Skip to content

Hello, web services

Narsil edited this page Dec 16, 2012 · 7 revisions

Hello, web services

Nowadays, numerous web applications offer their features as web services, APIs that can be used by other web applications or native clients. This is how Twitter, GitHub, Google Maps or countless others can be scripted by third-party applications, using cleanly defined and easily accessible protocols.

With Opa, offering a web service is just as simple as creating any other form of web application. In this chapter, instead of writing a new application, we will extend our wiki to make it accessible through such a web API. This task will lead us through REST web service design, command-line testing of Opa services, management of URI queries and more.

TIP:

Several protocols share the landscape of web services, in particular REST (Representational State Transfer, a simple standard which does not specify how messages should be formated, only how they should be exchanged), SOAP (a more complex standard imposing conventions on the formatting of messages) and WSDL (a higher-level protocol). In this chapter, we'll only cover REST.

Overview

In this chapter, we will modify our wiki to make it accessible by a REST API. This involves few changes from the original wiki, only the addition of a few cases to differentiate between several kinds of requests that can be sent by a client -- which does not need to be a browser anymore.

If you are curious, this is the full source code of the REST wiki server.

Run | Fork

We will now walk through the concepts introduced in this listing.

Removing topics

In the rest of this chapter (pun intended), we will want to be able to delete a topic previously added to the wiki. Adding this feature (without showing it in the user interface) is just the matter of one line, as follows:

function remove_topic(topic) {
    Db.remove(@/wiki[topic]);
}

In this extract, we use function Db.remove, a function whose sole role is to remove the contents of a database path. Notice the @ before /wiki[topic]? This symbol signifies that we are not working with the value /wiki[topic] but with the path itself. If we had omitted this symbol, the Opa compiler would have complained that Db.remove cannot work with a string -- which is absolutely true.

Resting a little

A web service behaves much like a web application, without the client part. In other word, as any Opa web application, it starts with a server declared with Server.start:

The server, with an entry point for rest
function topic_of_path(path) {
    String.capitalize(String.to_lower(List.to_string_using("", "", "::", path)));
}

function start(url) {
    match (url) {
    case {path: [] ... }: display("Hello");
    case {path: ["_rest_" | path] ...}: rest(topic_of_path(path));
    case {~path ...}: display(topic_of_path(path));
    }
}

Server.start(Server.http,
  [ {bundle: @static_include_directory("resources")}
  , {dispatch: start}
  ]
)

In this version of start, we have slightly altered our pattern-matching to handle the case of paths starting with "_rest_". We decide that such paths are actually entry points for REST-based requests and handle them as such. Here, we delegate the management to function rest, which we write immediately:

As you may see, this function is also quite simple:

Handling rest requests
function rest(topic) {
  match (HttpRequest.get_method()) {
  case {some: method} :
      match (method) {
      case {post}:
          _ = save_source(
              topic,
              match (HttpRequest.get_body()) {
              case ~{some}: some;
              case {none}: "";
              }
          );
          Resource.raw_status({success});
      case {delete}:
          remove_topic(topic);
          Resource.raw_status({success});
      case {get}:
          Resource.raw_response(load_source(topic), "text/plain", {success});
      default:
          Resource.raw_status({method_not_allowed});
      }
    default:
      Resource.raw_status({bad_request});
  }
}

First, notice that rest is based on pattern-matching. Expect to meet pattern-matching constantly in Opa. The first three patterns are built from some of to the distinct verbs of the standard vocabulary of REST (these verbs are called Http methods):

  • {post} is used to place information on a server, here to add some content to the wiki;
  • {delete} is used to remove information from the server, here remove a topic from the wiki;
  • {get} is used to get information from a server, here to download the source code of an entry.

From these verbs, we build the following patterns:

  • {some: {post}}, i.e. the Http method is defined and is post;
  • {some: {delete}}, i.e. the Http method is defined and is delete;
  • {some: {get}}, i.e. the Http method is defined and is get;
  • _, i.e. any other case, whether the Http method is not defined or whether it is a method that we do not wish to handle.

Everything else in rest is simply function calls. You can find the definition of each function in the API documentation, so we will just introduce quickly the functions you have not seen yet:

  • Function HttpRequest.get_method has type -> option(method). If the function is called from a request and this request has a method m, it produces {some: m}. Otherwise, it produces {none}.
  • Similarly, HttpRequest.get_body has type -> option(string). If the function is called from a request containing a body b, it produces {some: b}. Otherwise, it produces {none}.
  • Function Resource.raw_response has type string, string, status -> resource. It produces a resource with a body from its body, its MIME type, and a status. This function is commonly used to reply to REST requests.
  • Finally, function Resource.raw_status has type status -> resource. It produces an empty resource, and is generally used to return an error to a REST request.

