REST has quickly become the de-facto standard for building web services on the web because they’re easy to build and easy to consume.

There’s a much larger discussion to be had about how REST fits in the world of microservices, but — for this tutorial — let’s just look at building RESTful services.

Why REST? REST embraces the precepts of the web, including its architecture, benefits, and everything else. This is no surprise given its author, Roy Fielding, was involved in probably a dozen specs which govern how the web operates.

What benefits? The web and its core protocol, HTTP, provide a stack of features:

• Suitable actions (GET, POST, PUT, DELETE, …​)

• Caching

• Redirection and forwarding

• Security (encryption and authentication)

These are all critical factors on building resilient services. But that is not all. The web is built out of lots of tiny specs, hence it’s been able to evolve easily, without getting bogged down in "standards wars".

Developers are able to draw upon 3rd party toolkits that implement these diverse specs and instantly have both client and server technology at their fingertips.

By building on top of HTTP, REST APIs provide the means to build:

• Backwards compatible APIs

• Evolvable APIs

• Scaleable services

• Securable services

• A spectrum of stateless to stateful services

What’s important to realize is that REST, however ubiquitous, is not a standard, per se, but an approach, a style, a set of constraints on your architecture that can help you build web-scale systems. In this tutorial we will use the Spring portfolio to build a RESTful service while leveraging the stackless features of REST.

As we work through this tutorial, we’ll use Spring Boot. Go to Spring Initializr and add the following dependencies to a project:

• Web

• JPA

• H2

Change the Name to "Payroll" and then choose "Generate Project". A .zip will download. Unzip it. Inside you’ll find a simple, Maven-based project including a pom.xml build file (NOTE: You can use Gradle. The examples in this tutorial will be Maven-based.)

Spring Boot can work with any IDE. You can use Eclipse, IntelliJ IDEA, Netbeans, etc. The Spring Tool Suite is an open-source, Eclipse-based IDE distribution that provides a superset of the Java EE distribution of Eclipse. It includes features that make working with Spring applications even easier. It is, by no means, required. But consider it if you want that extra oomph for your keystrokes. Here’s a video demonstrating how to get started with STS and Spring Boot. This is a general introduction to familiarize you with the tools.

Let’s start off with the simplest thing we can construct. In fact, to make it as simple as possible, we can even leave out the concepts of REST. (Later on, we’ll add REST to understand the difference.)

Big picture: We’re going to create a simple payroll service that manages the employees of a company. We’ll store employee objects in a (H2 in-memory) database, and access them (via something called JPA). Then we’ll wrap that with something that will allow access over the internet (called the Spring MVC layer).

The following code defines an Employee in our system.

Despite being small, this Java class contains much:

• @Entity is a JPA annotation to make this object ready for storage in a JPA-based data store.

• id, name, and role are attributes of our Employee domain object. id is marked with more JPA annotations to indicate it’s the primary key and automatically populated by the JPA provider.

• a custom constructor is created when we need to create a new instance, but don’t yet have an id.

With this domain object definition, we can now turn to Spring Data JPA to handle the tedious database interactions.

Spring Data JPA repositories are interfaces with methods supporting creating, reading, updating, and deleting records against a back end data store. Some repositories also support data paging, and sorting, where appropriate. Spring Data synthesizes implementations based on conventions found in the naming of the methods in the interface.

Spring makes accessing data easy. By simply declaring the following EmployeeRepository interface we automatically will be able to

• Create new Employees

• Update existing ones

• Delete Employees

• Find Employees (one, all, or search by simple or complex properties)

To get all this free functionality, all we had to do was declare an interface which extends Spring Data JPA’s JpaRepository, specifying the domain type as Employee and the id type as Long.

Spring Data’s repository solution makes it possible to sidestep data store specifics and instead solve a majority of problems using domain-specific terminology.

Believe it or not, this is enough to launch an application! A Spring Boot application is, at a minimum, a public static void main entry-point and the @SpringBootApplication annotation. This tells Spring Boot to help out, wherever possible.

@SpringBootApplication is a meta-annotation that pulls in component scanning, autoconfiguration, and property support. We won’t dive into the details of Spring Boot in this tutorial, but in essence, it will fire up a servlet container and serve up our service.

Nevertheless, an application with no data isn’t very interesting, so let’s preload it. The following class will get loaded automatically by Spring:

What happens when it gets loaded?

• Spring Boot will run ALL CommandLineRunner beans once the application context is loaded.

• This runner will request a copy of the EmployeeRepository you just created.

