(Unofficial) VoxEngine SDK

This repository contains simple SDK to ease development of VoxImplant scenarios. It has:

• Promise-based HTTP client
• Promise-based REST client
• SLF4J-alike logger
• Some concurrent primitives that may be required during development (timeout, delay, task queue, completable promise)

It may be installed via classic npm call

npm i @ama-team/voxengine-sdk --save

And required as any other package:

var SDK = require('@ama-team/voxengine-sdk')

SLF4J-alike Logger

This library wraps standard Logger and adds support for logger names, log levels and log message parameters (substitutions):

var Slf4j = SDK.Logger.Slf4j
var logger = Slf4j.create('logger.name')

// ...

var user = 'Pavel'
var elapsed = 12.345
var metadata = {roles: ['lead'], extras: []}
    
call.addEventListener(CallEvents.Connected, function () {
  logger.info('{} has responded in {} seconds (meta: {})', user, elapsed, metadata)
  // [INFO] logger.name: Pavel has responded in 12.345 seconds (meta: {"roles": ["lead"], "extras": []})
});

This logger provides .trace(), .debug(), .info(), .notice(), .warn(), .error() and .log(Logger.Level.*, pattern, substitutions...) methods to print things.

Substitutions

This logger uses following algorithm for injecting provided arguments:

• While another placeholder is available, replace it with short representation of an argument
• When there are no more placeholders, jump to new line and print long argument representation

Most of the types, there is no difference between two representations. However,

• Custom .toString() implementation will be used in short representations
• If not provided with custom .toString(), errors will be represented as <$.name: $.message> in short mode
• Objects are represented using JSON in long representations
• Long error representations are basically concatenation of short representation with stack

logger.error('{} was caused by:', new Error('alpha'), new Error('beta'))
// ERROR logger.name: <Error: alpha> was caused by:
// <Error: beta>
// Stack:
//   ...

So in order to see stack traces you need to have at most as many placeholders as there are arguments before error.

MDC

There are .attach(key, value), .detach(key), .attachAll(object) and .detachAll() methods for functionality known as Mapped Diagnostic Context in Logback:

logger.attach('id', '123')
logger.info('Sending custom event')
// [INFO] logger.name [id=123]: Sending custom event 

This may help you if you have similar logging output but need to distinguish components one from another.

Configuring

Every logger may be created with particular level and writer, as well as swap them in runtime:

// please note that Logger symbol (last parameter) comes from VoxEngine, not from SDK
var logger = Slf4j.create('logger.name', Slf4j.Level.Info, Logger)
logger.setLevel(Slf4j.Level.Notice)
logger.setLevel('warn')
logger.setWriter(Logger)

This functionality is also exported via Slf4j static methods:

Slf4j.setLevel('logger.name', 'info')
Slf4j.setWriter('logger.name', Logger)

so you can control loggers buried deep in other components.

Logger level and writer are looked up hierarchically: if no level / writer is found for current name, search continues for parent one. Names are separated using dots, so logger.name will use logger.name, logger, (root) lookup chain.

Basic HTTP Client

This client is a simple wrapper around Net.httpRequestAsync primitive

var SDK = require('@ama-team/voxengine-sdk')
var options = {
  url: 'http://my.backend.com',
  retries: 9,
  throwOnServerError: true
}
var client = new SDK.Http.Basic(options);

var call = client
  .get('/magic-phone-number.txt')
  .then(function(response) {
    return VoxEngine.callPSTN(response.payload);
  });

Basic client provides you with following methods:

• get(url, [query, headers, timeout])
• head(url, [query, headers, timeout])
• post(url, [payload, headers, timeout])
• put(url, [payload, headers, timeout])
• patch(url, [payload, headers, timeout])
• delete(url, [payload, headers, timeout])
• request(method, url, [query, payload, headers, timeout])

with following rules:

• Query is an object where every key is associated with a string value or array of string values: {page: '1', filters: ['active', 'fresh']}.
• The same applies to headers object.
• Payload may be either a string or falsey value.
• URL is built by simple concatenation of options.url and url method argument, so it would be http://my.backend.com/magic-phone-number.txt in the example. In case you need to use same client to talk to different hosts, just don't specify url in options - it would be set to empty string.
• Method is a string and can be set using SDK.Http.Method enum.

