Fragmentary

Fragmentary augments the fragment caching capabilities of Ruby on Rails to support:

  • arbitrarily complex dependencies between the content of a fragment and your application data
  • multiple versions of individual fragments for different groups of users, e.g. admin vs regular users
  • post-cache insertion of user-specific content
  • automatic refreshing of cached content when application data changes, without an external client request
  • multiple application instances running concurrently with shared application data

Fragmentary has been extracted from Persuasive Thinking where it is currently in active use.

Background

In simple cases, Rails' native support for fragment caching assumes that a fragment's content is a representation of a specific application data record. The content is stored in the cache with a key value derived from theid and updated_at attributes of that record. If any attributes of the record change, the cached entry automatically expires and on the next browser request for that content the fragment is re-rendered using the current data. In the view, the cache helper is used to specify the record used to determine the key and define the content to be rendered within the fragment, e.g.:

<% cache product do %>
  <%= render product %>
<% end %>

For data models with a has_one or has_many association, nested (Russian Doll) caching is possible with the inner fragment representing the associated record. For example if a product has many games -

<% cache product do %>
  <% product.games.each do |game| %>
    <% cache game do %>
      <%= render game %>
    <% end %>
  <% end %>
<% end %>

By declaring the Game model with touch: true, any changes to a game not only cause the inner game fragment to be updated but also the updated_at attribute of the product it belongs to, thus updating the outer product fragment as well.

class Game < ApplicationRecord
  belongs_to :product, touch: true
end

Part of the beauty of this approach is the fact that any change in data requiring the cache to be refreshed is detected by inspecting just a single updated_at attribute of each record. However it is less effective at handling cases where the content of a fragment depends in more complex ways on multiple records from multiple models. The fact that automatic updating of nested fragments is limited to belongs_to associations is one aspect of this.

It is certainly possible to construct more complex keys from multiple records and models. However this requires retrieving all of those records from the database and computing keys from them for all fragments contained in the server's response every time a request is received, potentially undermining the benefit caching is intended to provide. A related challenge exists in dealing with user-specific content. Here again the user record can easily be incorporated into the key. However if the content of a fragment really must be customized to individual users, this can lead to the number of cache entries escalating dramatically.

A further limitation is that the rendering and storage of cache content relies on an explicit request being received from a user's browser, meaning that at least one user experiences a response time that doesn't benefit from caching every time a relevant change in data occurs. Nested fragments can mitigate this problem for pre-existing pages, since only part of the response need be re-built, but we are still left with the challenge of how to implement nesting in the case of more complex data associations.

Fragmentary - General Approach

Fragmentary uses a database table and corresponding ActiveRecord model that are separate from your application data specifically to represent view fragments. Records in this table serve only as metadata, recording the type of content each fragment contains, where it is located, and when it was last updated. These records play the same role with respect to caching as an application data record in Rails' native approach, i.e. internally a fragment record is passed to Rails' cache method in the same way that product was in the earlier example, and the cache key is derived from the fragment record's updated_at attribute. A publish-subscribe mechanism is used to automatically update this timestamp whenever any application data affecting the content of a fragment is added, modified or destroyed, causing the fragment's cached content to be expired.

To support fragment nesting, each fragment record also includes a parent_id attribute pointing to its immediate parent, or containing, fragment. Whenever the updated_at attribute on an inner fragment changes (initially as a result of a change in some application data the fragment is associated with), as well as expiring the cached content for that fragment, the parent fragment is automatically touched, thus expiring the containing cache as well. This process continues up through all successive ancestors, ensuring that the whole page (or part thereof in the case of some AJAX requests) is refreshed.

Rather than use Rails' native cache helper directly, Fragmentary provides some new helper methods that hide some of the necessary internals involved in working with an explicit fragment model. Additional features include the ability to cache separate versions of fragments for different types of users and the ability to insert user-specific content after cached content is retrieved from the cache store. Fragmentary also supports automatic background updating of cached content within seconds of application data changing, avoiding the need for a user to visit a page in order to update the cache.

Plainly, Fragmentary is more complex to use than Rails' native approach to fragment caching, but it provides much greater flexibility. It was initially developed for an application in which this flexibility was essential in order to achieve acceptable page load times for views derived from relatively complex data models and where data can be changed from multiple page contexts and can affect rendered content in multiple ways in multiple contexts.

