Ember Data

Ember Data is a library for loading models from a persistence layer (such as a JSON API), updating those models, then saving the changes. It provides many of the facilities you'd find in server-side ORMs like ActiveRecord, but is designed specifically for the unique environment of JavaScript in the browser.

This release is definitely alpha-quality. The basics work, but there are for sure edge cases that are not yet handled. Please report any bugs or feature requests, and pull requests are always welcome.

Is It Good?

Yes.

Is It "Production Ready™"?

No. Breaking changes, indexed by date, are listed in BREAKING_CHANGES.md.

Roadmap

• Handle error states
• Better built-in attributes
• Editing "forked" records
• Out-of-the-box support for Rails apps that follow the active_model_serializers gem's conventions.
• Handle partially-loaded records

Creating a Store

Every application has one or more stores. The store will be the repository that holds loaded models, and is responsible for retrieving models that have not yet been loaded.

App.store = DS.Store.create({
  revision: 4
});

NOTE: The revision property is used by ember-data to notify you of breaking changes to the public API before 1.0. For new applications, just set the revision to this number. See BREAKING_CHANGES.md for more information.

You can tell the store how to talk to your backend by specifying an adapter. Ember Data comes with a RESTful JSON API adapter. You can specify this adapter by setting the adapter property:

App.store = DS.Store.create({
  revision: 4,
  adapter: DS.RESTAdapter.create({ bulkCommit: false })
});

The RESTAdapter supports the following options:

• bulkCommit (default: false): If your REST API does not support bulk operations, you can turn them off by setting bulkCommit to false.

• namespace (default: undefined): A leading URL component under which all REST URLs reside, without leading slash, e.g. api (for /api/authors/1-type URLs).

The RESTful adapter is still in progress. For Rails applications, we plan to make it work seamlessly with the active_model_serializers gem's conventions. In the meantime, see the section on rolling your own adapter.

Defining Models

For every type of data you'd like to represent, create a new subclass of DS.Model:

App.Person = DS.Model.extend();

You can specify which attributes a model has by using DS.attr. An attribute represents a value that will exist in the underlying JSON representation of the object which you'd also want to expose through the Ember object.

You can use attributes just like any other property, including as part of a computed property. These attributes ensure that the values can be retrieved from the underlying JSON representation and persisted later as needed.

App.Person = DS.Model.extend({
    firstName: DS.attr('string'),
    lastName: DS.attr('string'),
    birthday: DS.attr('date'),

    fullName: function() {
        return this.get('firstName') + ' ' + this.get('lastName');
    }.property('firstName', 'lastName')
});

Valid attribute types are string, number, boolean, and date. You can also register custom attribute types. For example, here's a boolString attribute type that converts booleans into the string "Y" or "N":

DS.attr.transforms.boolString = {
    from: function(serialized) {
        if (serialized === 'Y') {
            return true;
        }

        return false;
    },

    to: function(deserialized) {
        if (deserialized) {
            return "Y";
        }
        return "N";
    }
}

Built-in attribute types are currently very primitive. Please help us flesh them out with patches and unit tests!

By default, the store uses a model's id attribute as its primary key. You can specify a different key by setting the primaryKey property:

DS.Model.extend({
    primaryKey: 'guid'
});

Associations

Models can be associated with other models. Ember Data includes several built-in types to help you define how your models relate to each other.

Has One

A hasOne association declares that a model is associated with exactly one other model. For example, imagine we're writing a blog application that allows authors to post entries. Each author has a profile associated with their account that is displayed when visitors want to learn more about them.

App.Profile = DS.Model.extend({
  about: DS.attr('string'),
  postCount: DS.attr('number')
});

App.Author = DS.Model.extend({
  profile: DS.hasOne('App.Profile'),
  name: DS.attr('string')
});

Now, when we have an Author record, we can easily find its related Profile:

var author = App.store.find(App.Author, 1);
author.get('name'); // "Timothy Leary"
author.get('profile'); // App.Profile
author.getPath('profile.postCount'); // 1969

Belongs To

Similar to hasOne, belongsTo sets up a one-to-one relationship from one model to another. Let's revise the example above so that, in addition to being able to find an author's profile, we can find the author associated with a profile:

App.Profile = DS.Model.extend({
  about: DS.attr('string'),
  postCount: DS.attr('number'),
  author: DS.belongsTo('App.Author')
});

App.Author = DS.Model.extend({
  profile: DS.hasOne('App.Profile'),
  name: DS.attr('string')
});

So, when should you use hasOne and when should you use belongsTo? The difference is where the information about the relationship is stored at the persistence layer.