Every method returns a promise that either resolves with response or rejects with SDK.Http.NetworkException, SDK.Http.HttpException, one of their children or SDK.Http.InvalidConfigurationException. In case of reject, received exception should have .name, .message, .code, .request and sometimes .response fields (except for InvalidConfigurationException).

Whether an error will be thrown and how many retries will be made is defined in client settings passed as first constructor argument:

var settings = {
    retryOnNetworkError: true,
    throwOnServerError: false,
    retryOnServerError: true,
    throwOnClientError: false,
    retryOnClientError: false,
    // NotFound is a special case for 404 response code
    throwOnNotFound: false,
    retryOnNotFound: false,
    // VoxImplant is capable only of emitting GET and POST requests,
    // but this may be overcome by providing real method in
    // additional header
    methodOverrideHeader: 'X-HTTP-Method-Override',
    // those will be used on every request unless headers 
    // specified with request override these 
    headers: {},
    // maximum amount of request retries
    retries: 4,
    // default timeout in milliseconds
    timeout: 1000,
    // alternative logger options
    logger: {
      level: SDK.Logger.Level.Info,
      name: 'custom-name',
      // mapping diagnostic context as described above
      mdc: {
        entity: 'call#123'
      },
      instance: new MyCustomLogger()
    }
};

You can tune client as you want to throw exceptions or return responses on certain outcomes. Network errors always result in exception, however, you may enforce several retries to be made.

REST client

REST client is a wrapper around basic HTTP client that operates over entities rather than HTTP requests/responses. It adds automatic serialization/deserialization support, provides exists, get, create, set, modify and delete methods and adds following rules:

• Any client error results in corresponding exception
• Any server error also results in exception, but retries for specified amount of times
• .exists() method is a wrapper around HEAD-request and returns boolean in promise, treating 404 as false and any 2xx as true
• .get() method is a wrapper around GET-request and returns null on 404
• Non-safe data-changing methods (all others) treat 404 as an error and trigger http.NotFoundException
• There are fallback methods .request() and .execute() in case you have some logic depending on response codes.

The signatures are:

• get(resource, [query, headers, timeout])
• exists(resource, [query, headers, timeout])
• create(resource, [payload, headers, query, timeout])
• set(resource, [payload, headers, query, timeout]) : PUT
• modify(resource, [payload, headers, query, timeout]) : PATCH
• delete(resource, [payload, headers, query, timeout])

So you may talk to your backend like that:

var rest = new SDK.Http.Rest()
var user = rest
  .get('/user', {phone: number, size: 1})
  .then(function (response) {
    return response ? response.content[0] : rest.create('/user', {phone: number});
  });

REST client is configured very similarly to HTTP client:

var options = {
  // this will be prepended to all routes you pass into client
  url: 'http://backend/api/v1',
  // you will certainly need to set this if you're going to use anything but .get/.create
  methodOverrideHeader: 'X-HTTP-Method-Override',
  // that's pretty much the default
  retries: 4,
  // this is set by default as well
  serializer: {
    serialize: JSON.stringify,
    deserialize: JSON.parse
  },
  // again, a set of headers that will always be present in 
  // requests (unless overriden in particular request)
  headers: {
    'Content-Type': 'application/json'
  },
  // default timeout
  timeout: 1000,
  // same logger options
  logger: {}
}
var client = new SDK.Http.Rest(options);

Future

This SDK also provides an externally-completable promise called Future for brevity. It has the same interface as standard promise, but also exposes #resolve() and #reject() instance methods:

var Future = SDK.Concurrent.Future

var race = Future.race([new Future(), new Future()])
race.resolve('hijack')
race.reject('will be ignored')
var future = new Future(function (resolve, reject) {
  Math.random() > 0.5 ? resolve() : reject()
})

Future passes all Promises/A+ compliance tests.