Installation

Add this line to your application's Gemfile:

gem 'fragmentary'

And then execute:

$ bundle

Or install it yourself as:

$ gem install fragmentary

Usage

Model Setup

Create a database migration for a table to hold the fragment records.

rails g migration CreateFragments

Here is a suggested migration (choose any table name you wish):

class CreateFragments < ActiveRecord::Migration
  def change
    create_table :fragments do |t|
      t.string :type
      t.index :type
      t.timestamps
      t.belongs_to :root
      t.belongs_to :parent
      t.index :parent_id
      t.integer :record_id
      t.index :record_id
      t.string :user_type
      t.string :key
      t.string :memo
    end
  end
end

Run the migration,

rake db:migrate

Create a Fragment model and include module Fragmentary::Fragment (again, choose whatever model name you wish), e.g.

class Fragment < ActiveRecord::Base
  include Fragmentary::Fragment
end

In any application data model whose records affect the content of any fragment, include module Fragmentary::Publisher, e.g.

class Product < ActiveRecord::Base
  include Fragmentary::Publisher
  ...
end

Create a Fragment subclass (e.g. in app/models/fragments.rb or elsewhere) for each distinct fragment type that your views contain. If a particular fragment type needs to be unique by the id of some application data record, specify the class of the application model as shown in the example below.

class ProductTemplate < Fragment
  needs_record_id :type => 'Product'
end

Here needs_record_id indicates that there is a separate ProductTemplate fragment associated with each Product record. If you need to define further subclasses of your initial subclass, you can if necessary declare needs_record_id on the latter without providing a type and specify the type separately on the individual subclasses using:

set_record_type 'SomeModelName'

We've used this, for example, for fragment types representing different kinds of list items that have certain generic characteristics but are used in different contexts to represent different kinds of content.

Within the body of the fragment subclass definition, for each application model whose records the content of the fragment depends upon, use the subscribe_to method with a block containing method definitions to handle create, update and destroy events on your application data, typically to touch the fragment records affected by the application data change. The names of these methods follow the form used in the wisper-activerecord gem, i.e. create_<model_name>_successful, update_<model_name>_successful and destroy_<model_name>_successful, each taking a single argument representing the application data record that has changed.

Within the body of each method you define within the subscribe_to block, you can retrieve and touch the fragment records affected by the change in application data. The method touch_fragments_for_record can be used for convenience. That method takes an individual application data record or record_id or an array of either. So, for example, if product listings include the names of all the categories the products belong to, with those categories being represented by a separate ActiveRecord model, and the wording of a category name changes, you could handle that as follows.

class ProductTemplate < Fragment
  needs_record_id :type => 'Product'

  subscribe_to 'ProductCategory' do
    def update_product_category_successful(product_category)
      touch_fragments_for_record(product_category.products)
    end
  end
end

The effect of this will be to expire the product template fragment for every product contained within the affected category.

When implementing the method definitions within the subscribe_to block, note that the block will be executed against an instance of a separate Fragmentary::Subscriber class that acts on behalf of the particular Fragment subclass we are defining (so the methods we define within the block are actually defined on that subscriber object). However, all other methods called on the Subscriber object, including those called from within the methods we define in the block, are delegated by method_missing to the fragment subclass. So in the example, touch_fragments_for_record called from within update_product_category_successful represents ProductTemplate.touch_fragments_for_record. This method is defined by the Fragment class, so is available to all fragment subclasses.

Note also that for fragment subclasses that declare needs_record_id, there is no need to define a destroy_<model_name>_successful method simply to remove a fragment whose record_id matches the id of a application record that is destroyed, e.g. to destroy a ProductTemplate whose record_id matches the id of a destroyed Product object. Fragmentary handles this clean-up automatically. This is not to say, however, that 'destroy' handlers are never needed at all. The destruction of an application data record will often require other fragments to be touched.

View Setup

Root Fragments

A 'root' fragment is one that has no parent. In the template in which a root fragment is to appear, define the content to be cached using the cache_fragment helper:

<% cache_fragment :type => 'ProductTemplate', :record_id => @product.id do |fragment| %>
  <%#  Content to be cached goes here  %>
  ...
<% end %>

cache_fragment takes a hash of options needed to uniquely identify the fragment. The inclusion of the :record_id option is only necessary if we have defined ProductTemplate with needs_record_id. The cache_fragment helper will retrieve an existing fragment record from the database based on the options provided and internally pass that record to Rails' native cache method, or if no matching record yet exists it will create one.