The record with the belongsTo relationship will save changes to the association on itself. Conversely, the record with the hasOne relationship asks the persistence layer what record belongs to it. If the relationship changes, only the record with the belongsTo relationship must be saved.

Has Many

Use the hasMany() method to describe a relationship where multiple models belong to a single model. In our blog engine example, a single blog post may have multiple comments:

App.Comment = DS.Model.extend({
    content: DS.attr('string'),
    post: DS.belongsTo('App.Post')
});

App.Post = DS.Model.extend({
    content: DS.attr('string'),
    comments: DS.hasMany('App.Comment')
});

Representing Associations

One-to-One

The default REST adapter supports several different variations for representing associations to best fit the needs of your application. We'll examine each of the different types of associations in turn.

Let us first discuss a typical one-to-one relationship. We'll use the example of the Profile and Author from above. In that example, we have a Profile that belongs to an Author, and an Author that has one Profile. The simplest way to represent these two records' JSON structure is as follows:

// Author
{
  "author": {
      "id": 1,
      "name": "Tom Dale"
    }
  }
}

// Profile
{
  "profile": {
    "id": 1,
    "about": "Tom Dale is a software engineer that drinks too much beer.",
    "postCount": 1984,
    "author_id": 1
  }
}

Note that in the above example, the JSON for our Author does not contain any information about how to find its related Profile. If you were to request the profile, like this:

author.get('profile');

…the REST adapter would send a request to the URL /profiles?author_id=1. (Asking the Profile for its Author would not generate an additional request, because the ID of the associated Author is built-in to the response.)

As a performance optimization, the REST API can return the ID of the Profile in the Author JSON:

// Author with included Profile id
{
  "author": {
    "id": 1,
    "name": "Tom Dale",
    "profile_id": 1
  }
}

Now, if you ask for the author's profile, one of two things will happen. If the Profile with that ID has already been loaded at any point during the execution of the app, it will be returned immediately without any additional requests. Otherwise, the REST adapter will make a request to /profile/1 to load that specific profile.

In some cases, if you know that you will always being using both records in an association, you may want to minimize the number of HTTP requests by including both records in the same JSON.

One option is to embed the association directly in the parent record. For example, we could represent the entirety of the association above like this:

{
  "authors": [{
    "id": 1,
    "name": "Tom Dale",
    "profile": {
      "id": 1,
      "about": "Tom Dale is a software engineer that drinks too much beer.",
      "postCount": 1984,
      "author_id": 1
    }
  }]
}

If you do this, note that Profile.find(1) will still trigger an Ajax request until you access the embedded profile record for the first time (Author.find(1).get('profile')).

Another option is to use the format described above (with the ID embedded), then "sideloading" the records. For example, we could represent the entirety of the association above like this:

{
  "authors": [{
    "id": 1,
    "name": "Tom Dale",
    "profile_id": 1
  }],

  "profiles": [{
    "id": 1,
    "about": "Tom Dale is a software engineer that drinks too much beer.",
    "postCount": 1984,
    "author_id": 1
  }]
}

However, imagine the JSON returned from the server for a Person looked like this:

{
  "person": {
    "id": 1,
    "name": "Tom Dale",
    "tags": [{
      "id": 1,
      "name": "good-looking"
    },

    {
      "id": 2,
      "name": "not-too-bright"
    }]
  }
}

In this case, instead of the association being an array of ids, it is an array of embedded objects. To have the store understand these correctly, set the embedded option to true:

App.Person = DS.Model.extend({
    tags: DS.hasMany('App.Tag', { embedded: true })
});

It is also possible to change the data attribute that an association is mapped to. Suppose the JSON for a Person looked like this:

{
  "person": {
    "id": 2,
    "name": "Carsten Nielsen",
    "tag_ids": [1, 2]
  }
}

In this case, you would specify the key in the association like this:

App.Person = DS.Model.extend({
    tags: DS.hasMany('App.Tag', { key: 'tag_ids' })
});

Finding a Specific Record Instance

You can retrieve a record by its unique ID by using the find method:

var model = App.store.find(App.Person, 1);

If that specific record has already been loaded, it will be returned immediately. Otherwise, an empty object will be returned. You can setup bindings and observers on the properties you're interested in; as soon as the data returns from the persistence layer, all of the attributes you specified will be updated automatically.