As pattern-matching against an option is very common, Opa provides an operator ? that can be used to make the above extract shorter and more readable. Expression a?b is equivalent to the following three lines:

match (a) {
case {none}: b
casse ~{some}: some

With this expression, we may rewrite our extract as follows:

Handling rest requests (shorter variant)
function rest(topic) {
    match (HttpRequest.get_method()) {
    case {some : {post}} :
        _ = save_source(topic, HttpRequest.get_body() ? "");
        Resource.raw_status({success});
    case {some : {delete}} :
        remove_topic(topic);
        Resource.raw_status({success});
    case {some : {get}} :
        Resource.raw_response(load_source(topic), "text/plain", {success});
    default :
        Resource.raw_status({method_not_allowed});
    }
}

And with this, we are done! Our wiki can now be scripted by external web applications:

SourceRun | Fork

All in all, the changes required a dozen lines of code.

Exercises will show you how to introduce more complex forms of scripting.

Testing it

The simplest way of testing a REST API is to use a command-line tool that lets you place requests directly, for instance curl or wget. Assuming that curl is installed on your system, the following command-line will test the result of placing a {get} request at address _rest_/hello:

curl localhost:8080/_rest_/hello

Execute this command-line and curl will show you the result of the call.

Similarly, the following command-line will test the result of placing a {post} request at the same address:

curl localhost:8080/_rest_/hello -d "I've just POSTed to change the contents of my wiki"

Now, we are not here to learn about curl, but to learn about Opa. And what best way to test the REST API of a wiki than by writing a web front-end that does not rely on its own database but on that of the wiki we have just defined?

We will do just this in the next chapter.

Questions

When is a method or a body not defined?

As mentioned, functions HttpServer.get_method and HttpServer.get_body can produce result {none} if the http method (respectively the body) does not exist.

This may be surprising, as, by definition of the protocols, every request has a method (not all have a body). Indeed, the only case in which HttpServer.get_method returns {none} is when there is no request, i.e. when the function has been called by the server for its own use and not during the execution of a request on behalf of a web browser or a distant web server.

On the other hand, many requests do not have a body. Function HttpServer.get_body returns {none} when there is no body, or when there is no request, as above.

Only one server?

If you have started thinking about large applications, at this stage, you might start worrying about having to centralize all your path management into only one pattern-matching, which could hurt modularity and hamper your work.

No need to worry, though, as we've already seen with Opa, you may combine any number of servers in an application. Take a look at all the variants in Server.handler, which is the type of the second argument to Server.start, to see other ways of constructing servers.

Exercises

Rest for chat

Add a REST API to your chat, with the following feature:

  • use a {post} request send a message for immediate display into the chat (for the moment, we will assume that the message has been written by author "ghost").

TIP:

To deal with several entry points, you will have to rewrite your server and replace one_page_bundle by a dispatcher. For these exercises, we decide that any request placed on path _rest_ is a REST request.

For testing, use the following command-line (assuming that curl is installed on your system):

curl localhost:8080/_rest_ -d "Whispers..."

Rest for chat, with logs

If you have not done so yet, update your chat to maintain conversation logs in the database.

Now, add the following REST API:

  • use a {get} request to get the log of messages as string containing one message per line.

Remember, use function List.to_string_using to convert a list to a string.

Rest for chat, with queries

For this exercise, we wish to extend the REST API for the chat to be able to send a message and give a name to the author of the message. For this purpose, we need to send more informations than simply {post}. In the REST world, there are typically two ways of passing additional informations: either in the URI itself or in the body of the request. For this exercise, we will see the first option:

  • if a {post} request is received on _rest_ and if the query of the request contains a pair ("author", x), use the value of x as the author name;
  • otherwise, use name "ghost", as above.

TIP: About queries

A query is an element of a URI. From the user's perspective, queries look like ?author=name&arg2=val2&arg3=val3. From the developer's perspective, the query is contained in field query of the URI, just as path. This field contains a list of pairs with the name of the argument and its value. So, for the previous query, the list will look like:

[("author", "name"), ("arg2", "val2"), ("arg3", "val3")]

Note that the order of these arguments is meaningless.

TIP: About association lists

Lists of pairs containing a name and a value (or, more generally, a key and a value) are generally called "association lists".

In Opa, the most common function to extract a value from an association list is List.assoc. This function takes two arguments: the key to search and the list in which to search. Its result is an option which may contain either {none} (if the key does not appear in the list) or {some: v} (if the key appears in the list, associated to value v).

Rest for chat, with JSON

Another common technique used among REST services is to pass additional information as part of the body of the request, often formated using the JavaScript Object Notation language (or JSON). The objective of this exercise is to use JSON instead of the URI to send the author name to the server.

  • if a {post} request is received on _rest_ and if the body of the request is a valid JSON construction containing a field "author", use the value associated to this field as the author name;
  • otherwise, use name "ghost", as above.

TIP: JSON requests

To obtain the JSON body of a request, use function HttpRequest.get_json_body.

TIP: About JSON

JSON is a format of strings which can be interpreted as simple data structures. In Opa, a string in JSON format can be transformed into a value with type RPC.Json.json by using function

Json.deserialize: string -> option(RPC.Json.json)

Note that this function can return {none} if the string was incorrectly formated.

The opposite operation is implemented by function

Json.serialize: RPC.Json.json -> string

Type RPC.Json.json is defined as follows:

type RPC.Json.json =
   { int   Int }
or { float Float }
or { string String }
or { bool Bool }
or { list(RPC.Json.json) List }
or { list((string, RPC.Json.json)) Record }

As above, case Record corresponds to a list of associations.