Application Structure

How to structure your Meteor app with ES2015 modules, ship code to the client and server, and split your code into multiple apps.

After reading this article, you’ll know:

  1. How a Meteor application compares to other types of applications in terms of file structure.
  2. How to organize your application both for small and larger applications.
  3. How to format your code and name the parts of your application in consistent and maintainable ways.

Universal JavaScript

Meteor is a full-stack framework for building JavaScript applications. This means Meteor applications differ from most applications in that they include code that runs on the client, inside a web browser or Cordova mobile app, code that runs on the server, inside a Node.js container, and common code that runs in both environments. The Meteor build tool allows you to easily specify what JavaScript code, including any supporting UI templates, CSS rules, and static assets, to run in each environment using a combination of ES2015 import and export and the Meteor build system default file load order rules.

ES2015 modules

As of version 1.3, Meteor ships with full support for ES2015 modules. The ES2015 module standard is the replacement for CommonJS and AMD, which are commonly used JavaScript module format and loading systems.

In ES2015, you can make variables available outside a file using the export keyword. To use the variables somewhere else, you must import them using the path to the source. Files that export some variables are called “modules”, because they represent a unit of reusable code. Explicitly importing the modules and packages you use helps you write your code in a modular way, avoiding the introduction of global symbols and “action at a distance”.

Since this is a new feature introduced in Meteor 1.3, you will find a lot of code online that uses the older, more centralized conventions built around packages and apps declaring global symbols. This old system still works, so to opt-in to the new module system code must be placed inside the imports/ directory in your application. We expect a future release of Meteor will turn on modules by default for all code, because this is more aligned with how developers in the wider JavaScript community write their code.

You can read about the module system in detail in the modules package README. This package is automatically included in every new Meteor app as part of the ecmascript meta-package, so most apps won’t need to do anything to start using modules right away.

Introduction to using import and export

Meteor allows you to import not only JavaScript in your application, but also CSS and HTML to control load order:

import '../../api/lists/methods.js';  // import from relative path
import '/imports/startup/client'; // import module with index.js from absolute path
import './loading.html'; // import Blaze compiled HTML from relative path
import '/imports/ui/style.css'; // import CSS from absolute path

For more ways to import styles, see the Build System article.

Meteor also supports the standard ES2015 modules export syntax:

export const listRenderHold = LaunchScreen.hold();  // named export
export { Todos }; // named export
export default Lists; // default export
export default new Collection('lists'); // default export

Importing from packages

In Meteor, it is also simple and straightforward to use the import syntax to load npm packages on the client or server and access the package’s exported symbols as you would with any other module. You can also import from Meteor Atmosphere packages, but the import path must be prefixed with meteor/ to avoid conflict with the npm package namespace. For example, to import moment from npm and HTTP from Atmosphere:

import moment from 'moment';          // default import from npm
import { HTTP } from 'meteor/http'; // named import from Atmosphere

For more details using imports with packages see Using Packages in the Meteor Guide.

Using require

In Meteor, import statements compile to CommonJS require syntax. However, as a convention we encourage you to use import.

With that said, in some situations you may need to call out to require directly. One notable example is when requiring client or server-only code from a common file. As imports must be at the top-level scope, you may not place them within an if statement, so you’ll need to write code like:

if (Meteor.isClient) {

Note that dynamic calls to require() (where the name being required can change at runtime) cannot be analyzed correctly and may result in broken client bundles.

If you need to require from an ES2015 module with a default export, you can grab the export with require("package").default.

Another situation where you’ll need to use require is in CoffeeScript files. As CS doesn’t support the import syntax yet, you should use require:

{ FlowRouter } = require 'meteor/kadira:flow-router'
React = require 'react'

Exporting with CoffeeScript

When using CoffeeScript, not only the syntax to import variables is different, but also the export has to be done in a different way. Variables to be exported are put in the exports object:

exports.Lists = ListsCollection 'lists'

File structure

To fully use the module system and ensure that our code only runs when we ask it to, we recommend that all of your application code should be placed inside the imports/ directory. This means that the Meteor build system will only bundle and include that file if it is referenced from another file using an import (also called “lazy evaluation or loading”).