• Using it, it will create two entities and store them.

Right-click and Run PayRollApplication, and this is what you get:

This isn’t the whole log, but just the key bits of preloading data. (Indeed, check out the whole console. It’s glorious.)

To wrap your repository with a web layer, you must turn to Spring MVC. Thanks to Spring Boot, there is little in infrastructure to code. Instead, we can focus on actions:

• @RestController indicates that the data returned by each method will be written straight into the response body instead of rendering a template.

• An EmployeeRepository is injected by constructor into the controller.

• We have routes for each operation (@GetMapping, @PostMapping, @PutMapping and @DeleteMapping, corresponding to HTTP GET, POST, PUT, and DELETE calls). (NOTE: It’s useful to read each method and understand what they do.)

• EmployeeNotFoundException is an exception used to indicate when an employee is looked up but not found.

When an EmployeeNotFoundException is thrown, this extra tidbit of Spring MVC configuration is used to render an HTTP 404:

• @ResponseBody signals that this advice is rendered straight into the response body.

• @ExceptionHandler configures the advice to only respond if an EmployeeNotFoundException is thrown.

• @ResponseStatus says to issue an HttpStatus.NOT_FOUND, i.e. an HTTP 404.

• The body of the advice generates the content. In this case, it gives the message of the exception.

To launch the application, either right-click the public static void main in PayRollApplication and select Run from your IDE, or:

Spring Initializr uses maven wrapper so type this:

Alternatively using your installed maven version type this:

When the app starts, we can immediately interrogate it.

This will yield:

Here you can see the pre-loaded data, in a compacted format.

If you try and query a user that doesn’t exist…​

You get…​

This message nicely shows an HTTP 404 error with the custom message Could not find employee 99.

It’s not hard to show the currently coded interactions…​

To create a new Employee record we use the following command in a terminal—the $ at the beginning signifies that what follows it is a terminal command:

Then it stores newly created employee and sends it back to us:

You can update the user. Let’s change his role.

And we can see the change reflected in the output.

Finally, you can delete users like this:

This is all well and good, but do we have a RESTful service yet? (If you didn’t catch the hint, the answer is no.)

What’s missing?

So far, you have a web-based service that handles the core operations involving employee data. But that’s not enough to make things "RESTful".

• Pretty URLs like /employees/3 aren’t REST.

• Merely using GET, POST, etc. isn’t REST.

• Having all the CRUD operations laid out isn’t REST.

In fact, what we have built so far is better described as RPC (Remote Procedure Call). That’s because there is no way to know how to interact with this service. If you published this today, you’d also have to write a document or host a developer’s portal somewhere with all the details.

This statement of Roy Fielding’s may further lend a clue to the difference between REST and RPC:

I am getting frustrated by the number of people calling any HTTP-based interface a REST API. Today’s example is the SocialSite REST API. That is RPC. It screams RPC. There is so much coupling on display that it should be given an X rating.

What needs to be done to make the REST architectural style clear on the notion that hypertext is a constraint? In other words, if the engine of application state (and hence the API) is not being driven by hypertext, then it cannot be RESTful and cannot be a REST API. Period. Is there some broken manual somewhere that needs to be fixed?

The side effect of NOT including hypermedia in our representations is that clients MUST hard code URIs to navigate the API. This leads to the same brittle nature that predated the rise of e-commerce on the web. It’s a signal that our JSON output needs a little help.

Introducing Spring HATEOAS, a Spring project aimed at helping you write hypermedia-driven outputs. To upgrade your service to being RESTful, add this to your build:

This tiny library will give us the constructs to define a RESTful service and then render it in an acceptable format for client consumption.

A critical ingredient to any RESTful service is adding links to relevant operations. To make your controller more RESTful, add links like this:

This is very similar to what we had before, but a few things have changed:

• The return type of the method has changed from Employee to EntityModel<Employee>. EntityModel<T> is a generic container from Spring HATEOAS that includes not only the data but a collection of links.

• linkTo(methodOn(EmployeeController.class).one(id)).withSelfRel() asks that Spring HATEOAS build a link to the EmployeeController 's one() method, and flag it as a self link.

• linkTo(methodOn(EmployeeController.class).all()).withRel("employees") asks Spring HATEOAS to build a link to the aggregate root, all(), and call it "employees".

What do we mean by "build a link"? One of Spring HATEOAS’s core types is Link. It includes a URI and a rel (relation). Links are what empower the web. Before the World Wide Web, other document systems would render information or links, but it was the linking of documents WITH this kind of relationship metadata that stitched the web together.