Nested Fragments

The variable fragment that is yielded to the block above is an object of class Fragmentary::FragmentsHelper::CacheBuilder, which contains both the actual ActiveRecord fragment record found or created by cache_fragment and the current template as instance variables. The class has one public instance method defined, cache_child, which can be used to define a child fragment nested within the first. The method is used within a block of content defined by cache_fragment and much like cache_fragment it takes a hash of options that uniquely identify the child fragment. Also like cache_fragment it yields another CacheBuilder object and wraps a block containing the content of the child fragment to be cached.

<% cache_fragment :type => 'ProductTemplate', :record_id => @product.id do |fragment| %>
  <% fragment.cache_child :type => 'StoresAvailable' do |child_fragment| %>
    <%#  Content of the nested fragment to be cached goes here  %>
    ...
  <% end %>
<% end %>

Within the body of the child fragment you can continue to define further nested fragments using child_fragment.cache_child etc, as long as appropriate fragment subclasses are defined.

Internally, the main difference between CacheBuilder#cache_child and cache_fragment is that in the former, if an existing fragment matching the options provided is not found and a new record needs to be created, the method will automatically set the parent_id attribute of the new child fragment to the id of its parent. This makes it possible for future changes to the child fragment's updated_at attribute to trigger similar updates to its parent.

Also note that if the parent fragment's class has been defined with needs_record_id but the child fragment's class has not, cache_child will automatically copy the parent's record_id to the child, i.e. the record_id propagates down the nesting tree until it reaches a fragment whose class declares needs_record_id, at which point it must be provided explicitly in the call to cache_child.

Lists

Special consideration is needed in the case of fragments that represent lists of variable length. Specifically, we need to be able to properly handle the creation of new list items. Suppose for example, we have a list of stores in which a product is available and a new store needs to be added to the list. In terms of application data, the availability of a product in a particular store might be represented by a ProductStore join model. Adding a new store involves creating a new record of this class. This record effectively acts as a 'list membership' association.

There are two fragment types involved in caching the list, one for the list as a whole and another for individual items within it. We might represent the list with a fragment of type StoresAvailable and individual stores in the list with fragments of type AvailableStore, each of those a child of the list fragment. Both of these fragment types are associated via their respective record_id attributes with corresponding application data records. The StoresAvailable list fragment is associated with a Product record (the product whose availability we are displaying) and each AvailableStore fragment is associated with a list membership ProductStore record.

We need to ensure that when a new ProductStore record is created, the corresponding AvailableStore fragment is also created and that the containing StoresAvailable fragment is updated as well. We can't simply rely on the creation of the AvailableStore to automatically trigger an update to StoresAvailable in the way we do when we update a child fragment, because the former doesn't exist until the latter is refreshed.

To address this situation, as a convenience Fragmentary provides a class method acts_as_list_fragment that is used when defining the fragment class for the list as a whole, e.g.

class StoresAvailable < Fragment
  acts_as_list_fragment :members => :product_stores, :list_record => :product
end

acts_as_list_fragment takes two named arguments:

  • members is the name of the list membership association in snake case, or tableized, form. The creation of an association of that type triggers the addition of a new item to the list. In the example, it is the creation of a new product_store that results in a new item being added to the list. The effect of declaring acts_as_list_fragment is to ensure that when that membership association is created, the list fragment it is being added to is touched, expiring the cache so that on the next request the list will be re-rendered, which has the effect of creating the required new AvailableStore fragment. Note that the value of the members argument should match the record type of the list item fragments. So in the example, the AvailableStore class should be defined with needs_record_id, :type => ProductStore (We recognize there's some implied redundancy here that could be problematic; some adjustment may be made in the future).
  • list_record is either the name of a method (represented by a symbol) or a Proc that defines how to obtain the record_id (or the record itself; either will work) associated with the list fragment from a given membership association. If the value is a method name, the list record is found by calling that method on the membership association. In the example, the membership association is a ProductStore record, say product_store. The list is represented by a StoresAvailable fragment whose record_id points to a Product record. We can get that Product record simply by calling product_store.product, so the list_record parameter passed to acts_as_list_fragment is just the method :product (:product_id would also work). However sometimes a simple method like this is insufficient and a Proc may be used instead. In this case the newly created membership association is passed as a parameter to the Proc and we can implement whatever functional relationship is necessary to obtain the list record. In this simple example, if we wanted (for no good reason) to use a Proc, it would look like ->(product_store){product_store.product}.

Note that in the example, the specified list_record method, :product, returns a single record for a given product_store membership association. However this isn't necessarily always the case. The method or Proc may also return an array of records or record_ids.

