Operations
This Page Under Construction
"A class that performs an action" A simple explanation of Service Objects for Ruby on Rails
Why do we need Service Objects? Because in any real world system you have logic that does not belong in models (or stores) because it effects multiple models or stores, and it does not belong in components because the logic of the task is independent of the specific user interface design. In MVC frameworks this kind of logic is often shoved in the controller, but it doesn't belong there either.
There are also those boundary areas between gathering and processing external data and getting data into or out of our stores and models. You don't want that kind of logic in your model or store, so where does it go? It belongs in a service object or Operation in Hyperstack terminology.
The term Operation, the key concepts of the Operation, and a lot of the implementation was taken from Trailblazer
Simply put an Operation is like a large standalone method that has no internal state of its own. You run an operation, it does it thing, and it returns an answer.
Any state that an operation needs to retrieve or save is stored somewhere else: in a model, a store, or even in a remote API. Once the operation completes, it has no memory of its own.
Being a stand-alone, glue and business logic method is an Operation's full time mission. The Hyperstack Operation
base class is therefor structured to make writing this kind of code easy.
An Operation may take parameters (params) just like any other method;
An Operation may validate the parameters;
An Operation then executes a number of steps;
The steps can be part of a success track or a failure track;
The value of the final step is returned to the caller;
And the results can be broadcast to any interested parties.
Hyperstack's Operations often involve asynchronous methods such as HTTP requests and so Operations always return Promises. Likewise each of the steps of an Operation can itself be an asynchronous action, and the Operation class will take care of chaining the promises together for you.
Another key feature of Operations is that because they are stateless they make a perfect RPC (Remote Procedure Call) mechanism. So an Operation can be called on the client, but will run on the server, and then return or broadcast the results to the clients. Thus Operations form the underlying data transport mechanism between the server and clients.
That is a lot to digest, and truly Operations are the swiss-army knife of Hyperstack. So let's dive into some examples.
In this simple example we are going to use a third-party API to determine our browser's IP address. First without Operations:
Nice and simple. Our App mounts, does a HTTP get from our API, and when it returns it updates the state. The problem is our view logic is cluttered up with low level specifics of how to get the address. Lets fix that by moving that logic to a separate service object:
Notice that the object is stateless and because it has no state it is simply a class method. We then use our service object like this:
If we were to change how we get the IP address, the Component now doesn't have to change.
Now we will redefine our service object using the Hyperstack::Operation class.
You invoke Operations using the run method, so our Component does not have to change at all.
The advantage is that the Operation syntax takes care of a lot of clutter, allows our promise to be chained neatly, and makes our intention clear to the reader.
We will see how these advantages multiply as our example becomes more complex.
Before moving on lets understand the basics of Operations.
Every Operation has as its external API a single
run
method.The work of the Operation is defined by a series of steps.
When the run method is called, the code associated with each step is executed.
If a step returns a promise the next step will wait till the promise is resolved.
The result of the final step is wrapped in a promise and is the result of the operation.
The final point means that regardless of the Operation's internal implementation, the Operation always returns a promise, so its API is consistent. As operations always return promises you can simply apply the then
and fail
promise methods directly to the Operation rather than saying Op.run.then
.
Let's say that rather than a simple ip address what we want is a full set of geo-location data. We can use another third party API to do the job. This API requires we supply our IP address, so we will reuse our IPAddress Op.
Here we can see one of the different ways to define a step: We simply delegate the first step to our already defined GetIPAddress
operation.
Again lets compare to a traditional ServiceObject:
Again its the same logic, but the body of our service object is over twice the number of lines and logic is obscured by the promise handlers.
It would be nice if we could include a flag icon to go with the country in the response data. Lets do that:
Of course its just Ruby, so we can further clean up our code by defining some helper methods:
Our GetGeoData
uses two remote third party operations, which may occasionally fail so we add a retry mechanism. This will introduce four new features of Operation: The failure track, parameters, and the abort!
and succeed!
methods.
Tracks
Operations have two tracks of execution. The normal success track which is defined by the step
method, and a failure track which is defined by a series of failed
methods.
Execution begins with the first step, and continues with each step until an exception is raised, or a promise fails. When that happens execution jumps to the next failed
step, and the continues executing failed
steps. The result of the Operation will be value of the last failed step, and the Operation's promise will be be rejected (i.e. will be in the fail state.)