Besides find(), all of the methods described below operate in a similar fashion.

Querying Record Instances

You can make a server query by passing an Object as the second parameter to find. In this case, you will get back a ModelArray object.

App.people = App.store.find(App.Person, { page: 1 });

At first, this people array will have no elements. Later, we will see how your adapter will populate the people. Because the people array is an Ember Array, you can immediately insert it into the DOM. When it becomes populated later, Ember's bindings will automatically update the DOM.

<ul>
{{#each App.people}}
  <li>{{fullName}}</li>
{{/each}}
</ul>

This will allow you to ask the store for an Array of information, and keep your view code completely agnostic to how the Array becomes populated.

Note: If manually retrieving records from a ModelArray, you must use the objectAt(index) method. Since the object is not a JavaScript Array, using the [] notation will not work.

Finding All Records of a Model Type

To find all records of a certain type, use the store's findAll() method:

var people = App.store.findAll(App.Person);

All currently loaded records of that type will be immediately returned in a ModelArray. Your adapter will also have an opportunity to load additional records of that type if necessary.

Whenever a new record is loaded into the store for the type in question, the ModelArray returned by findAll will update to reflect the new data. This means that you can pass it to a #each in an Ember template and it will stay up to date as new data is loaded.

Filtering Loaded Records

You can filter all records of a model type by calling the store's filter() method with a function that determines whether the record should be included or not. To avoid materializing record objects needlessly, only the raw data hash returned from the persistence layer is passed.

To include a record, return true. If a record should not be included, return false or undefined.

var oldPeople = App.store.filter(App.Person, function(data) {
    if (data.age > 80) { return true; }
});

Creating New Records

You can create new record based on a particular model definition with createRecord():

var wycats = App.store.createRecord(App.Person,  { name: "Brohuda" });

New records are not saved back to the persistence layer until the store's commit() method is called.

Updating Records

To update records, simply change a property on them. Updated records will not be saved until the store's commit() method is called, which allows you to batch changes.

Deleting Records

To delete a record, call its deleteRecord() method:

var person = App.store.find(App.Person, 1);
person.deleteRecord();

The record will not be deleted in the persistence layer until the store's commit() method is called. However, deleted records will immediately be removed from its ModelArray and associations.

Record Lifecycle

You can be notified when certain events occur in a record's lifecycle by implementing methods on them:

• didCreate - called when the record has been successfully created in the persistence layer
• didUpdate - called when changes have been successfully saved to the persistence layer
• didLoad - called when data has finished loading from the persistence layer

For example:

App.Person = DS.Model.extend({
    didLoad: function() {
        alert(this.get('firstName') + " finished loading.");
    }
});

You can also determine the state of a record by checking its state properties.

• isLoaded - true when the record has finished loading, always true for models created locally.
• isDirty - true for created, updated, or deleted records that have not yet been saved
• isSaving - true if the record is in the process of being saved
• isDeleted - true if the record has been deleted, either locally or on the server
• isError - true if the record is in an error state

Loading Data

You can "pre-load" data into the store, so it's ready for your users as soon as they need it.

To load an individual record, use the load() method:

App.store.load(App.Person, {
    id: 1,
    firstName: "Peter",
    lastName: "Wagenet"
});

You can load multiple records using loadMany():

App.store.loadMany(App.Person, [{
    id: 2,
    firstName: "Erik",
    lastName: "Brynjolsofosonsosnson"
},

{
    id: 3,
    firstName: "Yehuda",
    lastName: "Katz"
}]);

Adapter API

An adapter is an object that receives requests from a store and translates them into the appropriate action to take against your persistence layer. The persistence layer is usually an HTTP API, but may be anything, such as the browser's local storage.

Creating an Adapter

First, create a new instance of DS.Adapter:

App.adapter = DS.Adapter.create();

To tell your store which adapter to use, set its adapter property:

App.store = DS.Store.create({
  revision: 3,
  adapter: App.adapter
});

Next, implement the methods your adapter needs, as described below.

find()

Implement find() to fetch and populate a record with a specific ID. Once the record has been found, call the store's load() method:

App.Person = DS.Model.extend();
App.Person.reopenClass({
    url: '/people/%@'
});

DS.Adapter.create({
    find: function(store, type, id) {
        var url = type.url;
        url = url.fmt(id);

        jQuery.getJSON(url, function(data) {
            // data is a Hash of key/value pairs. If your server returns a
            // root, simply do something like:
            // store.load(type, id, data.person)
            store.load(type, id, data);
        });
    }
});

The store will call your adapter's find() method when you call store.find(type, id).

