Action based Sequence for REST Server | LoopBack Documentation

Page Contents

Warning: Action-based sequence is now being phased out. Please use middleware-based sequence.

What is a Sequence?

A Sequence is a stateless grouping of Actions that control how a Server responds to requests.

The contract of a Sequence is simple: it must produce a response to a request. Creating your own Sequence gives you full control over how your Server instances handle requests and responses. The DefaultSequence looks like this:

export class DefaultSequence implements SequenceHandler {
  /**
   * Optional invoker for registered middleware in a chain.
   * To be injected via SequenceActions.INVOKE_MIDDLEWARE.
   */
  @inject(SequenceActions.INVOKE_MIDDLEWARE, {optional: true})
  protected invokeMiddleware: InvokeMiddleware = () => false;

  /**
   * Constructor: Injects findRoute, invokeMethod & logError
   * methods as promises.
   *
   * @param findRoute - Finds the appropriate controller method,
   *  spec and args for invocation (injected via SequenceActions.FIND_ROUTE).
   * @param parseParams - The parameter parsing function (injected
   * via SequenceActions.PARSE_PARAMS).
   * @param invoke - Invokes the method specified by the route
   * (injected via SequenceActions.INVOKE_METHOD).
   * @param send - The action to merge the invoke result with the response
   * (injected via SequenceActions.SEND)
   * @param reject - The action to take if the invoke returns a rejected
   * promise result (injected via SequenceActions.REJECT).
   */
  constructor(
    @inject(SequenceActions.FIND_ROUTE) protected findRoute: FindRoute,
    @inject(SequenceActions.PARSE_PARAMS) protected parseParams: ParseParams,
    @inject(SequenceActions.INVOKE_METHOD) protected invoke: InvokeMethod,
    @inject(SequenceActions.SEND) public send: Send,
    @inject(SequenceActions.REJECT) public reject: Reject,
  ) {}

  /**
   * Runs the default sequence. Given a handler context (request and response),
   * running the sequence will produce a response or an error.
   *
   * Default sequence executes these steps
   *  - Executes middleware for CORS, OpenAPI spec endpoints
   *  - Finds the appropriate controller method, swagger spec
   *    and args for invocation
   *  - Parses HTTP request to get API argument list
   *  - Invokes the API which is defined in the Application Controller
   *  - Writes the result from API into the HTTP response
   *  - Error is caught and logged using 'logError' if any of the above steps
   *    in the sequence fails with an error.
   *
   * @param context - The request context: HTTP request and response objects,
   * per-request IoC container and more.
   */
  async handle(context: RequestContext): Promise<void> {
    try {
      const {request, response} = context;
      // Invoke registered Express middleware
      const finished = await this.invokeMiddleware(context);
      if (finished) {
        // The response been produced by the middleware chain
        return;
      }
      const route = this.findRoute(request);
      const args = await this.parseParams(request, route);
      const result = await this.invoke(route, args);

      debug('%s result -', route.describe(), result);
      this.send(response, result);
    } catch (error) {
      this.reject(context, error);
    }
  }
}

Elements

In the example above, route, params, and result are all Elements. When building sequences, you use LoopBack Elements to respond to a request:

InvokeMiddleware
FindRoute
Request - (TBD) missing API docs link
Response - (TBD) missing API docs link
OperationRetVal
ParseParams
OpenAPISpec

Actions

Actions are JavaScript functions that only accept or return Elements. Since the input of one action (an Element) is the output of another action (Element) you can easily compose them. Below is an example that uses several built-in Actions:

class MySequence extends DefaultSequence {
  async handle(context: RequestContext) {
    try {
      // Invoke registered Express middleware
      const finished = await this.invokeMiddleware(context);
      if (finished) {
        // The response been produced by the middleware chain
        return;
      }
      // findRoute() produces an element
      const route = this.findRoute(context.request);
      // parseParams() uses the route element and produces the params element
      const params = await this.parseParams(context.request, route);
      // invoke() uses both the route and params elements to produce the result (OperationRetVal) element
      const result = await this.invoke(route, params);
      // send() uses the result element
      this.send(context.response, result);
    } catch (error) {
      this.reject(context, error);
    }
  }
}

Warning: Starting from v4.0.0 of @loopback/rest. The sequence adds an InvokeMiddleware action for CORS and OpenAPI spec endpoints as well as other middleware. See Middleware and Express Middleware for more details. For applications generated using old version of lb4, the src/sequence.ts needs to be manually updated with the code above.