Parameters
Operations can take a series of named parameters defined by the param
method. Parameters can have type information, defaults, and can be validated. This helps Operations act like a firewall between various parts of the system, making debugging and error handling easier. For now we are just going to use a simple case of a parameter that takes a default value.
The abort!
and succeed!
methods
These provide an early exit like return
, break
and next statements. Calling abort!
and succeed!
immediately exits the Operation by the appropriate track.
Putting it together:
; they orchestrate the interactions between Components, external services, Models, and Stores. Operations provide a tidy place to keep your business logic.
Operations receive parameters, va and execute a series of steps They have a simple structure which is not dissimilar to a Component:
Hyperstack's Isomorphic Operations span the client and server divide automagically. Operations can run on the client, the server, and traverse between the two.
This goal of this documentation is to outline Operations classes and provides enough information and examples to show how to implement Operations in an application.
Operations have three core functions
Operations are packaged as one neat package but perform three different functions:
Operations encapsulate business logic into a series of steps
Operations can dispatch messages (either on the client or between the client and server)
ServerOps can be used to replace boiler-plate APIs through a bi-directional RPC mechanism
Important to understand: There is no requirement to use all three functions. Use only the functionality your application requires.
Operations encapsulate business logic
In a traditional MVC architecture, the business logic ends up either in Controllers, Models, Views or some other secondary construct such as service objects, helpers, or concerns. In Hyperstack, Operations are first class objects who's job is to mutate state in the Stores, Models, and Components. Operations are discreet logic, which is of course, testable and maintainable.
An Operation does the following things:
receives incoming parameters, and does basic validations
performs any further validations
executes the operation
dispatches to any listeners
returns the value of the execution (step 3)
These are defined by series of class methods described below.
Operation Structure
Hyperstack::Operation
is the base class for an Operation
As an example, here is an Operation which ensures that the Model being saved always has the current created_by
and updated_by
Member
.
This Operation is run from anywhere in the client or server code:
Operations always return Promises, and those Promises can be chained together. See the section on Promises later in this documentation for details on how Promises work.
Operations can invoke other Operations so you can chain a sequence of steps
and Promises which proceed unless the previous step
fails:
This approach allows you to build readable and testable workflows in your application.
Running Operations
To run an Operation:
use the
run
method:
passing params:
the
then
andfail
methods, which will dispatch the operation and attach a promise handler:
Parameters
Operations can take parameters when they are run. Parameters are described and accessed with the same syntax as Hyperstack Components.
The parameter filter types and options are taken from the Mutations gem with the following changes:
In Hyperstack::Operations all params are declared with the param macro
The type can be specified using the
type:
optionArray and hash types can be shortened to
[]
and{}
Optional params either have the default value associated with the param name or by having the
default
option presentAll other Mutation filter options (such as
:min
) will work the same
All incoming params are validated against the param declarations, and any errors are posted to the @errors
instance variable. Extra params are ignored, but missing params unless they have a default value will cause a validation error.
Defining Execution Steps
Operations may define a sequence of steps to be executed when the operation is run, using the step
, failed
and async
callback macros.
step
: runs a callback - each step is run in order.failed
: runs a callback if a previousstep
or validation has failed.async
: will be explained below.
Together step
and failed
form two railway tracks. Initially, execution proceeds down the success track until something goes wrong; then execution switches to the failure track starting at the next failed
statement. Once on the failed track execution continues performing each failed
callback and skipping any step
callbacks.
Failure occurs when either an exception is raised, or a Promise fails (more on this in the next section.) The Ruby fail
keyword can be used as a simple way to switch to the failed track.
Both step
and failed
can receive any results delivered by the previous step. If the last step raised an exception (outside a Promise), the failure track would receive the exception object.
The callback may be provided to step
and failed
either as a block, a symbol (which will name a method), a proc, a lambda, or an Operation.
FYI: You can also use the Ruby next
keyword as expected to leave the current step and move to the next one.
Promises and Operations
Within the browser, the code does not wait for asynchronous methods (such as HTTP requests or timers) to complete. Operations use Opal's Promise library to deal with these situations cleanly. A Promise is an object that has three states: It is either still pending, or has been rejected (i.e. failed), or has been successfully resolved. A Promise can have callbacks attached to either the failed or resolved state, and these callbacks will be executed once the Promise is resolved or rejected.
