Using Service Workers

Esta tradução está incompleta. Ajude atraduzir este artigo.

This is an experimental technology
Because this technology's specification has not stabilized, check the compatibility table for usage in various browsers. Also note that the syntax and behavior of an experimental technology is subject to change in future versions of browsers as the specification changes.

Esse artigo contém informações de como começar com service workes, incluindo a arquitura básica, registro de um service worker, o processo de instalação e ativação de um novo service worker, controle de cache e respostas customizadas, tudo isso no contexto de um simples app com a funcionalidade offline.

A premissa do Service Worker

Um problema predominante que os usuários sofreram durante anos foi a perda de conexão. A melhor web app consegue prover uma péssima experiência de usuário se você não conseguir baixa-lá. Houve várias tentativas de criar tecnologias para resolver esse problema, como mostra nossa página Offline, e algumas questões foram resolvidas. Mas o maior problema é que no geral não possui um mecanismo de controle geral para caching de assets e requisições customizadas para internet.

A tentativa anterior — AppCache — parecia ser uma boa ideia porque permitia que você especificasse os assets para cache de maneira mais fácil. Porém, fazia várias hipótese sobre o que você estava tentando fazer e depois quebrava horrivelmente quando seu app não seguia exatamente as hipóteses. Leia Jake Archibald's Application Cache is a Douchebag para mais detalhes.

Nota: Recentemente o Firefox 44, quando AppCache está sendo usado para fornecer suporte offline para uma página uma mensagem de aviso é mostrada no console para os desenvolvedores usarem Service workers em vez do AppCache (bug 1204581.)

Service workers deve finalmente consertar essas questões. A sintaxe do service worker é mais complexa do que o AppCache, mas equilibra pois você pode usar Javascript para controlar as implicações do comportamento do AppCache com um bom grau de qualidade, permitindo que você controle esse problema e muitos outros. Usando Service Worker você pode de maneira mais fácil configurar um app para usar assets em cache primeiro, assim provendo uma experiência padronizada mesmo que esteja offline, antes de pegar mais dados da internet (mais conhecido como Offline First). Isso já está disponível para apps nativos, que é uma das principais razões de escolha sobre web apps.

Configuração para usar service workers

Várias features do service workers são habilitadas por padrão em versões mais novas dos browsers. Se você por acaso encontrar um código demo que não esta funcionando na sua versão instalada, você precisará habilitar essa preferência:

  • Firefox Nightly: Vá em sobre: config e mude dom.serviceWorkers.enabled para true; reinicie o browser.
  • Chrome Canary: Vá em chrome://flags e mude para experimental-web-platform-features; reinicie o browser(note que algumas features do Chrome vem habilitada por padrão.)
  • Opera:  Vá em opera://flags e habilite Support for ServiceWorker; reinicie o browser.

Você também precisa fornecer esse código via HTTPS — Service workers possui uma restrição para rodar apenas via HTTPS por razões de segurança. GitHub serve como um bom host para os experimentos, pois suporta HTTPS. A fim de facilitar o desenvolvimento local, localhost é considerado uma origem segura para os browsers também. 

Arquitetura Básica

Com service workers, os seguintes passos geralmente são observados como configuração básica:

  1. A URL do service é worker é verificada é registrada via serviceWorkerContainer.register().
  2. Em caso de sucesso, o service worker é exectuado no ServiceWorkerGlobalScope; isto é basicamente um tipo especial para o contexto do worker, que está sendo executado fora thread principal, sem acesso ao DOM.
  3. O service worker agora está pronto para processar eventos.
  4. A tentativa de instalação do worker acontece quando as páginas controladas pelo service worker são acessadas. O evento de instalação sempre é o primeiro a ser enviado ao service worker (este pode ser o processo usado para iniciar o processo de população do IndexedDB, e o cache de assets do site). Este é o mesmo processo de instalação native ou do app do Firefox OS — fazendo tudo ficar disponível offline.
  5. Quando o manipulador oninstall estiver completo, considera-se que o service worker está instalado.
  6. O próximo é a ativação. Quando um service worker é instalado, ele recebe um evento de ativação. O uso primário do onactivate é para limpeza de recursos usado em versões anterioires do script do Service worker.
  7. O Service worker agora vai controlar páginas, mas apenas aquelas abertas após o register() tiver sido bem-sucedido. ex.: um documento começa a vida com ou sem Service worker e mantém isso para seu tempo de vida. Então documentos tem que ser recarregados para ser controlado.