Meteor will load all files outside of any directory named imports/ in the application using the default file load order rules (also called “eager evaluation or loading”). It is recommended that you create exactly two eagerly loaded files, client/main.js and server/main.js, in order to define explicit entry points for both the client and the server. Meteor ensures that any file in any directory named server/ will only be available on the server, and likewise for files in any directory named client/. This also precludes trying to import a file to be used on the server from any directory named client/ even if it is nested in an imports/ directory and vice versa for importing client files from server/.

These main.js files won’t do anything themselves, but they should import some startup modules which will run immediately, on client and server respectively, when the app loads. These modules should do any configuration necessary for the packages you are using in your app, and import the rest of your app’s code.

Example directory layout

To start, let’s look at our Todos example application, which is a great example to follow when structuring your app. Here’s an overview of its directory structure:

index.js # import client startup through a single index entry point
routes.js # set up all routes in the app
useraccounts-configuration.js # configure login templates
fixtures.js # fill the DB with example data on startup
index.js # import server startup through a single index entry point

lists/ # a unit of domain logic
publications.js # all list-related publications
publications.tests.js # tests for the list publications
lists.js # definition of the Lists collection
lists.tests.js # tests for the behavior of that collection
methods.js # methods related to lists
methods.tests.js # tests for those methods

components/ # all reusable components in the application
# can be split by domain if there are many
layouts/ # wrapper components for behaviour and visuals
pages/ # entry points for rendering used by the router

main.js # client entry point, imports all client code

main.js # server entry point, imports all server code

Structuring imports

Now that we have placed all files inside the imports/ directory, let’s think about how best to organize our code using modules. It makes sense to put all code that runs when your app starts in an imports/startup directory. Another good idea is splitting data and business logic from UI rendering code. We suggest using directories called imports/api and imports/ui for this logical split.

Within the imports/api directory, it’s sensible to split the code into directories based on the domain that the code is providing an API for — typically this corresponds to the collections you’ve defined in your app. For instance in the Todos example app, we have the imports/api/lists and imports/api/todos domains. Inside each directory we define the collections, publications and methods used to manipulate the relevant domain data.

Note: in a larger application, given that the todos themselves are a part of a list, it might make sense to group both of these domains into a single larger “list” module. The Todos example is small enough that we need to separate these only to demonstrate modularity.

Within the imports/ui directory it typically makes sense to group files into directories based on the type of UI side code they define, i.e. top level pages, wrapping layouts, or reusable components.

For each module defined above, it makes sense to co-locate the various auxiliary files with the base JavaScript file. For instance, a Blaze UI component should have its template HTML, JavaScript logic, and CSS rules in the same directory. A JavaScript module with some business logic should be co-located with the unit tests for that module.

Startup files

Some of your code isn’t going to be a unit of business logic or UI, it’s just some setup or configuration code that needs to run in the context of the app when it starts up. In the Todos example app, the imports/startup/client/useraccounts-configuration.js file configures the useraccounts login templates (see the Accounts article for more information about useraccounts). The imports/startup/client/routes.js configures all of the routes and then imports all other code that is required on the client:

import { FlowRouter } from 'meteor/kadira:flow-router';
import { BlazeLayout } from 'meteor/kadira:blaze-layout';
import { AccountsTemplates } from 'meteor/useraccounts:core';

// Import to load these templates
import '../../ui/layouts/app-body.js';
import '../../ui/pages/root-redirector.js';
import '../../ui/pages/lists-show-page.js';
import '../../ui/pages/app-not-found.js';

// Import to override accounts templates
import '../../ui/accounts/accounts-templates.js';

// Below here are the route definitions

We then import both of these files in imports/startup/client/index.js:

import './useraccounts-configuration.js';
import './routes.js';

This makes it easy to then import all the client startup code with a single import as a module from our main eagerly loaded client entry point client/main.js:

import '/imports/startup/client';

On the server, we use the same technique of importing all the startup code in imports/startup/server/index.js:

// This defines a starting set of data to be loaded if the app is loaded with an empty db.
import '../imports/startup/server/fixtures.js';

// This file configures the Accounts package to define the UI of the reset password email.
import '../imports/startup/server/reset-password-email.js';

// Set up some rate limiting and other important security settings.
import '../imports/startup/server/security.js';

// This defines all the collections, publications and methods that the application provides
// as an API to the client.
import '../imports/api/api.js';

Our main server entry point server/main.js then imports this startup module. You can see that here we don’t actually import any variables from these files - we just import them so that they execute in this order.