Roy Fielding encourages building APIs with the same techniques that made the web successful, and links are one of them.

If you restart the application and query the employee record of Bilbo, you’ll get a slightly different response than earlier:

This decompressed output shows not only the data elements you saw earlier (id, name and role), but also a _links entry containing two URIs. This entire document is formatted using HAL.

HAL is a lightweight mediatype that allows encoding not just data but also hypermedia controls, alerting consumers to other parts of the API they can navigate toward. In this case, there is a "self" link (kind of like a this statement in code) along with a link back to the aggregate root.

To make the aggregate root ALSO more RESTful, you want to include top level links while ALSO including any RESTful components within.

So we turn this

into this

Wow! That method, which used to just be repository.findAll(), is all grown up! Not to worry. Let’s unpack it.

CollectionModel<> is another Spring HATEOAS container; it’s aimed at encapsulating collections of resources—instead of a single resource entity, like EntityModel<> from earlier. CollectionModel<>, too, lets you include links.

Don’t let that first statement slip by. What does "encapsulating collections" mean? Collections of employees?

Not quite.

Since we’re talking REST, it should encapsulate collections of employee resources.

That’s why you fetch all the employees, but then transform them into a list of EntityModel<Employee> objects. (Thanks Java 8 Streams!)

If you restart the application and fetch the aggregate root, you can see what it looks like now.

For this aggregate root, which serves up a collection of employee resources, there is a top-level "self" link. The "collection" is listed underneath the "_embedded" section; this is how HAL represents collections.

And each individual member of the collection has their information as well as related links.

What is the point of adding all these links? It makes it possible to evolve REST services over time. Existing links can be maintained while new links can be added in the future. Newer clients may take advantage of the new links, while legacy clients can sustain themselves on the old links. This is especially helpful if services get relocated and moved around. As long as the link structure is maintained, clients can STILL find and interact with things.

In the code earlier, did you notice the repetition in single employee link creation? The code to provide a single link to an employee, as well as to create an "employees" link to the aggregate root, was shown twice. If that raised your concern, good! There’s a solution.

Simply put, you need to define a function that converts Employee objects to EntityModel<Employee> objects. While you could easily code this method yourself, there are benefits down the road of implementing Spring HATEOAS’s RepresentationModelAssembler interface—which will do the work for you.

This simple interface has one method: toModel(). It is based on converting a non-model object (Employee) into a model-based object (EntityModel<Employee>).

All the code you saw earlier in the controller can be moved into this class. And by applying Spring Framework’s @Component annotation, the assembler will be automatically created when the app starts.

To leverage this assembler, you only have to alter the EmployeeController by injecting the assembler in the constructor.

From here, you can use that assembler in the single-item employee method:

This code is almost the same, except instead of creating the EntityModel<Employee> instance here, you delegate it to the assembler. Maybe that doesn’t look like much.

Applying the same thing in the aggregate root controller method is more impressive:

The code is, again, almost the same, however you get to replace all that EntityModel<Employee> creation logic with map(assembler::toModel). Thanks to Java 8 method references, it’s super easy to plug it in and simplify your controller.

At this stage, you’ve created a Spring MVC REST controller that actually produces hypermedia-powered content! Clients that don’t speak HAL can ignore the extra bits while consuming the pure data. Clients that DO speak HAL can navigate your empowered API.

But that is not the only thing needed to build a truly RESTful service with Spring.

With one additional library and a few lines of extra code, you have added hypermedia to your application. But that is not the only thing needed to make your service RESTful. An important facet of REST is the fact that it’s neither a technology stack nor a single standard.

REST is a collection of architectural constraints that when adopted make your application much more resilient. A key factor of resilience is that when you make upgrades to your services, your clients don’t suffer from downtime.

In the "olden" days, upgrades were notorious for breaking clients. In other words, an upgrade to the server required an update to the client. In this day and age, hours or even minutes of downtime spent doing an upgrade can cost millions in lost revenue.

Some companies require that you present management with a plan to minimize downtime. In the past, you could get away with upgrading at 2:00 a.m. on a Sunday when load was at a minimum. But in today’s Internet-based e-commerce with international customers in other time zones, such strategies are not as effective.

SOAP-based services and CORBA-based services were incredibly brittle. It was hard to roll out a server that could support both old and new clients. With REST-based practices, it’s much easier. Especially using the Spring stack.

