i_win

  executed do |context|
    prize_num  = "#{context.name}__#{context.greeting}".length
    prizes     = ["jelly beans", "ice cream", "pie"]
    did_i_win  = rand((1..prize_num)) % 7 == 0
    did_i_lose = rand((1..prize_num)) % 13 == 0

    if did_i_lose
      # When failing, send a message as an argument, readable from the return context
      context.fail!("you are exceptionally unlucky")
    else
      # You can specify 'optional' context items by treating context like a hash.
      # Useful for when you may or may not be returning extra data. Ideally, selecting
      # a prize should be a separate action that is only run if you win.
      context[:prize]   = "lifetime supply of #{prizes.sample}" if did_i_win
      context.did_i_win = did_i_win
    end
  end
end

And here's the organizer that ties the two together. You implement a call class method that takes some arguments and from there sends them to with in key: value format which forms the initial state of the context. From there, chain reduce to with and send it a list of action class names in sequence. The organizer will call each action, one after the other, and build up the context as it goes along.

class WelcomeAPotentiallyLuckyPerson
  extend LightService::Organizer

  def self.call(name)
    with(:name => name).reduce(GreetsPerson, RandomlyAwardsPrize)
  end
end

When an organizer is run, you have access to the context as it passed through all actions, and the overall status of the organized execution. You can invoke an organizer by calling .call on the class with the expected arguments, and inspect its status and context just like you would an action:

outcome = WelcomeAPotentiallyLuckyPerson.call("Han")

if outcome.success?
  puts outcome.greeting # which was a promised context value

  if outcome.did_i_win
    puts "And you've won a prize! Lucky you. Please see the front desk for your #{outcome.prize}."
  end
else # outcome.failure? is true, and we can pull the failure message out of the context for feedback to the user.
  puts "Rats... I can't say hello to you, because #{outcome.message}."
end

Because organizers generally run through complex business logic, and every action has the potential to cause a failure, testing an organizer is functionally equivalent to an integration test.

For further examples, please visit the project's Wiki and review the "Why LightService" section above.

Stopping the Series of Actions

When nothing unexpected happens during the organizer's call, the returned context will be successful. Here is how you can check for this:

class SomeController < ApplicationController
  def index
    result_context = SomeOrganizer.call(current_user.id)

    if result_context.success?
      redirect_to foo_path, :notice => "Everything went OK! Thanks!"
    else
      flash[:error] = result_context.message
      render :action => "new"
    end
  end
end

However, sometimes not everything will play out as you expect it. An external API call might not be available or some complex business logic will need to stop the processing of the Series of Actions. You have two options to stop the call chain:

  1. Failing the context
  2. Skipping the rest of the actions

Failing the Context

When something goes wrong in an action and you want to halt the chain, you need to call fail! on the context object. This will push the context in a failure state (context.failure? # will evalute to true). The context's fail! method can take an optional message argument, this message might help describing what went wrong. In case you need to return immediately from the point of failure, you have to do that by calling next context.

In case you want to fail the context and stop the execution of the executed block, use the fail_and_return!('something went wrong') method. This will immediately leave the block, you don't need to call next context to return from the block.

Here is an example:

class SubmitsOrderAction
  extend LightService::Action
  expects :order, :mailer

  executed do |context|
    unless context.order.submit_order_successful?
      context.fail_and_return!("Failed to submit the order")
    end

    # This won't be executed
    context.mailer.send_order_notification!
  end
end

fail-actions

In the example above the organizer called 4 actions. The first 2 actions got executed successfully. The 3rd had a failure, that pushed the context into a failure state and the 4th action was skipped.

Skipping the rest of the actions

You can skip the rest of the actions by calling context.skip_remaining!. This behaves very similarly to the above-mentioned fail! mechanism, except this will not push the context into a failure state. A good use case for this is executing the first couple of action and based on a check you might not need to execute the rest. Here is an example of how you do it:

class ChecksOrderStatusAction
  extend LightService::Action
  expects :order

  executed do |context|
    if context.order.send_notification?
      context.skip_remaining!("Everything is good, no need to execute the rest of the actions")
    end
  end
end

skip-actions

In the example above the organizer called 4 actions. The first 2 actions got executed successfully. The 3rd decided to skip the rest, the 4th action was not invoked. The context was successful.

Benchmarking Actions with Around Advice

Benchmarking your action is needed when you profile the series of actions. You could add benchmarking logic to each and every action, however, that would blur the business logic you have in your actions.

Take advantage of the organizer's around_each method, which wraps the action calls as its reducing them in order.

