airship-server

Airship

Introduction

Airship is a framework for Node.JS & TypeScript that helps you to write big, scalable and maintainable API servers.

Main features:

• clean and simple architecture
• statically typed
• simple API
• automatic models serialization and deserialization
• automatic API Scheme generation
• automatic SDK generation for the web frontend
• automatic documentation generation
• ability to switch transport protocol (at this moment only HTTP is implemented)

Basic concepts

The main idea is very simple, every request has its model (its just a class), every response has a model too. To handle request there is a component called RequestHandler, every request handler handles a specific request and returns specific response. Its important that at this point whole system does not even know anything about the network and it should not. Because of that, your system is very abstract, it just handles specified requests and returns specified responses. This gives you the ability to change your network protocol or even stop using your system like a web server and use it as a part of local UI application.

Installation

To install the stable version:

npm install --save airship-server

This assumes you are using npm as your package manager. If you don’t, you can access these files on unpkg, download them, or point your package manager to them.

Basic example

Let's imagine that we need a web server with just one method /randomInt which returns a random integer in range.

First of all, we need to write our request and response models:

@queryPath('/getRandomInt')
export class RandomIntRequest extends ASRequest {
    @serializable()
    public readonly min: number
 
    @serializable()
    public readonly  max: number
 
    constructor(
        min: number,
        max: number
    ) {
        super()
 
        this.min = min
        this.max = max
    }
}

So, it's pretty simple, we just created a class that extends from base request class and have from and to properties. The only interesting thing is using of @serializable() decorator, it's used to have the ability to serialize and deserialize our model. And we also setting query path using @queryPath decorator, that string will be used for URL in HTTP & for methods names in SDK

Response model is also very simple and extends from base response model.

export class RandomIntResponse extends ASResponse {
    @serializable()
    public readonly integer: number
 
    constructor(integer: number) {
        super()
 
        this.integer = integer
    }
}

Now we need to write a handler for our method, it's also very simple, we just need to write a class that extends from BaseRequestHandler class and implement two methods:

export class RandomIntHandler extends BaseRequestHandler {
    public async handle(request: RandomIntRequest): Promise<RandomIntResponse> {
 
        let randomInt = Math.round(Math.random() * (request.max - request.min) + request.min)
 
        return new RandomIntResponse(randomInt)
    }
 
    public supports(request: Request): boolean {
        return request instanceof RandomIntRequest
    }
}

handle method gets an instance of our request method and returns an instance of response. supports method just tells the system which requests are supported by this handler.

Now we just need to set this up:

let logger = new ConsoleLogger()
 
const server = new AirshipAPIServer({
    requestsProvider: new HttpRequestsProvider(
        logger,
        7000,
 
        RandomIntRequest
    ),
 
    requestsHandler: new RequestHandlersManager([
        new RandomIntHandler()
    ])
})
 
server.start()

So the top level component is AirshipAPIServer, you need to pass at least two things to it: requests provider which somehow gets request. In our case, we are using HttpRequestsProvider, so requests are coming from the network over HTTP at 7000 port. HttpRequestsProvider needs a logger, the port, and list of supported request models.

The second required argument is requestsHandler, because in our case we have just one method - we could have been passed just an instance of RandomIntHandler. But when you have several methods you should use RequestHandlersManager, that class itself extends BaseRequestHandler and finds which handler can handle any request, you just need to pass your handlers.

The last thing we need to do is start the server, that's it!

Models serialization

Let's start with example:

class Square {
    public width: number
 
    public height: number
 
    constructor(
        width: number,
        height: number
    ) {
        this.width = width
        this.height = height
    }
}

Here we have simple Square model, first of all if we want this model to be serializable - we need to implement ISerializable interface:

class Square implements ISerializable {
    public width: number
 
    public height: number
 
    constructor(
        width: number,
        height: number
    ) {
        this.width = width
        this.height = height
    }
}

Now we need to point our fields using @serializable decorator:

class Square implements ISerializable {
    @serializable()
    public width: number
 
    @serializable()
    public height: number
 
    constructor(
        width: number,
        height: number
    ) {
        this.width = width
        this.height = height
    }
}

@serializable decorator is pretty simple:

serializable(name?: string, arrayType?: Function)

name - is the name of your property, you can leave it empty & system will automatically detect it, if property name starts with "_" system will replace it with empty string ("_name" -> "name")

arrayType - since typescript cant provide us type of array items you should pass it yourself, for example if you have property ids: number[] - @serializable call should look like this:

@serializable('ids', Number)
public ids: number[]

Now we can serialize and deserialize our model, there is a BaseSerializer class which implements common logic and we have JSONSerializer which extends from BaseSerializer.

There are two methods that we need:

public static serialize(entity: ISerializable): Object
 
public static deserialize<T extends ISerializable>(
    serializableType: ISerializable & Function,
    raw: { [key: string]: any }
)

First one gets your model instance and serializes in, second one deserializes it:

JSONSerializer.serialize(new Square(10, 10))
// { "width": 10, "height": 10 }

JSONSerializer.deserialize(Square, { "width": 10, "height": 10 })
// Square { width: 10, height: 10 }

deserialize method also does type checking and existing checking.