If a step
or failed
callback returns a pending Promise then the execution of the operation is suspended, and the Operation will return the Promise to the caller. If there is more track ahead, then execution will resume at the next step when the Promise is resolved. Likewise, if the pending Promise is rejected execution will resume on the next failed
callback. Because of the way Promises work, the operation steps will all be completed before the resolved state is passed along to the caller so that everything will execute in its original order.
Likewise, the Operation's dispatch occurs when the Promise resolves as well.
The async
method can be used to override the waiting behavior. If a step
returns a Promise, and there is an async
callback further down the track, execution will immediately pick up at the async
. Any steps in between will still be run when the Promise resolves, but their results will not be passed outside of the operation.
These features make it easy to organize, understand and compose asynchronous code:
Operations will always return a Promise. If an Operation has no steps that return a Promise the value of the last step will be wrapped in a resolved Promise. Operations can be easily changed regardless of their internal implementation:
You can also use Promise#when
if you don't care about the order of Operations
Early Exits
Any step
or failed
callback, can have an immediate exit from the Operation using the abort!
and succeed!
methods. The abort!
method returns a failed Promise with any supplied parameters. The succeed!
method does an immediate dispatch and returns a resolved Promise with any supplied parameters. If succeed!
is used in a failed
callback, it will override the failed status of the Operation. This is especially useful if you want to dispatch in spite of failures:
Validation
An Operation can also have some validate
callbacks which will run before the first step. This is a handy place to put any additional validations. In the validate method you can add validation type messages using the add_error
method, and these will be passed along like any other param validation failures.
If the validate method returns a Promise, then execution will wait until the Promise resolves. If the Promise fails, then the current validation fails.
abort!
can be called from within validate
or add_error
to exit the Operation immediately. Otherwise, all validations will be run and collected together, and the Operation will move onto the failed
track. If abort!
is called within an add_error
callback the error will be added before aborting.
You can also raise an exception directly in validate if appropriate. If a Hyperstack::AccessViolation
exception is raised the Operation will immediately abort, otherwise just the current validation fails.
To avoid further validations if there are any failures in the basic parameter validations, this can be added
before the first validate
or add_error
call.
Handling Failed Operations
Because Operations always return a promise, the Promise's fail
method can be used on the Operation's result to detect failures.
Failures to validate params result in Hyperstack::ValidationException
which contains a Mutations error object.
Instance Versus Class Execution Context
Typically the Operation's steps are declared and run in the context of an instance of the Operation. An instance of the Operation is created, runs and is thrown away.
Sometimes it's useful to run a step (or other macro such as validate
) in the context of the class. This is useful especially for caching values between calls to the Operation. This can be done by defining the steps in the class context, or by providing the option scope: :class
to the step.
Note that the primary use should be in interfacing to an outside APIs. Application state should not be hidden inside an Operation, and it should be moved to a Store.
An instance of the operation is always created to hold the current parameter values, dispatcher, etc. The first parameter to a class level step
block or method (if it takes parameters) will always be the instance.
The Boot Operation
Hyperstack includes one predefined Operation, Hyperstack::Application::Boot
, that runs at system initialization. Stores can receive Hyperstack::Application::Boot
to initialize their state. To reset the state of the application, you can just execute Hyperstack::Application::Boot
Operations can dispatch messages
Hyperstack Operations borrow from the Flux pattern where Operations are dispatchers and Stores are receivers. The choice to use Operations in this depends entirely on the needs and design of your application.
To illustrate this point, here is the simplest Operation:
To 'Reset' the system you would say
Elsewhere your HyperStores can receive the Reset Dispatch using the receives
macro:
Note that multiple stores can receive the same Dispatch.
Note: Flux pattern vs. Hyperstack Operations Operations serve the role of both Action Creators and Dispatchers described in the Flux architecture. We chose the name
Operation
rather thanAction
orMutation
because we feel it best captures all the capabilities of aHyperstack::Operation
. Nevertheless, Operations are fully compatible with the Flux Pattern.
Dispatching With New Parameters
The dispatch
method sends the params
object on to any registered receivers. Sometimes it's useful to add additional outbound params before dispatching. Additional params can be declared using the outbound
macro:
Dispatching messages or invoking steps (or both)?
Facebook is very keen on their Flux architecture where messages are dispatched between receivers. In an extensive and complicated front end application it is easy to see why they are drawn to this architecture as it creates an independence and isolation between Components.
As stated earlier in this documentation, the step
idea came from Trailblazer, which is an alternative Rails architecture that posits that business functionality should not be kept in the Models, Controllers or Views.