Check out this example:

class LogDuration
  def self.call(context)
    start_time = Time.now
    result = yield
    duration = Time.now - start_time
    LightService::Configuration.logger.info(
      :action   => context.current_action,
      :duration => duration
    )

    result
  end
end

class CalculatesTax
  extend LightService::Organizer

  def self.call(order)
    with(:order => order).around_each(LogDuration).reduce(
        LooksUpTaxPercentageAction,
        CalculatesOrderTaxAction,
        ProvidesFreeShippingAction
      )
  end
end

Any object passed into around_each must respond to #call with two arguments: the action name and the context it will execute with. It is also passed a block, where LightService's action execution will be done in, so the result must be returned. While this is a little work, it also gives you before and after state access to the data for any auditing and/or checks you may need to accomplish.

Before and After Action Hooks

In case you need to inject code right before and after the actions are executed, you can use the before_actions and after_actions hooks. It accepts one or multiple lambdas that the Action implementation will invoke. This addition to LightService is a great way to decouple instrumentation from business logic.

Consider this code:

class SomeOrganizer
  extend LightService::Organizer

  def self.call(ctx)
    with(ctx).reduce(actions)
  end

  def self.actions
    [
      OneAction,
      TwoAction,
      ThreeAction
    ]
  end
end

class TwoAction
  extend LightService::Action
  expects :user, :logger

  executed do |ctx|
    # Logging information
    if ctx.user.role == 'admin'
       ctx.logger.info('admin is doing something')
    end

    ctx.user.do_something
  end
end

The logging logic makes TwoAction more complex, there is more code for logging than for business logic.

You have two options to decouple instrumentation from real logic with before_actions and after_actions hooks:

  1. Declare your hooks in the Organizer
  2. Attach hooks to the Organizer from the outside

This is how you can declaratively add before and after hooks to the Organizer:

class SomeOrganizer
  extend LightService::Organizer
  before_actions (lambda do |ctx|
                           if ctx.current_action == TwoAction
                             return unless ctx.user.role == 'admin'
                             ctx.logger.info('admin is doing something')
                           end
                         end)
  after_actions (lambda do |ctx|
                          if ctx.current_action == TwoAction
                            return unless ctx.user.role == 'admin'
                            ctx.logger.info('admin is DONE doing something')
                          end
                        end)

  def self.call(ctx)
    with(ctx).reduce(actions)
  end

  def self.actions
    [
      OneAction,
      TwoAction,
      ThreeAction
    ]
  end
end

class TwoAction
  extend LightService::Action
  expects :user

  executed do |ctx|
    ctx.user.do_something
  end
end

Note how the action has no logging logic after this change. Also, you can target before and after action logic for specific actions, as the ctx.current_action will have the class name of the currently processed action. In the example above, logging will occur only for TwoAction and not for OneAction or ThreeAction.

Here is how you can declaratively add before_hooks or after_hooks to your Organizer from the outside:

SomeOrganizer.before_actions =
  lambda do |ctx|
    if ctx.current_action == TwoAction
      return unless ctx.user.role == 'admin'
      ctx.logger.info('admin is doing something')
    end
  end

These ideas are originally from Aspect Oriented Programming, read more about them here.

Expects and Promises

The expects and promises macros are rules for the inputs/outputs of an action. expects describes what keys it needs to execute, and promises makes sure the keys are in the context after the action is reduced. If either of them are violated, a LightService::ExpectedKeysNotInContextError or LightService::PromisedKeysNotInContextError exception respectively will be thrown.

This is how it's used:

class FooAction
  extend LightService::Action
  expects :baz
  promises :bar

  executed do |context|
    context.bar = context.baz + 2
  end
end

The expects macro will pull the value with the expected key from the context, and makes it available to you through a reader.

The promises macro will not only check if the context has the promised keys, it also sets them for you in the context if you use the accessor with the same name, much the same way as the expects macro works.

The context object is essentially a smarter-than-normal Hash. Take a look at this spec to see expects and promises used with and without accessors.

Default values for optional Expected keys

When you have an expected key that has a sensible default which should be used everywhere and only overridden on an as-needed basis, you can specify a default value. An example use-case is a flag that allows a failure from a service under most circumstances to avoid failing an entire workflow because of a non-critical action.