Another important consideration is that if the list fragment, a StoresAvailable fragment in the example, is nested within another, it must actually exist in order for the update to work. If your template is written in such a way that the list fragment is only generated if there are list items to be displayed, say like this,

<% cache_fragment :type => 'ProductTemplate', :record_id => @product.id do |product_fragment| %>
  <% if @product.product_stores.any? %>
    <% product_fragment.cache_child :type => 'StoresAvailable' do |stores_fragment| %>
      <% fragment.cache_child :type => 'AvailableStore', :record_id => store.id do |store_fragment| %>
        <%# ... %>
      <% end %>
    <% end %>
  <% end %>
<% end %>

then creating an initial product_store association won't add the initial item since the StoresAvailable list fragment doesn't exist. Instead you should do something like the following.

<% cache_fragment :type => 'ProductTemplate', :record_id => @product.id do |product_fragment| %>
  <% product_fragment.cache_child :type => 'StoresAvailable' do |stores_fragment| %>
    <% @product.product_stores.each do |store| %>
      <% fragment.cache_child :type => 'AvailableStore', :record_id => store.id do |store_fragment| %>
        <%# ... %>
      <% end %>
    <% end %>
  <% end %>
<% end %>

It is also worth noting that you are completely free to create lists without using acts_as_list_fragment at all. You just have to make sure that you explicitly provide a handler yourself in a subscribe_to block to touch any affected list fragments when a new membership association is created. In the example this would look like the following.

class StoresAvailable < Fragment
  subscribe_to 'ProductStore' do
    def create_product_store_successful(product_store)
      touch_fragments_for_record(product_store.product)
    end
  end
end

In fact there can be cases where it is actually necessary to take an explicit approach like this. Using acts_as_list_fragment assumes that we can identify the list fragments to be touched by identifying their associated record_ids (this is the point of the method's list_record parameter). However, we have seen a situation where the set of list fragments that needed to be touched required a complex inner join between the fragments table and multiple application data tables, and this produced the list fragments to be touched directly rather than a set of associated record_ids.

Accessing Fragments by Arbitrary Application Attributes

The examples above involved fragments associated with specific application data records via the record_id attribute. The fragments were uniquely identified by the fragment type, the parent_id if the fragment is nested, and the record_id.

Of course it is not always necessary to have a record_id, for example in the case of an index page. On the other hand, there are also cases where something other than a record_id is needed to uniquely identify a fragment. Suppose, for example, you have a fragment that renders a set of books published in a certain year, a set of restaurants in a certain postcode, or a set of individuals (say sporting event competitors) in a particular age-group category. In cases like these we need to be able to uniquely identify the fragment again by its type and its parent_id (if it is nested), but also by some other parameter, which we refer to as a 'key' (the terminology may be a little unfortunate; this is not the same thing as a cache key).

To facilitate this, the fragment model includes a key attribute that can be customized on a per-class basis using the needs_key method. For example,

class BooksInYear < Fragment
  needs_key :year_of_publication
end

With this definition we can define the cached content like this:

<% (first_year..last_year).each do |year| %>
  <% cache_fragment :type => 'BooksInYear', :year_of_publication => year do |fragment| %>
    <%# ... %>
  <% end %>
<% end %>

Internally, the key attribute is a string, but the value of the custom option passed to either cache_fragment or cache_child can be a string or anything that responds to to_s. Declaring needs_key also creates an instance method on your class with the same name as the key, so if you need to access the value of the key from the fragment, instead of writing fragment.key you could for example write fragment.year_of_publication.

User Specific Content

Customizing Based on User Type

In the context of a website that identifies users by some form of authentication, often the content served by the site needs to be different for different groups of users. A common example is the case where administrative users see special privileged information or capabilities on a page that are not available to regular users. Similarly, a signed-in user may see a different version of a page from a user who is not signed in.

In the context of caching, this means that separate versions of some content may need to be stored in the cache for each distinct group of users. Fragmentary supports this by means of the user_type attribute on the Fragment model. This is a string identifying which type of user the corresponding cached content is intended for. Typical examples would be "admin", "signed_in" and "signed_out".