Custom Sequences

Most use cases can be accomplished with DefaultSequence or by slightly customizing it. When an app is generated by the command lb4 app, a sequence file extending DefaultSequence at src/sequence.ts is already generated and bound for you so that you can easily customize it.

Here is an example where the application logs out a message before and after a request is handled:

import {DefaultSequence, Request, Response} from '@loopback/rest';

class MySequence extends DefaultSequence {
  log(msg: string) {
    console.log(msg);
  }
  async handle(context: RequestContext) {
    this.log('before request');
    await super.handle(context);
    this.log('after request');
  }
}

In order for LoopBack to use your custom sequence, you must register it before starting your Application:

import {RestApplication} from '@loopback/rest';

const app = new RestApplication();
app.sequence(MySequencce);

app.start();

Advanced topics

Customizing Sequence Actions

There might be scenarios where the default sequence ordering is not something you want to change, but rather the individual actions that the sequence will execute.

To do this, you’ll need to override one or more of the sequence action bindings used by the RestServer, under the RestBindings.SequenceActions constants.

As an example, we’ll implement a custom sequence action to replace the default “send” action. This action is responsible for returning the response from a controller to the client making the request.

To do this, we’ll register a custom send action by binding a Provider to the RestBindings.SequenceActions.SEND key.

First, let’s create our CustomSendProvider class, which will provide the send function upon injection.

/src/providers/custom-send.provider.ts

custom-send.provider.ts

import {Send, Response} from '@loopback/rest';
import {Provider, BoundValue, inject} from '@loopback/core';
import {writeResultToResponse, RestBindings, Request} from '@loopback/rest';

// Note: This is an example class; we do not provide this for you.
import {Formatter} from '../utils';

export class CustomSendProvider implements Provider<Send> {
  // In this example, the injection key for formatter is simple
  constructor(
    @inject('utils.formatter') public formatter: Formatter,
    @inject(RestBindings.Http.REQUEST) public request: Request,
  ) {}

  value() {
    // Use the lambda syntax to preserve the "this" scope for future calls!
    return (response: Response, result: OperationRetval) => {
      this.action(response, result);
    };
  }
  /**
   * Use the mimeType given in the request's Accept header to convert
   * the response object!
   * @param response - The response object used to reply to the  client.
   * @param result - The result of the operation carried out by the controller's
   * handling function.
   */
  action(response: Response, result: OperationRetval) {
    if (result) {
      // Currently, the headers interface doesn't allow arbitrary string keys!
      const headers = (this.request.headers as any) || {};
      const header = headers.accept || 'application/json';
      const formattedResult = this.formatter.convertToMimeType(result, header);
      response.setHeader('Content-Type', header);
      response.end(formattedResult);
    } else {
      response.end();
    }
  }
}

Our custom provider will automatically read the Accept header from the request context, and then transform the result object so that it matches the specified MIME type.

Next, in our application class, we’ll inject this provider on the RestBindings.SequenceActions.SEND key.

/src/application.ts

import {RestApplication, RestBindings} from '@loopback/rest';
import {
  RepositoryMixin,
  Class,
  Repository,
  juggler,
} from '@loopback/repository';
import {CustomSendProvider} from './providers/custom-send.provider';
import {Formatter} from './utils';
import {BindingScope} from '@loopback/core';

export class YourApp extends RepositoryMixin(RestApplication) {
  constructor() {
    super();
    // Assume your controller setup and other items are in here as well.
    this.bind('utils.formatter')
      .toClass(Formatter)
      .inScope(BindingScope.SINGLETON);
    this.bind(RestBindings.SequenceActions.SEND).toProvider(CustomSendProvider);
  }
}

As a result, whenever the send action of the DefaultSequence is called, it will make use of your function instead! You can use this approach to override any of the actions listed under the RestBindings.SequenceActions namespace.

Query string parameters and path parameters

OAI 3.0.x describes the data from a request’s header, query and path in an operation specification’s parameters property. In a Controller method, such an argument is typically decorated by @param(). We’ve made multiple shortcuts available to the @param() decorator in the form of @param.<http_source>.<OAI_primitive_type>. Using this notation, path parameters can be described as @param.path.string. Here is an example of a controller method which retrieves a Note model instance by obtaining the id from the path object.