LightService provides two mechanisms for specifying default values:

  1. A static value that is used as-is
  2. A callable that takes the current context as a param

Using the above use case, consider an action that sends a text message. In most cases, if there is a problem sending the text message, it might be OK for it to fail. We will expect an allow_failure key, but set it with a default, like so:

class SendSMS
  extend LightService::Action
  expects :message, :user
  expects :allow_failure, default: true

  executed do |context|
    sms_api = SMSService.new(key: ENV["SMS_API_KEY"])
    status  = sms_api.send(ctx.user.mobile_number, ctx.message)

    if !status.sent_ok?
      ctx.fail!(status.err_msg) unless ctx.allow_failure
    end
  end
end

Default values can also be processed dynamically by providing a callable. Any values already specified in the context are available to it via Hash key lookup syntax. e.g.

class SendSMS
  extend LightService::Action
  expects :message, :user
  expects :allow_failure, default: ->(ctx) { !ctx[:user].admin? } # Admins must always get SMS'

  executed do |context|
    sms_api = SMSService.new(key: ENV["SMS_API_KEY"])
    status  = sms_api.send(ctx.user.mobile_number, ctx.message)

    if !status.sent_ok?
      ctx.fail!(status.err_msg) unless ctx.allow_failure
    end
  end
end

Note that default values must be specified one at a time on their own line.

You can then call an action or organizer that uses an action with defaults without specifying the expected key that has a default.

Key Aliases

The aliases macro sets up pairs of keys and aliases in an organizer. Actions can access the context using the aliases.

This allows you to put together existing actions from different sources and have them work together without having to modify their code. Aliases will work with or without action expects.

Say for example you have actions AnAction and AnotherAction that you've used in previous projects. AnAction provides :my_key but AnotherAction needs to use that value but expects :key_alias. You can use them together in an organizer like so:

class AnOrganizer
  extend LightService::Organizer

  aliases :my_key => :key_alias

  def self.call(order)
    with(:order => order).reduce(
      AnAction,
      AnotherAction,
    )
  end
end

class AnAction
  extend LightService::Action
  promises :my_key

  executed do |context|
    context.my_key = "value"
  end
end

class AnotherAction
  extend LightService::Action
  expects :key_alias

  executed do |context|
    context.key_alias # => "value"
  end
end

Logging

Enable LightService's logging to better understand what goes on within the series of actions, what's in the context or when an action fails.

Logging in LightService is turned off by default. However, turning it on is simple. Add this line to your project's config file:

LightService::Configuration.logger = Logger.new(STDOUT)

You can turn off the logger by setting it to nil or /dev/null.

LightService::Configuration.logger = Logger.new('/dev/null')

Watch the console while you are executing the workflow through the organizer. You should see something like this:

I, [DATE]  INFO -- : [LightService] - calling organizer <TestDoubles::MakesTeaAndCappuccino>
I, [DATE]  INFO -- : [LightService] -     keys in context: :tea, :milk, :coffee
I, [DATE]  INFO -- : [LightService] - executing <TestDoubles::MakesTeaWithMilkAction>
I, [DATE]  INFO -- : [LightService] -   expects: :tea, :milk
I, [DATE]  INFO -- : [LightService] -   promises: :milk_tea
I, [DATE]  INFO -- : [LightService] -     keys in context: :tea, :milk, :coffee, :milk_tea
I, [DATE]  INFO -- : [LightService] - executing <TestDoubles::MakesLatteAction>
I, [DATE]  INFO -- : [LightService] -   expects: :coffee, :milk
I, [DATE]  INFO -- : [LightService] -   promises: :latte
I, [DATE]  INFO -- : [LightService] -     keys in context: :tea, :milk, :coffee, :milk_tea, :latte

The log provides a blueprint of the series of actions. You can see what organizer is invoked, what actions are called in what order, what do the expect and promise and most importantly what keys you have in the context after each action is executed.

The logger logs its messages with "INFO" level. The exception to this is the event when an action fails the context. That message is logged with "WARN" level:

I, [DATE]  INFO -- : [LightService] - calling organizer <TestDoubles::MakesCappuccinoAddsTwoAndFails>
I, [DATE]  INFO -- : [LightService] -     keys in context: :milk, :coffee
W, [DATE]  WARN -- : [LightService] - :-((( <TestDoubles::MakesLatteAction> has failed...
W, [DATE]  WARN -- : [LightService] - context message: Can't make a latte from a milk that's too hot!

The log message will show you what message was added to the context when the action pushed the context into a failure state.