In designing Hyperstack's Isomorphic Operations (which would run on the client and the server), we decided to borrow from the best of both architectures and let Operations work in either way. The decision as to adopt the dispatching or stepping based model is left down to the programmer as determined by their preference or the needs of their application.
ServerOps can be used to replace boiler-plate APIs
Some Operations simply do not make sense to run on the client as the resources they depend on may not be available on the client. For example, consider an Operation that needs to send an email - there is no mailer on the client so the Operation has to execute from the server.
That said, with our highest goal being developer productivity, it should be as invisible as possible to the developer where the Operation will execute. A developer writing front-end code should be able to invoke a server-side resource (like a mailer) just as easily as they might invoke a client-side resource.
Hyperstack ServerOps
replace the need for a boiler-plate HTTP API. All serialization and de-serialization of params are handled by Hyperstack. Hyperstack automagically creates the API endpoint needed to invoke a function from the client which executes on the server and returns the results (via a Promise) to the calling client-side code.
Server Operations
Operations will run on the client or the server. However, some Operations like ValidateUserDefaultCC
probably need to check information server side and make secure API calls to our credit card processor. Rather than build an API and controller to "validate the user credentials" you just specify that the operation must run on the server by using the Hyperstack::ServerOp
class.
A Server Operation will always run on the server even if invoked on the client. When invoked from the client, the ServerOp will receive the acting_user
param with the current value that your ApplicationController's acting_user
method returns. Typically the acting_user
method will return either some User model or nil (if there is no logged in user.) It's up to you to define how acting_user
is computed, but this is easily done with any of the popular authentication gems. Note that unless you explicitly add nils: true
to the param declaration, nil will not be accepted.
Note regarding Rails Controllers: Hyperstack is quite flexible and rides along side Rails, without interfering. So you could still have your old controllers, and invoke them the "non-Hyperstack" way by doing say an HTTP.post from the client, etc. Hyperstack adds a new mechanism for communicating between client and server called the Server Operation (which is a subclass of Operation.) A ServerOp has no implication on your existing controllers or code, and if used replaces controllers and client side API calls. HyperModel is built on top of Rails ActiveRecord models, and Server Operations, to keep models in sync across the application. ActiveRecord models that are made public (by moving them to the Hyperstack/models folder) will automatically be synchronized across the clients and the server (subject to permissions given in the Policy classes.) Like Server Operations, HyperModel completely removes the need to build controllers, and client side API code. However all of your current active record models, controllers will continue to work unaffected.
As shown above, you can also define a validation to ensure further that the acting user (with perhaps other parameters) is allowed to perform the operation. In the above case that is the only purpose of the Operation. Another typical use would be to make sure the current acting user has the correct role to perform the operation:
You can bake this kind logic into a superclass:
Because Operations always return a Promise, there is nothing to change on the client to call a Server Operation. A Server Operation will return a Promise that will be resolved (or rejected) when the Operation completes (or fails) on the server.
Isomorphic Operations
Unless the Operation is a Server Operation, it will run where it was invoked. This can be handy if you have an Operation that needs to run on both the server and the client. For example, an Operation that calculates the customers discount will want to run on the client so the user gets immediate feedback, and then will be run again on the server when the order is submitted as a double check.
Parameters and ServerOps
You cannot pass an object from the client to the server as a parameter as the server has no way of knowing the state of the object. Hyperstack takes a traditional implementation approach where an id (or some unique identifier) is passed as the parameter and the receiving code finds and created an instance of that object. For example:
Restricting server code to the server
There are valid cases where you will not want your ServerOp's code to be on the client yet still be able to invoke a ServerOp from client or server code. Good reasons for this would include:
Security concerns where you would not want some part of your code on the client
Size of code, where there will be unnecessary code downloaded to the client
Server code using backticks (`) or the %x{ ... } sequence, both of which are interpreted on the client as escape to generate JS code.
To accomplish this, you wrap the server side implementation of the ServerOp in a RUBY_ENGINE == 'opal'
test which acts as a compiler directive so that this code is not compiled by Opal.
There are several strategies you can use to apply the RUBY_ENGINE == 'opal' guard to your code.
Here is a fuller example:
Dispatching From Server Operations
You can also broadcast the dispatch from Server Operations to all authorized clients. The dispatch_to
will determine a list of channels to broadcast the dispatch to:
Channels
As seen above broadcasting is done over a Channel. Any Ruby class (including Operations) can be used as class channel. Any Ruby class that responds to the id
method can be used as an instance channel.