O gráfico abaixo mostra um resumo de eventos disponíveis do service worker:

install, activate, message, fetch, sync, push

Promises

Promises são ótimos mecanismos para rodar operações async, com sucesso dependendo de outra. Este é o principal para o modo de trabalho dos service workers.

Promises pode fazer ótimas coisas, mas por agora, tudo que precisa saber é se alguma coisa retorna uma promise, você pode inserir .then() ao final e incluir dentro dele callbacks para sucesso, falha, etc., ou poder inserir .catch() ao final se quiser incluir um callback para falha.

Vamos comparar a estrutura de um callback tradicional síncrono para uma promise assíncrona equivalente.

sync

try {
  var value = myFunction();
  console.log(value);
} catch(err) {
  console.log(err);
}

async

myFunction().then(function(value) {
  console.log(value);
}).catch(function(err) {
  console.log(err);
});

In the first example, we have to wait for myFunction() to run and return value before any more of the code can execute. In the second example, myFunction() returns a promise for value, then the rest of the code can carry on running. When the promise resolves, the code inside then will be run, asynchronously.

Now for a real example — what if we wanted to load images dynamically, but we wanted to make sure the images were loaded before we tried to display it? This is a standard thing to want to do, but it can be a bit of a pain. We can use .onload to only display the image after it’s loaded, but what about events that start happening before we start listening to them? We could try to work around this using .complete, but it’s still not foolproof, and what about multiple images? And, ummm, it’s still synchronous, so blocks the main thread.

Instead, we could build our own promise to handle this kind of case. (See our Promises test example for the source code, or look at it running live.)

Note: A real service worker implementation would use caching and onfetch rather than the deprecated XMLHttpRequest API. Those features are not used here so that you can focus on understanding Promises.

function imgLoad(url) {
  return new Promise(function(resolve, reject) {      
    var request = new XMLHttpRequest();
    request.open('GET', url);
    request.responseType = 'blob';

    request.onload = function() {
      if (request.status == 200) {
        resolve(request.response);
      } else {
        reject(Error('Image didn\'t load successfully; error code:' + request.statusText));
      }
    };

    request.onerror = function() {
      reject(Error('There was a network error.'));
    };

    request.send();
  });
}

We return a new promise using the Promise() constructor, which takes as an argument a callback function with resolve and reject parameters. Somewhere in the function, we need to define what happens for the promise to resolve successfully or be rejected — in this case return a 200 OK status or not — and then call resolve on success, or reject on failure. The rest of the contents of this function is fairly standard XHR stuff, so we won’t worry about that for now.

When we come to call the imgLoad() function, we call it with the url to the image we want to load, as we might expect, but the rest of the code is a little different:

var body = document.querySelector('body');
var myImage = new Image();

imgLoad('myLittleVader.jpg').then(function(response) {
  var imageURL = window.URL.createObjectURL(response);
  myImage.src = imageURL;
  body.appendChild(myImage);
}, function(Error) {
  console.log(Error);
});

On to the end of the function call, we chain the promise then() method, which contains two functions — the first one is executed when the promise successfully resolves, and the second is called when the promise is rejected. In the resolved case, we display the image inside myImage and append it to the body (it’s argument is the request.response contained inside the promise’s resolve method); in the rejected case we return an error to the console.

This all happens asynchronously.

Note: You can also chain promise calls together, for example:
myPromise().then(success, failure).then(success).catch(failure);

Note: You can find a lot more out about promises by reading Jake Archibald’s excellent JavaScript Promises: there and back again.

Service workers demo

To demonstrate just the very basics of registering and installing a service worker, we have created a simple demo called sw-test, which is a simple Star wars Lego image gallery. It uses a promise-powered function to read image data from a JSON object and load the images using Ajax, before displaying the images in a line down the page. We’ve kept things static and simple for now. It also registers, installs, and activates a service worker, and when more of the spec is supported by browsers it will cache all the files required so it will work offline!




You can see the source code on GitHub, and view the example live. The one bit we’ll call out here is the promise (see app.js lines 22-47), which is a modified version of what you read about above, in the Promises test demo. It is different in the following ways:

  1. In the original, we only passed in a URL to an image we wanted to load. In this version, we pass in a JSON fragment containing all the data for a single image (see what they look like in image-list.js). This is because all the data for each promise resolve has to be passed in with the promise, as it is asynchronous. If you just passed in the url, and then tried to access the other items in the JSON separately when the for() loop is being iterated through later on, it wouldn’t work, as the promise wouldn’t resolve at the same time as the iterations are being done (that is a synchronous process.)
  2. We actually resolve the promise with an array, as we want to make the loaded image blob available to the resolving function later on in the code, but also the image name, credits and alt text (see app.js lines 31-34). Promises will only resolve with a single argument, so if you want to resolve with multiple values, you need to use an array/object.
  3. To access the resolved promise values, we then access this function as you’d then expect (see app.js lines 60-64). This may seem a bit odd at first, but this is the way promises work.