The event of skipping the rest of the actions is also captured by its logs:

I, [DATE]  INFO -- : [LightService] - calling organizer <TestDoubles::MakesCappuccinoSkipsAddsTwo>
I, [DATE]  INFO -- : [LightService] -     keys in context: :milk, :coffee
I, [DATE]  INFO -- : [LightService] - ;-) <TestDoubles::MakesLatteAction> has decided to skip the rest of the actions
I, [DATE]  INFO -- : [LightService] - context message: Can't make a latte with a fatty milk like that!

You can specify the logger on the organizer level, so the organizer does not use the global logger.

class FooOrganizer
  extend LightService::Organizer
  log_with Logger.new("/my/special.log")
end

Error Codes

You can add some more structure to your error handling by taking advantage of error codes in the context. Normally, when something goes wrong in your actions, you fail the process by setting the context to failure:

class FooAction
  extend LightService::Action

  executed do |context|
    context.fail!("I don't like what happened here.")
  end
end

However, you might need to handle the errors coming from your action pipeline differently. Using an error code can help you check what type of expected error occurred in the organizer or in the actions.

class FooAction
  extend LightService::Action

  executed do |context|
    unless (service_call.success?)
      context.fail!("Service call failed", error_code: 1001)
    end

    # Do something else

    unless (entity.save)
      context.fail!("Saving the entity failed", error_code: 2001)
    end
  end
end

Action Rollback

Sometimes your action has to undo what it did when an error occurs. Think about a chain of actions where you need to persist records in your data store in one action and you have to call an external service in the next. What happens if there is an error when you call the external service? You want to remove the records you previously saved. You can do it now with the rolled_back macro.

class SaveEntities
  extend LightService::Action
  expects :user

  executed do |context|
    context.user.save!
  end

  rolled_back do |context|
    context.user.destroy
  end
end

You need to call the fail_with_rollback! method to initiate a rollback for actions starting with the action where the failure was triggered.

class CallExternalApi
  extend LightService::Action

  executed do |context|
    api_call_result = SomeAPI.save_user(context.user)

    context.fail_with_rollback!("Error when calling external API") if api_call_result.failure?
  end
end

Using the rolled_back macro is optional for the actions in the chain. You shouldn't care about undoing non-persisted changes.

The actions are rolled back in reversed order from the point of failure starting with the action that triggered it.

See this acceptance test to learn more about this functionality.

You may find yourself directly using an action that can roll back by calling .execute instead of using it from within an Organizer. If this action fails and attempts a rollback, a FailWithRollbackError exception will be raised. This is so that the organizer can rollback the actions one by one. If you don't want to wrap your call to the action with a begin, rescue FailWithRollbackError block, you can introspect the context like so, and keep your usage of the action clean:

class FooAction
  extend LightService::Action

  executed do |context|
    # context.organized_by will be nil if run from an action,
    # or will be the class name if run from an organizer
    if context.organized_by.nil?
      context.fail!
    else
      context.fail_with_rollback!
    end
  end
end

Localizing Messages

Built-in localization adapter

The built-in adapter simply uses a manually created dictionary to search for translations.

# lib/light_service_translations.rb
LightService::LocalizationMap.instance[:en] = {
  :foo_action => {
    :light_service => {
      :failures => {
        :exceeded_api_limit => "API limit for service Foo reached. Please try again later."
      },
      :successes => {
        :yay => "Yaaay!"
      }
    }
  }
}
class FooAction
  extend LightService::Action

  executed do |context|
    unless service_call.success?
      context.fail!(:exceeded_api_limit)

      # The failure message used here equates to:
      # LightService::LocalizationMap.instance[:en][:foo_action][:light_service][:failures][:exceeded_api_limit]
    end
  end
end

Nested classes will work too: App::FooAction, for example, would be translated to app/foo_action hash key.

:en is the default locale, but you can switch it whenever you want with

LightService::Configuration.locale = :it

If you have I18n loaded in your project the default adapter will automatically be updated to use it. But would you want to opt for the built-in localization adapter you can force it with

LightService::Configuration.localization_adapter = LightService::LocalizationAdapter

I18n localization adapter

If I18n is loaded into your project, LightService will automatically provide a mechanism for easily translating your error or success messages via I18n.

class FooAction
  extend LightService::Action

  executed do |context|
    unless service_call.success?
      context.fail!(:exceeded_api_limit)

      # The failure message used here equates to:
      # I18n.t(:exceeded_api_limit, scope: "foo_action.light_service.failures")
    end
  end
end

This also works with nested classes via the ActiveSupport #underscore method, just as ActiveRecord performs localization lookups on models placed inside a module.