Note that for the rest of this documentation, we will use the url property in our adapter. This is not the only way to write an adapter. For instance, you could simply put a case statement in each method and do something different per type. Or you could expose different information on your types that you use in the adapter. We are simply using url to illustrate how an adapter is written.

findMany()

Implement findMany() to fetch and populate all of the records for a given list of IDs. The default findMany() will repeatedly invoke find(), but this may be extremely inefficient. If you can, your server should support a way to find many items by a list of IDs.

Once you are ready to populate the store with the data for the requested IDs, use the loadMany method:

App.Person = DS.Model.extend();
App.Person.reopenClass({
    url: '/people?ids=%@'
});

DS.Adapter.create({
    findMany: function(store, type, ids) {
        var url = type.url;
        url = url.fmt(ids.join(','));

        jQuery.getJSON(url, function(data) {
            // data is an Array of Hashes in the same order as the original
            // Array of IDs. If your server returns a root, simply do something
            // like:
            // store.loadMany(type, ids, data.people)
            store.loadMany(type, ids, data);
        });
    }
});

Implementing findMany in Rails

It is extremely easy to implement an endpoint that will find many items in Ruby on Rails. Simply define the index action in a standard resourceful controller to understand an :ids parameter.

class PostsController < ApplicationController
  def index
    if ids = params[:ids]
      @posts = Post.where(:id => ids)
    else
      @posts = Post.scoped
    end

    respond_with @posts
  end
end

findQuery()

Called when the store's find() method is called with a query. Your adapter's findQuery() method will be passed a ModelArray that you should populate with the results returned by the server.

App.Person = DS.Model.extend();
App.Person.reopenClass({
    collectionUrl: '/people'
});

DS.Adapter.create({
    findQuery: function(store, type, query, modelArray) {
        var url = type.collectionUrl;

        jQuery.getJSON(url, query, function(data) {
            // data is expected to be an Array of Hashes, in an order
            // determined by the server. This order may be specified in
            // the query, and will be reflected in the view.
            //
            // If your server returns a root, simply do something like:
            // modelArray.load(data.people)
            modelArray.load(data);
        });
    }
});

You can do whatever you want with the query in your adapter, but most commonly, you will just send it along to the server as the data portion of an Ajax request.

Your server will then be responsible for returning an Array of JSON data. When you load the data into the modelArray, the elements of that Array will be loaded into the store at the same time.

findAll()

Invoked when findAll() is called on the store. If you do nothing, only models that have already been loaded will be included in the results. Otherwise, this is your opportunity to load any unloaded records of this type. The implementation is similar to findMany(); see above for an example.

createRecord()

When commit() is called on the store and there are records that need to be created on the server, the store will call the adapter's create() method.

Once the store calls the adapter's create method, it will be put into a saving state, and further attempts to edit the model will result in an error.

Implementing a create method is straight forward:

App.Person = DS.Model.extend();
App.Person.reopenClass({
    url: '/people/%@'
});

DS.Adapter.create({
    createRecord: function(store, type, model) {
        var url = type.url;

        jQuery.ajax({
            url: url.fmt(model.get('id')),
            data: model.get('data'),
            dataType: 'json',
            type: 'POST',

            success: function(data) {
                // data is a hash of key/value pairs representing the record.
                // In general, this hash will contain a new id, which the
                // store will now use to index the record. Future calls to
                // store.find(type, id) will find this record.
                store.didCreateRecord(model, data);
            }
        });
    })
});

createRecords()

For better efficiency, you can implement a createRecords method on your adapter, which should send all of the new models to the server at once.

App.Person = DS.Model.extend();
App.Person.reopenClass({
    collectionUrl: '/people'
});

DS.Adapter.create({
    createRecords: function(store, type, array) {
        jQuery.ajax({
            url: type.collectionUrl,
            data: array.mapProperty('data'),
            dataType: 'json',
            type: 'POST',

            success: function(data) {
                // data is an array of hashes in the same order as
                // the original records that were sent.
                store.didCreateRecords(type, array, data);
            }
        });
    })
});

updateRecord()

Update is implemented the same as createRecord(), except after the record has been saved, you should call the store's didUpdateRecord() method.