Enter service workers

Now let’s get on to service workers!

Registering your worker

The first block of code in our app’s JavaScript file — app.js — is as follows. This is our entry point into using service workers.

if ('serviceWorker' in navigator) {
  navigator.serviceWorker.register('/sw-test/sw.js', {scope: '/sw-test/'})
  .then(function(reg) {
    // registration worked
    console.log('Registration succeeded. Scope is ' + reg.scope);
  }).catch(function(error) {
    // registration failed
    console.log('Registration failed with ' + error);
  });
}
  1. The outer block performs a feature detection test to make sure service workers are supported before trying to register one.
  2. Next, we use the ServiceWorkerContainer.register() function to register the service worker for this site, which is just a JavaScript file residing inside our app (note this is the file's URL relative to the origin, not the JS file that references it.)
  3. The scope parameter is optional, and can be used to specify the subset of your content that you want the service worker to control. In this case, we have specified '/sw-test/', which means all content under the app's origin. If you leave it out, it will default to this value anyway, but we specified it here for illustration purposes.
  4. The .then() promise function is used to chain a success case onto our promise structure.  When the promise resolves successfully, the code inside it executes.
  5. Finally, we chain a .catch() function onto the end that will run if the promise is rejected.

This registers a service worker, which runs in a worker context, and therefore has no DOM access. You then run code in the service worker outside of your normal pages to control their loading.

A single service worker can control many pages. Each time a page within your scope is loaded, the service worker is installed against that page and operates on it. Bear in mind therefore that you need to be careful with global variables in the service worker script: each page doesn’t get its own unique worker.

Note: Your service worker functions like a proxy server, allowing you to modify requests and responses, replace them with items from its own cache, and more.

Note: One great thing about service workers is that if you use feature detection like we’ve shown above, browsers that don’t support service workers can just use your app online in the normal expected fashion. Furthermore, if you use AppCache and SW on a page, browsers that don’t support SW but do support AppCache will use that, and browsers that support both will ignore the AppCache and let SW take over.

Why is my service worker failing to register?

This could be for the following reasons:

  1. You are not running your application through HTTPS.
  2. The path to your service worker file is not written correctly — it needs to be written relative to the origin, not your app’s root directory. In our example, the worker is at https://mdn.github.io/sw-test/sw.js, and the app’s root is https://mdn.github.io/sw-test/. But the path needs to be written as /sw-test/sw.js, not /sw.js.
  3. The service worker being pointed to is on a different origin to that of your app. This is also not allowed.

Also note:

  • The service worker will only catch requests from clients under the service worker's scope.
  • The max scope for a service worker is the location of the worker.
  • If your server worker is active on a client being served with the Service-Worker-Allowed header, you can specify a list of max scopes for that worker.
  • In Firefox, Service Worker APIs are hidden and cannot be used when the user is in private browsing mode.

Install and activate: populating your cache

After your service worker is registered, the browser will attempt to install then activate the service worker for your page/site.

The install event is fired when an install is successfully completed. The install event is generally used to populate your browser’s offline caching capabilities with the assets you need to run your app offline. To do this, we use Service Worker’s brand new storage API — cache — a global on the service worker that allows us to store assets delivered by responses, and keyed by their requests. This API works in a similar way to the browser’s standard cache, but it is specific to your domain. It persists until you tell it not to — again, you have full control.

Note: The Cache API is not supported in every browser. (See the Browser support section for more information.) If you want to use this now, you could consider using a polyfill like the one available in Google's Topeka demo, or perhaps store your assets in IndexedDB.

