• Introduction
• The simplest possible use case
  • Additional examples
• Use cases
  • On-demand server-side rendering and caching
  • Single-use server-side rendering as part of a build process
  • Variants
    • Client code
    • Server code
• State transfer from server to client
• More details on server-side rendering code
  • Snapshot<V>
• Example projects
  • CLI based project that uses @angular/material
  • On-demand rendering using express
  • On-demand rendering using koa
• FAQ
• Comments, queries, or rants

The purpose of this library is to allow your application to support server-side rendering of your Angular 4+ applications with minimal code changes and mimimal difficulty. It supports both Angular CLI projects and projects that use custom webpack configurations. It works out of the box with @angular/material with no hot-fixes or workarounds! It also requires zero changes to your application code: you won't have to create separate @NgModules, one for the server-side rendered application and one for the regular client application. You can just take your Angular code as-is and follow the steps below to get server-side rendering working.

There are two ways you can use angular-ssr:

1. If your application is an Angular CLI application with no custom webpack configuration, you can simply install it as a dependency, run a normal ng build, and then invoke ng-render from node_modules/.bin. This will result in several steps being taken:
   • It will use tsconfig.json and some other configuration elements to compile your application to a temporary directory and load the resulting JavaScript code (application code + .ngfactory.js files) into memory.
   • It will query your router configuration and collect all your application routes into a flattened array (eg. /, /foo, /bar)
   • For each of the discovered routes, it will instantiate your application and render that route to a static .html file in dist (or, if you specified an alternate output directory using --output, it will write the files there). It instantiates the application using the existing dist/index.html file that was produced as part of your normal application build as a template. The pre-rendered content will be inserted into that template and written out as a new .html file based on the route: e.g., /foo/index.html.
   • The drawback to this approach is that the content is generated at build time, so if your routes contain some dynamic data that needs to be rendered on the server, you will instead need to write a simple HTTP server using express and koa and do on-demand server-side rendering.
2. If your application has custom webpack configurations and loaders, you probably will not be able to use ng-render. But that's alright. It just means that you will have to build a separate webpack program output: either a NodeJS HTTP server, or a NodeJS application whose sole purpose is to do prerendering. You will follow these rough steps:
   • Install angular-ssr as a dependency: npm install angular-ssr --save
   • If you already have multiple webpack configs (one for server and one for client), then you can skip down to the next section and begin writing code to interface with angular-ssr.
   • Otherwise, you will need to add an additional output to your existing webpack configurations. This can take two forms: either you modify your existing webpack.config.js and just add an additional output, or you create an entirely new webpack-server.config.js which will serve as your SSR webpack configuration. Regardless of how you accomplish it, you will ultimately need to produce two programs from webpack:
     • Your normal client-side JavaScript application
     • An additional server-side application that you will use to do server-side rendering. You have a couple choices here, as well:
       • If you want your application to use a NodeJS application with an HTTP server inside of it that will do on-demand pre-rendering of your application routes, then do that. We can then write a few lines of code to do the actual pre-rendering / caching inside of your route handlers. It doesn't matter if you use koa or express or any other HTTP server you wish to use -- angular-ssr will not integrate directly with the HTTP server anyway. It just exposes a very simple API to get pre-rendered HTML documents, and you can integrate this with your server in whichever way makes the most sense for your application.
       • Alternatively, you can build an application whose sole purpose is to do server-side rendering at build-time. This application will produce some static pre-rendered application content and then exit. This use-case makes sense if your application will not need to do on-demand server-side rendering. Let's say for example you just have an application with a few routes (/a, /b, /c, etc.). In this case, since all routes are known in advance and none of them take any URL parameters, we can just pre-render each route at build time and spit out some .html files.
       • Let's say that your application does need on-demand rendering, though. For example, you are writing a blog application that has URLs like /blog/post/1, /blog/user/3, etc. In this case, you will need to do on-demand server-side rendering. No problem! In this use-case, it makes sense to build a small HTTP server using express or koa and to write a few lines of code to integrate with angular-ssr. Then from inside your server, you can demand render and cache particular routes with arguments like /blog/post/1. I will show you some examples of how to do this below.

If your application was generated by ng new and does not use any custom webpack configuration, then you will be able to use the ng-render CLI tooll to automatically pre-render your application routes into static .html files. It is worth emphasizing that this use case is the easiest, but also the least flexible. If you need on-demand rendering, or if you have custom webpack configurations, then you should skip down to the examples below as they will cover your use-case better than this section.