App.Person = DS.Model.extend();
App.Person.reopenClass({
    url: '/people/%@'
});

DS.Adapter.create({
    updateRecord: function(store, type, model) {
        var url = type.url;

        jQuery.ajax({
            url: url.fmt(model.get('id')),
            dataType: 'json',
            type: 'PUT',

            success: function(data) {
                // data is a hash of key/value pairs representing the record
                // in its current state on the server.
                store.didUpdateRecord(model, data);
            }
        });
    })
});

updateRecords()

Again, updateRecords() is very similar to createRecords().

App.Person = DS.Model.extend();
App.Person.reopenClass({
    collectionUrl: '/people'
});

DS.Adapter.create({
    updateRecords: function(store, type, array) {
        jQuery.ajax({
            url: type.collectionUrl,
            data: array.mapProperty('data'),
            dataType: 'json',
            type: 'PUT',

            success: function(data) {
                // data is an array of hashes in the same order as
                // the original records that were sent.
                store.didUpdateRecords(array);
            }
        });
    })
});

deleteRecord()

To delete a record, implement the deleteRecord() method, and call the store's didDeleteRecord() method when completed.

App.Person = DS.Model.extend();
App.Person.reopenClass({
    url: '/people/%@'
});

DS.Adapter.create({
    deleteRecord: function(store, type, model) {
        var url = type.url;

        jQuery.ajax({
            url: url.fmt(model.get('id')),
            dataType: 'json',
            type: 'DELETE',

            success: function() {
                store.didDeleteRecord(model);
            }
        });
    })
});

deleteRecords()

Are you getting it?

App.Person = DS.Model.extend();
App.Person.reopenClass({
    collectionUrl: '/people'
});

DS.Adapter.create({
    deleteRecords: function(store, type, array) {
        jQuery.ajax({
            url: type.collectionUrl,
            data: array.mapProperty('data'),
            dataType: 'json',
            type: 'DELETE',

            success: function(data) {
                store.didDeleteRecords(array);
            }
        });
    })
});

commit()

For maximum turbo-efficiency, you can package all pending changes (creates, updates, and deletes) into one mega package of data awesomeness. To do so, implement commit(), which will be called with everything that needs to be sent to the persistence layer.

Here's what the default adapter's commit() method looks like:

commit: function(store, commitDetails) {
  commitDetails.updated.eachType(function(type, array) {
    this.updateRecords(store, type, array.slice());
  }, this);

  commitDetails.created.eachType(function(type, array) {
    this.createRecords(store, type, array.slice());
  }, this);

  commitDetails.deleted.eachType(function(type, array) {
    this.deleteRecords(store, type, array.slice());
  }, this);
}

Connecting to Views

Ember Data will always return records or arrays of records of a certain type immediately, even though the underlying JSON objects have not yet been returned from the server.

In general, this means that you can insert them into the DOM using Ember's Handlebars template engine, and they will automatically update when your adapter has populated them.

For example, if you request a ModelArray:

App.people = App.store.find(App.Person, { firstName: "Tom" });

You will get back a ModelArray that is currently empty. Ember Data will then ask your adapter to populate the ModelArray with records, which will usually make an Ajax request. However, you can immediately refer to it in your templates:

<ul>
{{#each App.people}}
    <li>{{fullName}}</li>
{{/each}}
</ul>

Once the Adapter calls modelArray.load(array), the DOM will automatically populate with the new information.

The same is true of records themselves. For instance, you can make a request for a single record:

App.person = App.store.find(App.Person, 1);

You will immediately receive back a new unpopulated Person object. You can refer to it in the view right away:

{{App.person.fullName}}

Initially, this will be empty, but when your adapter calls store.load(hash), it will update with the information provided.

If you'd like to show different content while a record is in the process of being loaded, you can use the record's isLoaded property:

{{#with App.person}}
    {{#if isLoaded}}
        Hello, {{fullName}}!
    {{else}}
        Loading...
    {{/if}}
{{/with}}

Note that the same principle applies to ModelArrays, as well. Like records, a ModelArray has an isLoaded property that you can use to display different content.

You can also indicate to users when a record is saving, for example:

{{#with App.person}}
    <h1 {{bindAttr class="isSaving"}}>{{fullName}}</h1>
{{/with}}

In this case, you could make the is-saving class in your CSS grey out the content or add a spinner alongside it, for instance.

Unit Tests

To run unit tests, run bundle exec rackup from the root directory and visit http://localhost:9292/tests/index.html?package=ember-data.

What next?

Profit.