Let’s start this section by looking at a code sample — this is the first block you’ll find in our service worker:

this.addEventListener('install', function(event) {
  event.waitUntil(
    caches.open('v1').then(function(cache) {
      return cache.addAll([
        '/sw-test/',
        '/sw-test/index.html',
        '/sw-test/style.css',
        '/sw-test/app.js',
        '/sw-test/image-list.js',
        '/sw-test/star-wars-logo.jpg',
        '/sw-test/gallery/',
        '/sw-test/gallery/bountyHunters.jpg',
        '/sw-test/gallery/myLittleVader.jpg',
        '/sw-test/gallery/snowTroopers.jpg'
      ]);
    })
  );
});
  1. Here we add an install event listener to the service worker (hence this), and then chain a ExtendableEvent.waitUntil() method onto the event — this ensures that the service worker will not install until the code inside waitUntil() has successfully occurred.
  2. Inside waitUntil() we use the caches.open() method to create a new cache called v1, which will be version 1 of our site resources cache. This returns a promise for a created cache; once resolved, we then call a function that calls addAll() on the created cache, which for its parameter takes an array of origin-relative URLs to all the resources you want to cache.
  3. If the promise is rejected, the install fails, and the worker won’t do anything. This is ok, as you can fix your code and then try again the next time registration occurs.
  4. After a successful installation, the service worker activates. This doesn’t have much of a distinct use the first time your service worker is installed/activated, but it means more when the service worker is updated (see the Updating your service worker section later on.)

Note: localStorage works in a similar way to service worker cache, but it is synchronous, so not allowed in service workers.

Note: IndexedDB can be used inside a service worker for data storage if you require it.

Custom responses to requests

Now you’ve got your site assets cached, you need to tell service workers to do something with the cached content. This is easily done with the fetch event.

A fetch event fires every time any resource controlled by a service worker is fetched, which includes the documents inside the specified scope, and any resources referenced in those documents (for example if index.html makes a cross origin request to embed an image, that still goes through its service worker.)

You can attach a fetch event listener to the service worker, then call the respondWith() method on the event to hijack our HTTP responses and update them with your own magic.

this.addEventListener('fetch', function(event) {
  event.respondWith(
    // magic goes here
  );
});

We could start by simply responding with the resource whose url matches that of the network request, in each case:

this.addEventListener('fetch', function(event) {
  event.respondWith(
    caches.match(event.request)
  );
});

caches.match(event.request) allows us to match each resource requested from the network with the equivalent resource available in the cache, if there is a matching one available. The matching is done via url and vary headers, just like with normal HTTP requests.

Let’s look at a few other options we have when defining our magic (see our Fetch API documentation for more information about Request and Response objects.)

  1. The Response() constructor allows you to create a custom response. In this case, we are just returning a simple text string:

    new Response('Hello from your friendly neighbourhood service worker!');
  2. This more complex Response below shows that you can optionally pass a set of headers in with your response, emulating standard HTTP response headers. Here we are just telling the browser what the content type of our synthetic response is:

    new Response('<p>Hello from your friendly neighbourhood service worker!</p>', {
      headers: { 'Content-Type': 'text/html' }
    });
  3. If a match wasn’t found in the cache, you could tell the browser to simply fetch the default network request for that resource, to get the new resource from the network if it is available:

    fetch(event.request);
  4. If a match wasn’t found in the cache, and the network isn’t available, you could just match the request with some kind of default fallback page as a response using match(), like this:

    caches.match('/fallback.html');
  5. You can retrieve a lot of information about each request by calling parameters of the Request object returned by the FetchEvent:

    event.request.url
    event.request.method
    event.request.headers
    event.request.body

Recovering failed requests

So caches.match(event.request) is great when there is a match in the service worker cache, but what about cases when there isn’t a match? If we didn’t provide any kind of failure handling, our promise would reject and we would just come up against a network error when a match isn’t found.

Fortunately service workers’ promise-based structure makes it trivial to provide further options towards success. We could do this:

this.addEventListener('fetch', function(event) {
  event.respondWith(
    caches.match(event.request).then(function(response) {
      return response || fetch(event.request);
    })
  );
});

If the resources isn't in the cache, it is requested from the network.

If we were being really clever, we would not only request the resource from the network; we would also save it into the cache so that later requests for that resource could be retrieved offline too! This would mean that if extra images were added to the Star Wars gallery, our app could automatically grab them and cache them. The following would do the trick:

this.addEventListener('fetch', function(event) {
  event.respondWith(
    caches.match(event.request).then(function(resp) {
      return resp || fetch(event.request).then(function(response) {
        return caches.open('v1').then(function(cache) {
          cache.put(event.request, response.clone());
          return response;
        });  
      });
    })
  );
});

Here we return the default network request with return fetch(event.request), which returns a promise. When this promise is resolved, we respond by running a function that grabs our cache using caches.open('v1'); this also returns a promise. When that promise resolves, cache.put() is used to add the resource to the cache. The resource is grabbed from event.request, and the response is then cloned with response.clone() and added to the cache. The clone is put in the cache, and the original response is returned to the browser to be given to the page that called it.