For example, the User
active record model could be a used as a channel to broadcast to all users. Each user instance could also be a separate instance channel that would be used to broadcast to a specific user.
The purpose of having channels is to restrict what gets broadcast to who, therefore typically channels represent connections to
the application (represented by the
Hyperstack::Application
class)or some function within the application (like an Operation)
or some class which is authenticated like a User or Administrator,
instances of those classes,
or instances of classes in some relationship - like a
team
that auser
belongs to.
A channel can be created by including the Hyperstack::Policy::Mixin
, which gives three class methods: regulate_class_connection
always_allow_connection
and regulate_instance_connections
.
For example...
will attach the current acting user to the User
channel (which is shared with all users) and to that user's private channel.
Both blocks execute with self
set to the current acting user, but the return value has a different meaning. If regulate_class_connection
returns any truthy value, then the class level connection will be made on behalf of the acting user. On the other hand, if regulate_instance_connection
returns an array (possibly nested) or Active Record relationship then an instance connection is made with each object in the list. So, for example, you could add:
To broadcast to all users, the Operation would have
or to send an announcement to a specific user
The above will work if PrivateAnnouncement
is invoked from the server, but usually, some other client would be sending the message so the operation could look like this:
On the client::
and elsewhere in the client code, there would be a component like this:
This will (in only 28 lines of code)
associate a channel with each logged in user
invoke the PrivateAnnouncement Operation on the server (remotely from the client)
validate that there is a logged in user at that client
validate that we have a non-nil, non-blank receiver and message
validate that the acting_user is an admin
look up the receiver in the database under their login name
dispatch the parameters back to any clients where the receiver is logged in
those clients will update their alert_messages state and
display the message
The dispatch_to
callback takes a list of classes, representing Channels. The Operation will be dispatched to all clients connected to those Channels. Alternatively dispatch_to
can take a block, a symbol (indicating a method to call) or a proc. The block, proc or method should return a single Channel, or an array of Channels, which the Operation will be dispatched to. The dispatch_to callback has access to the params object. For example, we can add an optional to
param to our Operation, and use this to select which Channel we will broadcast to.
Defining Connections in ServerOps
The policy methods always_allow_connection
and regulate_class_connection
may be used directly in a ServerOp class. This will define a channel dedicated to that class, and will also dispatch to that channel when the Operation completes.
Regulating Dispatches in Policy Classes
Regulations and dispatch lists can be grouped and specified in Policy files, which are by convention kept in the Rails app/policies
directory.
Serialization
If you need to control serialization and deserialization across the wire you can define the following class methods:
Accessing the Controller
ServerOps have the ability to receive the "controller" as a param. This is handy for low-level stuff (like login) where you need access to the controller. There is a subclass of ServerOp called ControllerOp that simply declares this param and will delegate any controller methods to the controller param. So within a ControllerOp
if you say session
you will get the session object from the controller.
Here is a sample of the SignIn operation using the Devise Gem:
In the code above there is another parameter type in ServerOps, called inbound, which will not get dispatched.
Broadcasting to the current_session
Let's say you would like to be able to broadcast to the current session. For example, after the user signs in we want to broadcast to all the browser windows the user happens to have open so that they can update.
For this, we have a current_session
method in the ControllerOp
that you can dispatch to.
The Session channel is special so to attach to the application to it you would say in the top level component:
Additional information
Operation Capabilities
Operations have the following capabilities:
Can easily be chained because they always return Promises
declare both their parameters and what they will dispatch
Parameters can be validated and type checked
Can run remotely on the server
Can be dispatched from the server to all authorized clients.
Can hold their own state data when appropriate
Operations also serves as the bridge between client and server
An operation can run on the client or the server and can be invoked remotely.
Use Operations as you choose. This architecture is descriptive but not prescriptive. Depending on the needs of your application and your overall thoughts on architecture, you may need a little or a lot of the functionality provided by Operations. If you chose, you could keep all your business logic in your Models, Stores or Components - we suggest that it is better application design not to do this, but the choice is yours.
Background
The design of Hyperstack's Operations have been inspired by three concepts: Trailblazer Operations (for encapsulating business logic in steps
), the Flux pattern (for dispatchers and receivers), and the Mutation Gem (for validating params).
Hyperstack Operations compared to Flux
Last updated