Supporting changes to the API

{
  "id": 1,
  "firstName": "Bilbo",
  "lastName": "Baggins",
  "role": "burglar",
  "name": "Bilbo Baggins",
  "_links": {
    "self": {
      "href": "http://localhost:8080/employees/1"
    },
    "employees": {
      "href": "http://localhost:8080/employees"
    }
  }
}

link:evolution/src/main/java/payroll/Employee.java[]

link:evolution/src/main/java/payroll/LoadDatabase.java[]

link:evolution/src/main/java/payroll/EmployeeController.java[]

$ curl -v -X POST localhost:8080/employees -H 'Content-Type:application/json' -d '{"name": "Samwise Gamgee", "role": "gardener"}'

> POST /employees HTTP/1.1
> Host: localhost:8080
> User-Agent: curl/7.54.0
> Accept: */*
> Content-Type:application/json
> Content-Length: 46
>
< Location: http://localhost:8080/employees/3
< Content-Type: application/hal+json;charset=UTF-8
< Transfer-Encoding: chunked
< Date: Fri, 10 Aug 2018 19:44:43 GMT
<
{
  "id": 3,
  "firstName": "Samwise",
  "lastName": "Gamgee",
  "role": "gardener",
  "name": "Samwise Gamgee",
  "_links": {
    "self": {
      "href": "http://localhost:8080/employees/3"
    },
    "employees": {
      "href": "http://localhost:8080/employees"
    }
  }
}

link:evolution/src/main/java/payroll/EmployeeController.java[]

$ curl -v -X PUT localhost:8080/employees/3 -H 'Content-Type:application/json' -d '{"name": "Samwise Gamgee", "role": "ring bearer"}'

* TCP_NODELAY set
* Connected to localhost (::1) port 8080 (#0)
> PUT /employees/3 HTTP/1.1
> Host: localhost:8080
> User-Agent: curl/7.54.0
> Accept: */*
> Content-Type:application/json
> Content-Length: 49
>
< HTTP/1.1 201
< Location: http://localhost:8080/employees/3
< Content-Type: application/hal+json;charset=UTF-8
< Transfer-Encoding: chunked
< Date: Fri, 10 Aug 2018 19:52:56 GMT
{
	"id": 3,
	"firstName": "Samwise",
	"lastName": "Gamgee",
	"role": "ring bearer",
	"name": "Samwise Gamgee",
	"_links": {
		"self": {
			"href": "http://localhost:8080/employees/3"
		},
		"employees": {
			"href": "http://localhost:8080/employees"
		}
	}
}

link:evolution/src/main/java/payroll/EmployeeController.java[]

$ curl -v -X DELETE localhost:8080/employees/1

* TCP_NODELAY set
* Connected to localhost (::1) port 8080 (#0)
> DELETE /employees/1 HTTP/1.1
> Host: localhost:8080
> User-Agent: curl/7.54.0
> Accept: */*
>
< HTTP/1.1 204
< Date: Fri, 10 Aug 2018 21:30:26 GMT

So far, you’ve built an evolvable API with bare bones links. To grow your API and better serve your clients, you need to embrace the concept of Hypermedia as the Engine of Application State.

What does that mean? In this section, you’ll explore it in detail.

Business logic inevitably builds up rules that involve processes. The risk of such systems is we often carry such server-side logic into clients and build up strong coupling. REST is about breaking down such connections and minimizing such coupling.

To show how to cope with state changes without triggering breaking changes in clients, imagine adding a system that fulfills orders.

As a first step, define an Order record:

• The class requires a JPA @Table annotation changing the table’s name to CUSTOMER_ORDER because ORDER is not a valid name for table.

• It includes a description field as well as a status field.

Orders must go through a certain series of state transitions from the time a customer submits an order and it is either fulfilled or cancelled. This can be captured as a Java enum:

This enum captures the various states an Order can occupy. For this tutorial, let’s keep it simple.

To support interacting with orders in the database, you must define a corresponding Spring Data repository:

With this in place, you can now define a basic OrderController:

• It contains the same REST controller setup as the controllers you’ve built so far.

• It injects both an OrderRepository as well as a (not yet built) OrderModelAssembler.

• The first two Spring MVC routes handle the aggregate root as well as a single item Order resource request.

• The third Spring MVC route handles creating new orders, by starting them in the IN_PROGRESS state.

• All the controller methods return one of Spring HATEOAS’s RepresentationModel subclasses to properly render hypermedia (or a wrapper around such a type).