Cloning the response is necessary because request and response streams can only be read once.  In order to return the response to the browser and put it in the cache we have to clone it. So the original gets returned to the browser and the clone gets sent to the cache.  They are each read once.

The only trouble we have now is that if the request doesn’t match anything in the cache, and the network is not available, our request will still fail. Let’s provide a default fallback so that whatever happens, the user will at least get something:

this.addEventListener('fetch', function(event) {
  event.respondWith(
    caches.match(event.request).then(function(resp) {
      return resp || fetch(event.request).then(function(response) {
        caches.open('v1').then(function(cache) {
          cache.put(event.request, response.clone());
        });
        return response;
      });
    }).catch(function() {
      return caches.match('/sw-test/gallery/myLittleVader.jpg');
    })
  );
});

We have opted for this fallback image because the only updates that are likely to fail are new images, as everything else is depended on for installation in the install event listener we saw earlier.

Updating your service worker

If your service worker has previously been installed, but then a new version of the worker is available on refresh or page load, the new version is installed in the background, but not yet activated. It is only activated when there are no longer any pages loaded that are still using the old service worker. As soon as there are no more such pages still loaded, the new service worker activates.

You’ll want to update your install event listener in the new service worker to something like this (notice the new version number):

this.addEventListener('install', function(event) {
  event.waitUntil(
    caches.open('v2').then(function(cache) {
      return cache.addAll([
        '/sw-test/',
        '/sw-test/index.html',
        '/sw-test/style.css',
        '/sw-test/app.js',
        '/sw-test/image-list.js',
        
        …

        // include other new resources for the new version...
      ]);
    })
  );
});

While this happens, the previous version is still responsible for fetches. The new version is installing in the background. We are calling the new cache v2, so the previous v1 cache isn't disturbed.

When no pages are using the current version, the new worker activates and becomes responsible for fetches.

Deleting old caches

You also get an activate event. This is a generally used to do stuff that would have broken the previous version while it was still running, for example getting rid of old caches. This is also useful for removing data that is no longer needed to avoid filling up too much disk space — each browser has a hard limit on the amount of cache storage that a given service worker can use. The browser does its best to manage disk space, but it may delete the Cache storage for an origin.  The browser will generally delete all of the data for an origin or none of the data for an origin.

Promises passed into waitUntil() will block other events until completion, so you can rest assured that your clean-up operation will have completed by the time you get your first fetch event on the new cache.

this.addEventListener('activate', function(event) {
  var cacheWhitelist = ['v2'];

  event.waitUntil(
    caches.keys().then(function(keyList) {
      return Promise.all(keyList.map(function(key) {
        if (cacheWhitelist.indexOf(key) === -1) {
          return caches.delete(key);
        }
      }));
    })
  );
});

Developer tools

Chrome has chrome://inspect/#service-workers, which shows current service worker activity and storage on a device, and chrome://serviceworker-internals, which shows more detail and allows you to start/stop/debug the worker process. In the future they will have throttling/offline modes to simulate bad or non-existent connections, which will be a really good thing.

Firefox has also started to implement some useful tools related to service workers:

  • You can navigate to about:debugging to see what SWs are registered and update/remove them.
  • When testing you can get around the HTTPS restriction by checking the "Enable Service Workers over HTTP (when toolbox is open)" option in the Firefox Developer Tools settings.

Note: You may serve your app from http://localhost (e.g. using me@localhost:/my/app$ python -m SimpleHTTPServer) for local development. See Security considerations

Specifications

Specification Status Comment
Service Workers Working Draft Initial definition.

Browser compatibility

Feature Chrome Firefox (Gecko) Internet Explorer Opera Safari (WebKit)
Basic support 40.0 33.0 (33.0)[1] Não suportado 24 Não suportado
Feature Android Chrome for Android Firefox Mobile (Gecko) Firefox OS IE Phone Opera Mobile Safari Mobile
Basic support Não suportado 40.0 (Yes) (Yes) Não suportado (Yes) Não suportado

[1] Service workers (and Push) have been disabled in the Firefox 45 Extended Support Release (ESR.)

See also

Etiquetas do documento e colaboradores

 Colaboradores desta página: Lucasktrindade
 Última atualização por: Lucasktrindade,