Importing Meteor “pseudo-globals”

For backwards compatibility Meteor 1.3 still provides Meteor’s global namespacing for the Meteor core package as well as for other Meteor packages you include in your application. You can also still directly call functions such as Meteor.publish, as in previous versions of Meteor, without first importing them. However, it is recommended best practice that you first load all the Meteor “pseudo-globals” using the import { Name } from 'meteor/package' syntax before using them. For instance:

import { Meteor } from 'meteor/meteor';
import { EJSON } from 'meteor/ejson';

Default file load order

Even though it is recommended that you write your application to use ES2015 modules and the imports/ directory, Meteor 1.3 continues to support eager loading of files, using these default load order rules, to provide backwards compatibility with applications written for Meteor 1.2 and earlier. You may combine both eager loading and lazy loading using import in a single app. Any import statements are evaluated in the order they are listed in a file when that file is loaded and evaluated using these rules.

There are several load order rules. They are applied sequentially to all applicable files in the application, in the priority given below:

  1. HTML template files are always loaded before everything else
  2. Files beginning with main. are loaded last
  3. Files inside any lib/ directory are loaded next
  4. Files with deeper paths are loaded next
  5. Files are then loaded in alphabetical order of the entire path

For example, the files above are arranged in the correct load order. main.html is loaded second because HTML templates are always loaded first, even if it begins with main., since rule 1 has priority over rule 2. However, it will be loaded after nav.html because rule 2 has priority over rule 5.

client/lib/styles.js and lib/feature/styles.js have identical load order up to rule 4; however, since client comes before lib alphabetically, it will be loaded first.

You can also use Meteor.startup to control when run code is run on both the server and the client.

Special directories

By default, any JavaScript files in your Meteor application folder are bundled and loaded on both the client and the server. However, the names of the files and directories inside your project can affect their load order, where they are loaded, and some other characteristics. Here is a list of file and directory names that are treated specially by Meteor:

  • imports

    Any directory named imports/ is not loaded anywhere and files must be imported using import.

  • node_modules

    Any directory named node_modules/ is not loaded anywhere. node.js packages installed into node_modules directories must be imported using import or by using Npm.depends in package.js.

  • client

    Any directory named client/ is not loaded on the server. Similar to wrapping your code in if (Meteor.isClient) { ... }. All files loaded on the client are automatically concatenated and minified when in production mode. In development mode, JavaScript and CSS files are not minified, to make debugging easier. CSS files are still combined into a single file for consistency between production and development, because changing the CSS file’s URL affects how URLs in it are processed.

    HTML files in a Meteor application are treated quite a bit differently from a server-side framework. Meteor scans all the HTML files in your directory for three top-level elements: <head>, <body>, and <template>. The head and body sections are separately concatenated into a single head and body, which are transmitted to the client on initial page load.

  • server

    Any directory named server/ is not loaded on the client. Similar to wrapping your code in if (Meteor.isServer) { ... }, except the client never even receives the code. Any sensitive code that you don’t want served to the client, such as code containing passwords or authentication mechanisms, should be kept in the server/ directory.

    Meteor gathers all your JavaScript files, excluding anything under the client, public, and private subdirectories, and loads them into a Node.js server instance. In Meteor, your server code runs in a single thread per request, not in the asynchronous callback style typical of Node.

  • public

    All files inside a top-level directory called public/ are served as-is to the client. When referencing these assets, do not include public/ in the URL, write the URL as if they were all in the top level. For example, reference public/bg.png as <img src='/bg.png' />. This is the best place for favicon.ico, robots.txt, and similar files.

  • private

    All files inside a top-level directory called private/ are only accessible from server code and can be loaded via the Assets API. This can be used for private data files and any files that are in your project directory that you don’t want to be accessible from the outside.