module PaymentGateway
  class CaptureFunds
    extend LightService::Action

    executed do |context|
      if api_service.failed?
        context.fail!(:funds_not_available)
      end

      # this failure message equates to:
      # I18n.t(:funds_not_available, scope: "payment_gateway/capture_funds.light_service.failures")
    end
  end
end

If you need to provide custom variables for interpolation during localization, pass that along in a hash.

module PaymentGateway
  class CaptureFunds
    extend LightService::Action

    executed do |context|
      if api_service.failed?
        context.fail!(:funds_not_available, last_four: "1234")
      end

      # this failure message equates to:
      # I18n.t(:funds_not_available, last_four: "1234", scope: "payment_gateway/capture_funds.light_service.failures")

      # the translation string itself being:
      # => "Unable to process your payment for account ending in %{last_four}"
    end
  end
end

Custom localization adapter

You can also provide your own custom localization adapter if your application's logic is more complex than what is shown here.

To provide your own custom adapter, use the configuration setting and subclass the default adapter LightService provides.

LightService::Configuration.localization_adapter = MyLocalizer.new

# lib/my_localizer.rb
class MyLocalizer < LightService::I18n::LocalizationAdapter

  # I just want to change the default lookup path
  # => "light_service.failures.payment_gateway/capture_funds"
  def i18n_scope_from_class(action_class, type)
    "light_service.#{type.pluralize}.#{action_class.name.underscore}"
  end
end

To get the value of a fail! or succeed! message, simply call #message on the returned context.

Orchestrating Logic in Organizers

The Organizer - Action combination works really well for simple use cases. However, as business logic gets more complex, or when LightService is used in an ETL workflow, the code that routes the different organizers becomes very complex and imperative.

In the past, this was solved using Orchestrators. As of Version 0.9.0 Orchestrators have been deprecated. All their functionality is now usable directly within Organizers. Read on to understand how to orchestrate workflows from within a single Organizer.

Let's look at a piece of code that does basic data transformations:

class ExtractsTransformsLoadsData
  def self.run(connection)
    context = RetrievesConnectionInfo.call(connection)
    context = PullsDataFromRemoteApi.call(context)

    retrieved_items = context.retrieved_items
    if retrieved_items.empty?
      NotifiesEngineeringTeamAction.execute(context)
    end

    retrieved_items.each do |item|
      context[:item] = item
      TransformsData.call(context)
    end

    context = LoadsData.call(context)

    SendsNotifications.call(context)
  end
end

The LightService::Context is initialized with the first action, that context is passed around among organizers and actions. This code is still simpler than many out there, but it feels very imperative: it has conditionals, iterators in it. Let's see how we could make it a bit more simpler with a declarative style:

class ExtractsTransformsLoadsData
  extend LightService::Organizer

  def self.call(connection)
    with(:connection => connection).reduce(actions)
  end

  def self.actions
    [
      RetrievesConnectionInfo,
      PullsDataFromRemoteApi,
      reduce_if(->(ctx) { ctx.retrieved_items.empty? }, [
        NotifiesEngineeringTeamAction
      ]),
      iterate(:retrieved_items, [
        TransformsData
      ]),
      LoadsData,
      SendsNotifications
    ]
  end
end

This code is much easier to reason about, it's less noisy and it captures the goal of LightService well: simple, declarative code that's easy to understand.

The 9 different orchestrator constructs an organizer can have:

  1. reduce_until
  2. reduce_if
  3. reduce_if_else
  4. reduce_case
  5. iterate
  6. execute
  7. with_callback
  8. add_to_context
  9. add_aliases

reduce_until behaves like a while loop in imperative languages, it iterates until the provided predicate in the lambda evaluates to true. Take a look at this acceptance test to see how it's used.

reduce_if will reduce the included organizers and/or actions if the predicate in the lambda evaluates to true. This acceptance test describes this functionality.

reduce_if_else takes three arguments, a condition lambda, a first set of "if true" steps, and a second set of "if false" steps. If the lambda evaluates to true, the "if true" steps are executed, otherwise the "else steps" are executed. This acceptance test describes this functionality.

reduce_case behaves like a Ruby case statement. The first parameter value is the key of the value within the context that will be worked with. The second parameter when is a hash where the keys are conditional values and the values are steps to take if the condition matches. The final parameter else is a set of steps to take if no conditions within the when parameter are met. This acceptance test describes this functionality.

iterate gives your iteration logic, the symbol you define there has to be in the context as a key. For example, to iterate over items you will use iterate(:items) in your steps, the context needs to have items as a key, otherwise it will fail. The organizer will singularize the collection name and will put the actual item into the context under that name. Remaining with the example above, each element will be accessible by the name item for the actions in the iterate steps. This acceptance test should provide you with an example.