@get('/notes/{id}', {
  responses: {
    '200': {
      description: 'Note model instance',
      content: {
        'application/json': {
          schema: getModelSchemaRef(Note, {includeRelations: true}),
        },
      },
    },
  },
})
async findById(
  @param.path.string('id') id: string,
  @param.filter(Note, {exclude: 'where'}) filter?: FilterExcludingWhere<Note>
): Promise<Note> {
  return this.noteRepository.findById(id, filter);
}

(Notice: the filter for findById() method only supports the include clause for now.)

You can also specify a parameter which is an object value encoded as a JSON string or in multiple nested keys. For a JSON string, a sample value would be location={"lang": 23.414, "lat": -98.1515}. For the same location object, it can also be represented as location[lang]=23.414&location[lat]=-98.1515. Here is the equivalent usage for @param.query.object() decorator. It takes in the name of the parameter and an optional schema or reference object for it.

@param.query.object('location', {
  type: 'object',
  properties: {lat: {type: 'number', format: 'float'}, long: {type: 'number', format: 'float'}},
})

The parameters are retrieved as the result of parseParams Sequence action. Please note that deeply nested properties are not officially supported by OAS yet and is tracked by OAI/OpenAPI-Specification#1706. Therefore, our REST API Explorer does not allow users to provide values for such parameters and unfortunately has no visible indication of that. This problem is tracked and discussed in swagger-api/swagger-js#1385.

Parsing Requests

Parsing and validating arguments from the request url, headers, and body. See page Parsing requests.

Invoking controller methods

The invoke sequence action simply takes the parsed request parameters from the parseParams action along with non-decorated arguments, calls the corresponding controller method or route handler method, and returns the value from it. The default implementation of invoke action calls the handler function for the route with the request specific context and the arguments for the function. It is important to note that controller methods use invokeMethod from @loopback/core and can be used with global and custom interceptors. See Interceptor docs for more details. The request flow for two route flavours is explained below.

For controller methods:

A controller instance is instantiated from the context. As part of the instantiation, constructor and property dependencies are injected. The appropriate controller method is invoked via the chain of interceptors.
Arguments decorated with @param are resolved using data parsed from the request. Arguments decorated with @inject are resolved from the context. Arguments with no decorators are set to undefined, which is replaced by the argument default value if it’s provided.

For route handlers, the handler function is invoked via the chain of interceptors. The array of method arguments is constructed using OpenAPI spec provided at handler registration time (either via .api() for full schema or .route() for individual route registration).

Writing the response

The send sequence action is responsible for writing the result of the invoke action to the HTTP response object. The default sequence calls send with (transformed) data. Under the hood, send performs all steps required to send back the response, from content-negotiation to serialization of the response body. In Express, the handler is responsible for setting response status code, headers and writing the response body. In LoopBack, controller methods and route handlers return data describing the response and it’s the responsibility of the Sequence to send that data back to the client. This design makes it easier to transform the response before it is sent.

LoopBack 4 does not yet provide first-class support for streaming responses, see Issue#2230. As a short-term workaround, controller methods are allowed to send the response directly, effectively bypassing send action. The default implementation of send is prepared to handle this case here.

Handling errors

There are many reasons why the application may not be able to handle an incoming request:

The requested endpoint (method + URL path) was not found.
Parameters provided by the client were not valid.
A backend database or a service cannot be reached.
The response object cannot be converted to JSON because of cyclic dependencies.
A programmer made a mistake and a TypeError is thrown by the runtime.
And so on.

In the Sequence implementation described above, all errors are handled by a single catch block at the end of the sequence, using the Sequence Action called reject.

The default implementation of reject does the following steps:

Call strong-error-handler to send back an HTTP response describing the error.
Log the error to stderr if the status code was 5xx (an internal server error, not a bad request).

To prevent the application from leaking sensitive information like filesystem paths and server addresses, the error handler is configured to hide error details.