Before building the OrderModelAssembler, let’s discuss what needs to happen. You are modeling the flow of states between Status.IN_PROGRESS, Status.COMPLETED, and Status.CANCELLED. A natural thing when serving up such data to clients is to let the clients make the decision on what it can do based on this payload.

But that would be wrong.

What happens when you introduce a new state in this flow? The placement of various buttons on the UI would probably be erroneous.

What if you changed the name of each state, perhaps while coding international support and showing locale-specific text for each state? That would most likely break all the clients.

Enter HATEOAS or Hypermedia as the Engine of Application State. Instead of clients parsing the payload, give them links to signal valid actions. Decouple state-based actions from the payload of data. In other words, when CANCEL and COMPLETE are valid actions, dynamically add them to the list of links. Clients only need show users the corresponding buttons when the links exist.

This decouples clients from having to know WHEN such actions are valid, reducing the risk of the server and its clients getting out of sync on the logic of state transitions.

Having already embraced the concept of Spring HATEOAS RepresentationModelAssembler components, putting such logic in the OrderModelAssembler would be the perfect place to capture this business rule:

This resource assembler always includes the self link to the single-item resource as well as a link back to the aggregate root. But it also includes two conditional links to OrderController.cancel(id) as well as OrderController.complete(id) (not yet defined). These links are ONLY shown when the order’s status is Status.IN_PROGRESS.

If clients can adopt HAL and the ability to read links instead of simply reading the data of plain old JSON, they can trade in the need for domain knowledge about the order system. This naturally reduces coupling between client and server. And it opens the door to tuning the flow of order fulfillment without breaking clients in the process.

To round out order fulfillment, add the following to the OrderController for the cancel operation:

It checks the Order status before allowing it to be cancelled. If it’s not a valid state, it returns an RFC-7807 Problem, a hypermedia-supporting error container. If the transition is indeed valid, it transitions the Order to CANCELLED.

And add this to the OrderController as well for order completion:

This implements similar logic to prevent an Order status from being completed unless in the proper state.

Let’s update LoadDatabase to pre-load some Order​s along with the Employee​s it was loading before.

Now you can test things out!

To use the newly minted order service, just perform a few operations:

This HAL document immediately shows different links for each order, based upon its present state.

• The first order, being COMPLETED only has the navigational links. The state transition links are not shown.

• The second order, being IN_PROGRESS additionally has the cancel link as well as the complete link.

Try cancelling an order:

This response shows an HTTP 200 status code indicating it was successful. The response HAL document shows that order in its new state (CANCELLED). And the state-altering links gone.

If you try the same operation again…​

…​you see an HTTP 405 Method Not Allowed response. DELETE has become an invalid operation. The Problem response object clearly indicates that you aren’t allowed to "cancel" an order already in the "CANCELLED" status.

Additionally, trying to complete the same order also fails:

With all this in place, your order fulfillment service is capable of conditionally showing what operations are available. It also guards against invalid operations.

By leveraging the protocol of hypermedia and links, clients can be built sturdier and less likely to break simply because of a change in the data. And Spring HATEOAS eases building the hypermedia you need to serve to your clients.

Throughout this tutorial, you have engaged in various tactics to build REST APIs. As it turns out, REST isn’t just about pretty URIs and returning JSON instead of XML.

Instead, the following tactics help make your services less likely to break existing clients you may or may not control:

• Don’t remove old fields. Instead, support them.

• Use rel-based links so clients don’t have to hard code URIs.

• Retain old links as long as possible. Even if you have to change the URI, keep the rels so older clients have a path onto the newer features.

• Use links, not payload data, to instruct clients when various state-driving operations are available.

It may appear to be a bit of effort to build up RepresentationModelAssembler implementations for each resource type and to use these components in all of your controllers. But this extra bit of server-side setup (made easy thanks to Spring HATEOAS) can ensure the clients you control (and more importantly, those you don’t) can upgrade with ease as you evolve your API.

This concludes our tutorial on how to build RESTful services using Spring. Each section of this tutorial is managed as a separate subproject in a single github repo:

• nonrest — Simple Spring MVC app with no hypermedia

• rest — Spring MVC + Spring HATEOAS app with HAL representations of each resource

• evolution — REST app where a field is evolved but old data is retained for backward compatibility

• links — REST app where conditional links are used to signal valid state changes to clients

To view more examples of using Spring HATEOAS see https://github.com/spring-projects/spring-hateoas-examples.

To do some more exploring check out the following video by Spring teammate Oliver Gierke:

https://raw.githubusercontent.com/spring-guides/getting-started-macros/master/footer.adoc