But, in the event that you do have a simple ng cli application, you can give angular-ssr a whirl just by doing:

npm install angular-ssr --save
ng build
./node_modules/.bin/ng-render

It should spit out some messages like:

[info] Writing rendered route / to /Users/bond/proj/dist/index.html
[info] Writing rendered route /foo to /Users/bond/proj/dist/foo/index.html
[info] Writing rendered route /bar to /Users/bond/proj/dist/bar/index.html

You can then do cd dist and run:

npm install -g http-server
http-server .

Then when you load the application by hitting http://localhost:8080, you should see the pre-rendered document in the initial HTTP response (for each route in your application).

An example application like the one I have just described is available in the examples/cli directory. It also uses @angular/material to prove that Material works with angular-ssr.

Additional examples are available in the Examples section.

Let's get this use-case out of the way first, because I think it is likely to be the most common usage of angular-ssr.

You have an HTTP server application that you build as part of your application using webpack. Your HTTP server is written in TypeScript. (If your HTTP server is written in JavaScript, the library will still work in the same way, but you won't be able to copy-paste the code below.)

When you build your application, you are outputting two targets: your actual Angular client application, and your HTTP server application. We are going to focus on the server application here because there will be zero changes to your application code.

Your actual HTTP server code will look something like the following:

import {ApplicationFromModule, DocumentStore} from 'angular-ssr';

import {join} from 'path';

import {AppModule} from '../src/app/app.module';

const dist = join(process.cwd(), 'dist');

const application = new ApplicationFromModule(AppModule, join(dist, 'index.html'));

// Pre-render all routes that do not take parameters (angular-ssr will discover automatically)
const prerender = async () => {
  const snapshots = await application.prerender();

  return snapshots.subscribe(
    snapshot => {
      app.get(snapshot.uri, (req, res) => res.send(snapshot.renderedDocument));
    })
    .toPromise();
};

prerender();

const documentStore = new DocumentStore(application);

// Demand render and cache all other routes (eg /blog/post/12)
app.get('*', async (req, res) => {
  try {
    const snapshot = await documentStore.load(req.url);
    res.send(snapshot.renderedDocument);
  }
  catch (exception) {
    res.send(application.templateDocument()); // fall back on client-side rendering
  }
});

If your application does not fall into the categories described above (i.e., you do not need on-demand server-side rendering of all URLs), then perhaps your application falls into another category: single-use server-side rendering as part of the application build process.

In this case, your code will look similar to the HTTP server code above, but instead of integrating with express, you will simply use ApplicationRenderer to pre-render all application routes and write them to static .html files, which you can then serve with the HTTP server of your choosing. Again: this case only makes sense if you do not need on-demand rendering of all application routes.

In this case, your code will look something like this:

import {
  ApplicationFromModule,
  ApplicationRenderer,
  HtmlOutput,
} from 'angular-ssr';

import {join} from 'path';

import {AppModule} from '../src/app.module';

const dist = join(process.cwd(), 'dist');

const application = new ApplicationFromModule(ServerModule, join(dist, 'index.html'));

const html = new HtmlOutput(dist);

const renderer = new ApplicationRenderer(application);

renderer.renderTo(html)
  .catch(exception => {
    console.error('Failed to render due to uncaught exception', exception);
  });

Now we arrive at the most complex use case. Here we wish to do prerendering and demand rendering inside a NodeJS HTTP server, but we also wish to render variants of each page. For example, our application may support multiple languages. angular-ssr supports this using a concept called a variant. A variant is essentially a key, a set of unique values, and a transition function which can place the application in the specified state.

To illustrate, let's again use the example of locales / languages. Your application has multiple languages and you want to support server-side rendering for each of them. The first time someone loads your application, we set the current language selection to the value of navigator.language (eg, "en-US"). We set an application cookie using document.cookie so that subsequent loads of the application will include as part of the request the language that the user wishes to view the application in. Assume we have some simple code like this somewhere in the application:

import {Component, Injectable, OnInit} from '@angular/core';

@Component({
  selector: 'app',
  template: `<locale-selector [value]="localeService.locale" (select)="onLocaleChanged($event)"></locale-selector>`
})
export class LocaleSelector implements OnInit {
  constructor(public localeService: LocaleService) {}

  onLocaleChanged(locale: string) {
    this.localeService.locale = locale;
  }
}

import {Injectable} from '@angular/core';

import {Observable, ReplaySubject} from 'rxjs';

@Injectable()
export class LocaleService {
  subject = new ReplaySubject<string>();

  constructor(private cookies: CookieService) {
    this.update(cookies.get<string>('locale') || navigator.language || 'en-US');
  }

  locale(locale?: string): Observable<string> {
    if (locale) {
      this.update(locale);
    }
    return this.subject;
  }

  private update(value: string) {
    this.subject.next(value);

    this.cookies.set('locale', value);
  }
}

export type CookieValue = string | number;