  • client/compatibility

    This folder is for compatibility with JavaScript libraries that rely on variables declared with var at the top level being exported as globals. Files in this directory are executed without being wrapped in a new variable scope. These files are executed before other client-side JavaScript files.

    It is recommended to use npm for 3rd party JavaScript libraries and use import to control when files are loaded.

  • tests

    Any directory named tests/ is not loaded anywhere. Use this for any test code you want to run using a test runner outside of Meteor’s built-in test tools.

The following directories are also not loaded as part of your app code:

  • Files/directories whose names start with a dot, like .meteor and .git
  • packages/: Used for local packages
  • cordova-build-override/: Used for advanced mobile build customizations
  • programs: For legacy reasons

Files outside special directories

All JavaScript files outside special directories are loaded on both the client and the server. Meteor provides the variables Meteor.isClient and Meteor.isServer so that your code can alter its behavior depending on whether it’s running on the client or the server.

CSS and HTML files outside special directories are loaded on the client only and cannot be used from server code.

Splitting into multiple apps

If you are writing a sufficiently complex system, there can come a time where it makes sense to split your code up into multiple applications. For example you may want to create a separate application for the administration UI (rather than checking permissions all through the admin part of your site, you can check once), or separate the code for the mobile and desktop versions of your app.

Another very common use case is splitting a worker process away from your main application so that expensive jobs do not impact the user experience of your visitors by locking up a single web server.

There are some advantages of splitting your application in this way:

  • Your client JavaScript bundle can be significantly smaller if you separate out code that a specific type of user will never use.

  • You can deploy the different applications with different scaling setups and secure them differently (for instance you might restrict access to your admin application to users behind a firewall).

  • You can allow different teams at your organization to work on the different applications independently.

However there are some challenges to splitting your code in this way that should be considered before jumping in.

Sharing code

The primary challenge is properly sharing code between the different applications you are building. The simplest approach to deal with this issue is to simply deploy the same application on different web servers, controlling the behavior via different settings. This approach allows you to easily deploy different versions with different scaling behavior but doesn’t enjoy most of the other advantages stated above.

If you want to create Meteor applications with separate code, you’ll have some modules that you’d like to share between them. If those modules are something the wider world could use, you should consider publishing them to a package system, either npm or Atmosphere, depending on whether the code is Meteor-specific or otherwise.

If the code is private, or of no interest to others, it typically makes sense to simply include the same module in both applications (you can do this with private npm modules). There are several ways to do this:

  • a straightforward approach is simply to include the common code as a git submodule of both applications.

  • alternatively, if you include both applications in a single repository, you can use symbolic links to include the common module inside both apps.

Sharing data

Another important consideration is how you’ll share the data between your different applications.

The simplest approach is to point both applications at the same MONGO_URL and allow both applications to read and write from the database directly. This works well thanks to Meteor’s support for reactivity through the database. When one app changes some data in MongoDB, users of any other app connected to the database will see the changes immediately thanks to Meteor’s livequery.

However, in some cases it’s better to allow one application to be the master and control access to the data for other applications via an API. This can help if you want to deploy the different applications on different schedules and need to be conservative about how the data changes.

The simplest way to provide a server-server API is to use Meteor’s built-in DDP protocol directly. This is the same way your Meteor client gets data from your server, but you can also use it to communicate between different applications. You can use DDP.connect() to connect from a “client” server to the master server, and then use the connection object returned to make method calls and read from publications.

Sharing accounts

If you have two servers that access the same database and you want authenticated users to make DDP calls across the both of them, you can use the resume token set on one connection to login on the other.

If your user has connected to server A, then you can use DDP.connect() to open a connection to server B, and pass in server A’s resume token to authenticate on server B. As both servers are using the same DB, the same server token will work in both cases. The code to authenticate looks like this:

// This is server A's token as the default `Accounts` points at our server
const token = Accounts._storedLoginToken();

// We create a *second* accounts client pointing at server B
const app2 = DDP.connect('url://of.server.b');
const accounts2 = new AccountsClient({ connection: app2 });

// Now we can login with the token. Further calls to `accounts2` will be authenticated

You can see a proof of concept of this architecture in an example repository.

Edit on GitHub