timeout

timeout function allows you to wrap promise with another one which will auto-reject after specified time. Most of the time you won't need this, because VoxEngine automatically times out long dials and requests, but sometimes you will need more strict time bounds or set a timeout on non-standard resource:

var perfectMatch = api.getPerfectMatch(call)
SDK.Concurrent.timeout(perfectMatch, 10000)
  .then(null, function (error) {
    logger.info('It looks like api can\'t find match quickly: ', error)
    logger.info('Falling back to standard match')
    return api.getStandardMatch(call)
  })

The TimeoutException used to reject such promise can be located as SDK.Concurrent.TimeoutException. Resulting promise also has #cancel(silent) method that allows to clear internal timeout id. Silent parameter defines whether it it will be cancelled silently (pretending that there never was a timeout) or with CancellationException.

delay

delay function will wrap your callback and invoke it in future, or just create a timer promise if callback is omitted:

SDK.Concurrent.delay(10000, function () {
  call.say('Your answering time has expired')
})

SDK.Concurrent.delay(10000).then(function () {
  call.say('Your answering time has expired')
})

You can use it with Promise.race() to create some basic time-bounded scenarios without explicit rejection:

var toneDetection = new Promise(function (resolve) {
  call.addEventListener(CallEvents.ToneDetected, resolve)
})
var timeBound = SDK.Concurrent.delay(10000)
Promise.race(toneDetection, timeBound).then(function() {
  call.say('I\'m sorry, Dave. I\'m afraid i can\'t do that.')
})

Delay function result has #cancel(silent) method as well that will either instantly resolve delay or reject promise with CancellationException depending on silent parameter.

throttle

The last helper function prevents promise from resolving too fast. It wraps promise as well and ensures that it will resolve not earlier than specified timeout allows:

logger.info('Emulating hard work so our clients would think we\'re smart')
var balance = rest.get('/users/' + id + '/balance')
var throttled = SDK.Concurrent.throttle(balance, 10000)

Throttle function result provides similar #cancel(silent) method.

TaskQueue

When you're using tons of async code, quite often you need to enforce some ordering on processing. TaskQueue is created to run tasks sequentially - for example, it is often necessary for HTTP requests to come in order:

var queue = SDK.Concurrent.TaskQueue.started({name: 'Call ' + id + ' event stream'})
queue.push(function () {
  return rest.create('/calls/' + id, {number: number})
})
queue.push(function() {
  return rest.modify('/users/' + userId + '/balance', {amount: -33.3})
}, {name: 'User balance withdrawal', timeout: 1000})
queue
  .close()
  .then(function() {
    logger.info('Everything has been finished!')
  })

Queue exposes #start() and #pause() methods to start and pause processing, #push() for adding new tasks, #close() and #terminate() for normal (wait for all tasks) and emergency (discard waiting tasks) queue finalization. #push will return a promise that will be resolved once task has been run, while closing methods will return promise which will resolve once queue will become empty.

How do i require this stuff in VoxEngine?

If you don't already know, VoxImplant scripts are basically just a single file which is executed on external trigger. This limits your ability to require anything, including this particular library, but you can always bundle things down with a build tool of your choice. I personally use Webpack, but was recommended of Browserify as well, and, of course good old Grunt and Gulp should work too.

Don't forget that VoxImplant has scenario size limit of 256 kb, so don't require anything at sight, don't forget to minify things you require and don't forget to strip comments off.

ES5 & standards

This repository is developed as ES5 module because it is implied that built code would be injected directly in VoxImplant scenarios, which, in turn, do not support ES6 yet, and transpiling would dramatically harden bug hunt.

This repository is developed using standardjs, sorry if you have feelings for semicolons.

Testing

This package is using Mocha with Chai for test running, Istanbul for recording coverage metrics and Allure framework for reporting. To run tests, simply invoke jake:

npx jake test

If you want full-blown feedback, use npx jake test:with-report to generate Allure report (don't forget to install allure-commandline before) and coverage report, which will be placed in tmp/report directory.

Anything else?

We have @ama-team/voxengine-definitions package that helps with autocompletion and scenario framework which aims at more sophisticated approach for VoxImplant scenario development. Also there is a chance that script publishing tool will be finally developed at the moment you're reading this, so it worth to check it.

There is no particular roadmap for this project, so you may use GitHub issues to propose any ideas you got. However, there is not much time that could be devoted to this project.

Dev branch state / incubating