@Injectable()
export class CookieService {
  get map(): Map<string, CookieValue> {
    if (!document.cookie) {
      return new Map<string, string>();
    }

    const components = document.cookie.split(/(;\s?)/g);

    const tuples = components.map(
      (pair: string): [string, CookieValue] => {
        const [key, value] = pair.split(/=/);

        return /^([\d+])$/.test(value)
          ? [key, parseInt(value, 10)];
          : [key, value];
      });

    return new Map<string, CookieValue>(tuples);
  }

  get<T>(key: string): T {
    return this.map.get(key) as any;
  }

  set(key: string, value: CookieValue) {
    this.delete(key);

    document.cookie = `${key}=${value.toString()}; path=/; domain=${location.hostname};`;
  }

  delete(key: string) {
    const criterion: Array<[string, string | number]> = [
      ['expires', 'Thu, 01 Jan 1970 00:00:01 GMT'],
      ['path', '/'],
      ['domain', location.hostname],
      ['max-age', 0]
    ];

    while (criterion.length > 0) {
      const serialized = criterion.map(([k, v]) => `${k}=${v}`).join('; ');

      document.cookie = `${key}=;${serialized ? ' ' + serialized : String()}`.trim();

      criterion.pop();
    }
  }
}

Essentially what this code is doing is setting a cookie in two events:

1. The user loads the application for the first time and there is no cookie, so we set the cookie value to navigator.language, to respect their system locale settings.
2. If the user changes the locale, we update document.cookie with the new locale setting. Then subsequent HTTP requests will include a correct locale value, and we can use that to determine whether to serve them an English or a French page.

The code above means that our HTTP requests will match one of two cases:

1. The first time the user loads the application, no cookie will be set, in which case we can default to returning the English variant of the server-side rendered app and wait until we have access to navigator.language to select the system-correct locale
2. All subsequent requests will have a locale cookie which we can use to determine which language we should return when we are querying our document store.

We can handle this by rendering different variants of our application. Let's assume that our application supports en-US, en-CA and fr-FR locales. This is how we would configure the server:

import {Injector, Injectable} from '@angular/core';

import {StateTransition} from 'angular-ssr';

@Injectable()
export class LocaleTransition implements StateTransition<string> {
  constructor(private localeService: LocaleService) {}

  // This is the bit of code that actually executes the transition to set the locale
  // to whichever value is being rendered (but value is guaranteed to be one of the
  // values from the Set we created when we first described the locale variant below).
  // Note that this class can use the ng dependency injection system to retrieve any
  // services that it needs in order to execute the state transition.
  execute(value: string) {
    this.localeService.locale(value);
  }
}

const application = new ApplicationFromModule(AppModule, join(process.cwd(), 'dist', 'index.html'));

application.variants({
  locale: {
    values: new Set<string>([
      'en-CA',
      'en-US',
      'fr-FR'
    ]),
    transition: LocaleTransition
  }
});

type ApplicationVariants = {locale: string};

// DocumentVariantStore is a variant-aware special-case of DocumentStore. When you
// query it, you must provide values for each variant key that we described in the
// call to variants() above. But in this application, there is only one key: locale.
const documentStore = new DocumentVariantStore<ApplicationVariants>(application);

app.get('*', async (req, res) => {
  try {
    // Remember that we set locale in document.cookie, so all requests after the
    // first-ever application load will have a locale cookie that we can use to
    // decide whether to give the user an English or French pre-rendered page.
    const snapshot = await documentStore.load(req.url, {locale: req.cookies.locale});

    res.send(snapshot.renderedDocument);
  }
  catch (exception) {
    res.send(application.templateDocument()); // fall back on client-side rendering
  }
});

Voila! Now whenever the user reloads our application or comes back to it in a few days, we are going to hand them a pre-rendered document that is in the language of their choosing! Simple.

The example in examples/demand-express has working code that implements what was just described. Give it a shot!

Many applications may wish to transfer some state from the server to the client as part of application bootstrap. angular-ssr makes this easy. Simply tell your ApplicationBuilder object about your state reader class or function, and any state returned from it will be made available in a global variable called bootstrapApplicationState:

const application = new ApplicationFromModule(AppModule);

application.stateReader(ServerStateReader);

And your ServerStateReader class implementation might look like this:

import {Injectable} from '@angular/core';

import {Store} from '@ngrx/store';

import {StateReader} from 'angular-ssr';

@Injectable()
export class ServerStateReader implements StateReader {
  constructor(private store: Store<AppState>) {}

  getState(): Promise<MyState> {
    return this.store.select(s => s.someState).take(1).toPromise();
  }
}

Note that you can inject any service you wish into your state reader. angular-ssr will query the constructor arguments using the ng dependency injector the same way it works in application code. Alternatively, you can supply a function which just accepts a bare Injector and you can query the DI yourself:

import {Store} from '@ngrx/store';

application.stateReader(
  (injector: Injector) => {
    return injector.get(Store).select(s => s.fooBar).take(1).toPromise();
  });

Both solutions are functionally equivalent.