Logger

There is a BaseLogger interface that specifies basic logging capabilities:

export default interface BaseLogger {
    log(prefix: string, data?: any, recursiveDepth?: boolean): void
 
    warn(prefix: string, data?: any, recursiveDepth?: boolean): void
 
    error(prefix: string, data?: any, recursiveDepth?: boolean): void
}

You can implement your own logger using this interface or you can use ConsoleLogger which implements BaseLogger and writes logs to stdout & stderr

Handlers

There are two ways to write handlers at this moment. The first option is to extend BaseRequestHandler:

export default class TestHandler extends BaseRequestHandler {
    // here you handle request and return response
    // it's better to specify concrete types of request & response
    public async handle(request: TestRequest): Promise<SuccessResponse|ErrorResponse> {
        return new SuccessResponse()
    }
 
    // here you must return true if you support request 
    public supports(request: Request): boolean {
        return request instanceof TestRequest
    }
}

Because BaseRequestHandler is on great for handling multiply request - there is an subclass of it called MultiRequestHandler. You can use MultiRequestHandler like this:

 
export default class UsersHandler extends MultiRequestHandler {
    // here you pass your request class to handles decorator
    @handles(GetUserRequest)
    // all GetUserRequest`s will be passed to this method at this moment
    public async handleGetUser(request: GetUserRequest): Promise<ASResponse> {
        
    }
 
    @handles(SaveUserRequest)
    public async handleSaveUser(request: SaveUserRequest): Promise<ASResponse> {
    
    }
}

All requests are subclasses of ASRequest and all responses are subclasses of ASResponse. There are two already implemented responses:

• ASSuccessResponse
• ASErrorResponse

Cache

There is a BaseCache abstract class which specifies basic caching capabilities:

export abstract class BaseCache<K, V> {
    public abstract async cache(key: K, value: V|null, ttl?: number): Promise<void>
 
    public abstract async get(key: K): Promise<V|undefined>
 
    public abstract async getTTL(key: K): Promise<number>
 
    public abstract async del(key: K): Promise<number>
 
    public abstract async setnx(key: K, value: V): Promise<number>
 
    public abstract async getset(key: K, value: V): Promise<V>
 
    public abstract async expire(key: K, ttl: number): Promise<V>
 
    public abstract async keys(key: string): Promise<V[]>
 
    public abstract async exists(key: K): Promise<boolean>
}

You can implement your own cache using this interface or you can use MemoryCache which implements BaseCache and stores data in memory (not all methods are implemented in MemoryCache). BaseCache is not used in system, but it's may be useful in your project.

API Server

Main logic for API Server is implemented at AirshipAPIServer, implementation is pretty simple: it just wait's for requests from RequestsProvider, passes them to BaseRequestHandler and returns responses back to RequestsProvider.

To create server you need to pass config object to AirshipAPIServer constructor, config object must mach this interface:

export interface AirshipAPIServerConfig {
    // Your requests handler
    requestsHandler: BaseRequestHandler,
    // Your RequestsProvider e.g. HttpRequestsProvider
    requestsProvider: RequestsProvider,
    // optional statistics counter
    statisticsCounter?: BaseStatisticsCounter,
    // optional logger
    logger?: BaseLogger
}

Because AirshipAPIServer uses just one request handler it's expected that handler is capable of handling all types of request of your system. There is a subclass of BaseRequestHandler called RequestHandlerManager for that purposes. RequestHandlerManager receives array of all of your handlers and passes each request to handler that supports it.

Requests provider

Request provider is a component thet provides requests to AirshipAPIServer. There is a RequestsProvider class:

export abstract class RequestsProvider {
    public abstract getRequests(
        callback: (
            request: ASRequest,
            answerRequest: (response: ASResponse) => void
        ) => void
    ): void
}

getRequests method is called by AirshipAPIServer to subscribe to requests, response for request is returned by AirshipAPIServer to answerRequest function.

At this moment we have requests provider for http: HttpRequestsProvider, it uses dietjs module for networking. Usage is pretty simple:

new HttpRequestsProvider(
        logger, // instance of BaseLogger
        7000,   // port
 
        // list of your requests
        GetUserRequest,
        SaveUserRequest
    )

Feel free to create more requests providers!

Statistics

If you pass subclass of BaseStatisticsCounter to AirshipAPIServer server will call countRequestHit when request comes and doneRequest when request handled. There is a subclass called LocalStatisticsCounter which prints stats using your logger:

let statsCounter = new LocalStatisticsCounter(
    logger,
    false, // silence flag, handy for debugging
    1000 * 60 // logFrequency in ms, default is 5000 
)

Generating api scheme

What is api scheme? It's a json file which describes your server models, responses and requests. Api scheme is used to generate client adk & docs. To generate scheme first you need to create config file which implements ApiServerConfig interface:

import {AirshipAPIServerConfig} from "airship-server"
 
const config: ApiServerConfig = {
    endpoints: [
        [TestRequest, TestResponse],
        [GetUserRequest, GetUserResponse]
    ]
}
 
export default config

endpoints is just and array of request response pairs.

After you done with that you can use aschemegen tool to generate scheme:

node_modules/.bin/aschemegen  --o=/Users/altox/Desktop/test-server/scheme  --c=/Users/altox/Desktop/test-server/build/config.js

o argument is the absolute path where scheme will be saved c argument is the absolute path of compiled config file

Generating client SDK

To generate client SDK you just need to run asdkgen and pass to it path to your api schemes and output path:

node_modules/.bin/asdkgen --s=/Users/altox/Desktop/test-server/scheme --o=/Users/altox/Desktop/test-server/sdk

SDK is fully statically typed and written in TypeScript, so you can use it in your TS projects and you can compile it and use in your JS projects. SDK uses fetch, so you might need some polyfill.

Generating docs

Generating docs is just like generating SDK, but you want to use asdocgen:

node_modules/.bin/asdocgen --s=/Users/altox/Desktop/test-server/scheme --o=/Users/altox/Desktop/test-server/docs