For 5xx errors, the output contains only the status code and the status name from the HTTP specification. For example:
```
{
  "error": {
    "statusCode": 500,
    "message": "Internal Server Error"
  }
}
```
For 4xx errors, the output contains the full error message (error.message) and the contents of the details property (error.details) that ValidationError typically uses to provide machine-readable details about validation problems. It also includes error.code to allow a machine-readable error code to be passed through which could be used, for example, for translation.
```
{
  "error": {
    "statusCode": 422,
    "name": "Unprocessable Entity",
    "message": "Missing required fields",
    "code": "MISSING_REQUIRED_FIELDS"
  }
}
```

During development and testing, it may be useful to see all error details in the HTTP response returned by the server. This behavior can be enabled by enabling the debug flag in error-handler configuration as shown in the code example below. See strong-error-handler docs for a list of all available options.

app.bind(RestBindings.ERROR_WRITER_OPTIONS).to({debug: true});

An example error message when the debug mode is enabled:

{
  "error": {
    "statusCode": 500,
    "name": "Error",
    "message": "ENOENT: no such file or directory, open '/etc/passwords'",
    "errno": -2,
    "syscall": "open",
    "code": "ENOENT",
    "path": "/etc/passwords",
    "stack": "Error: a test error message\n    at Object.openSync (fs.js:434:3)\n    at Object.readFileSync (fs.js:339:35)"
  }
}

Keeping your Sequences

For most use cases, the default sequence supplied with LoopBack 4 applications is good enough for request-response handling pipeline. Check out Custom Sequences on how to extend it and implement custom actions.

Working with Express middleware

Warning: First-class support for Express middleware has been added to LoopBack since v4.0.0 of @loopback/rest. Please refer to Using Express Middleware. The following information only applies to earlier versions of @loopback/rest.

Under the hood, LoopBack leverages Express framework and its concept of middleware. To avoid common pitfalls, it is not possible to mount Express middleware directly on a LoopBack application. Instead, LoopBack provides and enforces a higher-level structure.

In a typical Express application, there are four kinds of middleware invoked in the following order:

Request-preprocessing middleware like cors or body-parser.
Route handlers handling requests and producing responses.
Middleware serving static assets (files).
Error handling middleware.

In LoopBack, we handle the request in the following steps:

The built-in request-preprocessing middleware is invoked.
The registered Sequence is started. The default implementation of findRoute and invoke actions will try to match the incoming request against the following resources:
1. Native LoopBack routes (controller methods, route handlers).
2. External Express routes (registered via mountExpressRouter API)
3. Static assets
Errors are handled by the Sequence using reject action.

Let’s see how different kinds of Express middleware can be mapped to LoopBack concepts:

Request-preprocessing middleware

At the moment, LoopBack does not provide API for mounting arbitrary middleware, we are discussing this feature in issues #1293 and #2035. Please up-vote them if you are interested in using Express middleware in LoopBack applications.

All applications come with cors enabled, this middleware can be configured via RestServer options - see Customize CORS.

While it is not possible to add additional middleware to a LoopBack application, it is possible to mount the entire LoopBack application as component of a parent top-level Express application where you can add arbitrary middleware as needed. You can find more details about this approach in Creating an Express Application with LoopBack REST API

Route handlers

In Express, a route handler is a middleware function that serves the response and does not call next(). Handlers can be registered using APIs like app.get(), app.post(), but also a more generic app.use().

In LoopBack, we typically use Controllers and Route handlers to implement request handling logic.

To support interoperability with Express, it is also possible to take an Express Router instance and add it to a LoopBack application as an external router - see Mounting an Express Router. This way it is possible to implement server endpoints using Express APIs.

Static files

LoopBack provides native API for registering static assets as described in Serve static files. Under the hood, static assets are served by serve-static middleware from Express.

The main difference between LoopBack and vanilla Express applications: LoopBack ensures that static-asset middleware is always invoked as the last one, only when no other route handled the request. This is important for performance reasons to avoid costly filesystem calls.

Error handling middleware

In Express, errors are handled by a special form of middleware, one that’s accepting four arguments: err, request, response, next. It’s up to the application developer to ensure that error handler is registered as the last middleware in the chain, otherwise not all errors may be routed to it.

In LoopBack, we use async functions instead of callbacks and thus can use simple try/catch flow to receive both sync and async errors from individual sequence actions. A typical Sequence implementation then passes these errors to the Sequence action reject.

You can learn more about error handling in Handling errors.