Note that your state reader will not be called until your application zone becomes stable. That is to say, when all macro and microtasks have finished. (For example, if your application has some pending HTTP requests, angular-ssr will wait for those to finish before asking your state reader for its state. This ensures that your application has finished initializing itself by the time the state reader is invoked.)

The main contract that you use to define your application in a server context is called ApplicationBuilder. It has thorough comments and explains all the ways that you can configure your application when doing server-side rendering.

But ApplicationBuilder is an interface. It has three concrete implementations:

• ApplicationFromModule<V, M>
  • If your code has access to the root @NgModule of your application, then this is probably the ApplicationBuilder that you want to use. It takes a module type and a template HTML document (dist/index.html) as its constructor arguments.
• ApplicationFromModuleFactory<V>
  • If your application code has already been run through ngc and produced .ngfactory.js files, then you can pass your root @NgModule's NgFactory -- not the module definition itself, but its compilation output -- to ApplicationFromModuleFactory<V> and you can skip the template compilation process.
• ApplicationFromSource<V>
  • You can use this for projects that use @angular/cli if you wish to use inplace compilation to generate an NgModuleFactory from raw source code. It's fairly unlikely that you will ever use this class: its main purpose is for the implementation of the ng-render command.

Other classes of interest are DocumentStore and DocumentVariantStore. You can use these in conjunction with ApplicationBuilder to maintain and query a cache of rendered pages.

Another interesting one is Snapshot. This is the data structure you get back from the server-side rendering process. It takes a type argument that represents the variants your application is aware of, or void if you are not using variants.

One thing to note about Snapshot is that it contains far more information than just the final rendered document. It has:

• console: Array<ConsoleLog>
  • This is an array of console logs that your application emitted during bootstrap and during the rendering process. It includes errors, infos, warnings, assertions, and anything else that you can emit via console.
• exceptions: Array<Error>
  • This is an array containing any uncaught exceptions that were generated during the bootstrap and rendering process. Generally it is possible to get back a snapshot that has an empty document and at least one exception in exceptions, so you should usually check this in your retrieval methods to ensure that everything worked properly. You don't want to send a mangled document to the user.
• renderedDocument: string
  • This is what you are primarily interested in: the final version of the rendered application HTML. This contains the document you will need to send in your HTTP response.
• variant: V
  • If you are using variants, this describes the particular set of variants that were used to generate this snapshot.
• uri: string
  • This is the URI that was originally given to the renderer when this snapshot was generated.

The examples/cli folder contains a project that was generated with ng new, and which also integrates with @angular/material, and uses the ng-render command to render itself as part of the build process. This is the simplest possible usage of angular-ssr and covers very basic applications.

A project using express and angular-ssr lives in the examples/demand-express directory.

A project using koa and angular-ssr lives in the examples/demand-koa directory.

1. Do I have to create separate modules for client and server?
   • No. You can use your existing root module even if it imports BrowserModule, BrowserAnimationsModule or any other such imports.
2. Can I send HTTP requests when the application is running on the server? Do I have to import a different HttpModule?
   • Yes, you can use the HTTP libraries in any way that you normally would. But if you want to avoid requesting the same data on both the server and the client, I would suggest that you make use of the state transfer feature to transmit this data to the client from the server. This way the client does not need to make a duplicate request to retrieve the same data that the server already got.
3. Can I use document or window? What if I am using a library that accesses document or window?
   • While it is generally inadvisable to access document and window from inside an Angular application, I recognize that real-world applications often use libraries that manipulate or query the DOM in some way or another. Therefore, you can access and manipulate document and window from an angular-ssr application running on the server. Each rendering context gets its own instance of document and window and using these objects should generally not produce any issues.
   • However, it is important to note that while your application will have access to document and window on the server, any operations designed to get the pixel size or location of any particular elements is likely to fail or return all zeros. This is because there is no rendering happening in the process. You are given a working DOM implementation, but that doesn't make it a browser. So adding, removing and manipulating DOM elements is OK, but if you are trying to query the size of certain elements, that is not a strategy that is going to work on code running on the server. You must avoid that, but mostly everything else is fair game.
4. Can I use jQuery plugins?
   • For the most part, yes, but again with the caveat that anything which tries to query pixel sizes is going to fail or return zeros. Generally speaking it is best to avoid jQuery plugins in Angular 4 applications, but you can probably still make it work with angular-ssr.

Direct your vitriol to [email protected] or post an issue on this GitHub repo!