To take advantage of another organizer or action, you might need to tweak the context a bit. Let's say you have a hash, and you need to iterate over its values in a series of action. To alter the context and have the values assigned into a variable, you need to create a new action with 1 line of code in it. That seems a lot of ceremony for a simple change. You can do that in a execute method like this execute(->(ctx) { ctx[:some_values] = ctx.some_hash.values }). This test describes how you can use it.

Use with_callback when you want to execute actions with a deferred and controlled callback. It works similar to a Sax parser, I've used it for processing large files. The advantage of it is not having to keep large amount of data in memory. See this acceptance test as a working example.

add_to_context can add key-value pairs on the fly to the context. This functionality is useful when you need a value injected into the context under a specific key right before the subsequent actions are executed. Keys are also made available as accessors on the context object and can be used just like methods exposed via expects and promises. This test describes its functionality.

Your action needs a certain key in the context but it's under a different one? Use the function add_aliases to alias an existing key in the context under the desired key. Take a look at this test to see an example.

ContextFactory for Faster Action Testing

As the complexity of your workflow increases, you will find yourself spending more and more time creating a context (LightService::Context it is) for your action tests. Some of this code can be reused by clever factories, but still, you are using a context that is artificial, and can be different from what the previous actions produced. This is especially true, when you use LightService in ETLs, where you start out with initial data and your actions are mutating its state.

Here is an example:

class SomeOrganizer
  extend LightService::Organizer

  def self.call(ctx)
    with(ctx).reduce(actions)
  end

  def self.actions
    [
       ETL::ParsesPayloadAction,
       ETL::BuildsEnititiesAction,
       ETL::SetsUpMappingsAction,
       ETL::SavesEntitiesAction,
       ETL::SendsNotificationAction
    ]
  end
end

You should test your workflow from the outside, invoking the organizer’s call method and verify that the data was properly created or updated in your data store. However, sometimes you need to zoom into one action, and setting up the context to test it is tedious work. This is where ContextFactory can be helpful.

In order to test the third action ETL::SetsUpMappingAction, you have to have several entities in the context. Depending on the logic you need to write code for, this could be a lot of work. However, by using the ContextFactory in your spec, you could easily have a prepared context that’s ready for testing:

require 'spec_helper'
require 'light-service/testing'

RSpec.describe ETL::SetsUpMappingsAction do
  let(:context) do
    LightService::Testing::ContextFactory
      .make_from(SomeOrganizer)
      .for(described_class)
      .with(:payload => File.read(‘spec/data/payload.json’)
  end

  it ‘works like it should’ do
    result = described_class.execute(context)
    expect(result).to be_success
  end
end

This context then can be passed to the action under test, freeing you up from the 20 lines of factory or fixture calls to create a context for your specs.

In case your organizer has more logic in its call method, you could create your own test organizer in your specs like you can see it in this acceptance test. This is reusable in all your action tests.

Rails support

LightService includes Rails generators for creating both Organizers and Actions along with corresponding tests. Currently only RSpec is supported (PR's for supporting MiniTest are welcome)

Note: Generators are namespaced to light_service not light-service due to Rake name constraints.

Organizer generation

rails generate light_service:organizer My::SuperFancy::Organizer
# -- or
rails generate light_service:organizer my/super_fancy/organizer

Options for this generator are:

Action generation

rails generate light_service:action My::SuperFancy::Action
# -- or
rails generate light_service:action my/super_fancy/action

Options for this generator are:

Advanced action generation

You are able to optionally specify expects and/or promises keys during generation

rails generate light_service:action CrankWidget expects:one_fish,two_fish promises:red_fish,blue_fish

When specifying expects, convenience variables will be initialized in the executed block so that you don't have to call them through the context. A stub context will be created in the test file using these keys too.

When specifying promises, specs will be created testing for their existence after executing the action.

Other implementations

Language Repo Author
Python pyservice @adomokos
PHP light-service @douglasgreyling
JavaScript light-service.js @douglasgreyling

Contributing

  1. Fork it
  2. Create your feature branch (git checkout -b my-new-feature)
  3. Commit your changes (git commit -am 'Added some feature')
  4. Push to the branch (git push origin my-new-feature)
  5. Create new Pull Request

Huge thanks to the contributors!

Release Notes

Follow the release notes in this document.

License

LightService is released under the MIT License.