The fragment subclass definition for a fragment representing content that needs to be customized by user type should include the needs_user_type declaration:

class MyFragment < Fragment
  needs_user_type
  ...
end

For a fragment subclass like this, when you define an individual fragment in the view, Fragmentary needs to be able to initially create and then subsequently retrieve the fragment record using the correct user_type value. One way to do this is to pass a user type to either the cache_fragment or cache_child method explicitly. This presumes that in the case of authenticated users, a user object is available in the template and that you can derive a corresponding user_type string for that object. For example, if you use the Devise gem to provide authentication, by default an authenticated user is represented by the method current_user. Since the method returns nil if the user is not signed in, the user_type passed to cache_fragment could be provided as follows (assuming here that a method is_an_admin? is defined for your user model),

<% user_type = current_user ? (current_user.is_an_admin? ? "admin" : "signed_in") : "signed_out" %>
<% cache_fragment :type => 'MyFragment', :user_type => user_type do %>
  ...
<% end %>

This will store the user_type string in the fragment record when it is first created and then use that string to retrieve the fragment record on subsequent requests.

However specifying the user_type explicitly like this every time you insert a fragment is actually not necessary. Fragmentary allows you to pre-configure both the method used in your template to obtain the user object and the method used to map that object to the user_type string. That string will then be automatically inserted into the fragment specification implicitly whenever you use cache_fragment or cache_child with a fragment type that references a class defined with needs_user_type. So in the template you don't need to provide any additional parameters:

<% cache_fragment :type => 'MyFragment' do %>
  ...
<% end %>

To configure Fragmentary to use this approach, create a file, say fragmentary.rb, in config/initializers and add something like the following:

Fragmentary.setup do |config|
  config.current_user_method = :current_user
  config.default_user_type_mapping = ->(user) {
    user ? (user.is_an_admin? ? "admin" : "signed_in") : "signed_out"
  }
end

The variable config yielded to the setup block above represents Fragmentary.config which is an instance of class Fragmentary::Config. The value assigned to config.current_user_method is a symbol representing the method Fragmentary will call on the current template to obtain the current user object. It is up to you to ensure that the method exists for your application. The default value is :current_user (so you wouldn't actually need to specify the value used in the example above).

The value assigned to config.default_user_type_mapping is a Proc to which Fragmentary will pass whatever user object is returned by config.current_user_method. When called, the method returns the required user_type string.

As well as configuring a default_user_type_mapping as shown above, it is also possible to specify a mapping on a per-class basis when declaring needs_user_type:

class MyFragment < Fragment
  needs_user_type
    :user_type_mapping => -> (user) {
      user ? (user.paid_subscriber? ? "paid" : "unpaid") : nil
    }
end

Whether you specify the user type mapping in Fragmentary's configuration or in a class declaration, declaring needs_user_type results in a class method MyFragment.user_type(user) being added to your fragment subclass.

Per-User Customization

While storing different versions of fragments is a workable solution for different groups of users, sometimes content needs to be customized for individual users. It is possible to accomplish this by defining a fragment subclass with the declaration needs_user_id:

class MyFragment < Fragment
  needs_user_id
  ...
end

You will also need to include a user_id column in the migration used to create your fragments database table. The needs_user_id declaration will result in separate fragments being created for each individual user. As in the case of needs_user_type, the user_id parameter is automatically extracted from your configured (or default) current_user_method.

However this approach is not generally recommended as doing so at any realistic scale is likely to introduce significant cost and performance challenges. Instead, we provide a way for user-specific content (in general any content) to be inserted into a cached fragment after it has been retrieved from the cache through the use of a special placeholder string in a view template. To accomplish this, two classes are defined, Fragmentary::Widget and a subclass Fragmentary::UserWidget. Instances of these classes represent chunks of content that will be inserted into a fragment after it has been either rendered and stored or retrieved from the cache. For each type of insertable content you wish your application to support, define a specific subclass of one of these classes (UserWidget is preferred if the content is to be user-specific) with two instance methods:

  • pattern, which returns a Regexp that will be used to detect a placeholder you will use in your template at the point where you wish the inserted content to be placed. The placeholder consists of a string matching the regular expression you specify wrapped in special delimiter characters %{...}.
  • content, which returns the string that will be inserted in place of the placeholder.

To give a very simple and contrived example, you could define a widget as:

class SoupWidget < Fragmentary::Widget
  def pattern
    Regexp.new('random_soup')
  end

  def content
    ["vegetable", "chicken noodle", "French onion", "beef lentil",
     "cream of mushroom", "minestrone, "Thai coconut chicken", "pumpkin"].sample
  end
end

and in your template insert that widget content by writing,

<% cache_fragment :type => 'Menu' do |fragment| %>
  ...
  <p>The soup of the day is <%= "%{random_soup}" %></p>
  ...
<% end %>

The point to note is that the specific soup in the example is not stored in the cache; it is inserted only after the cached content is retrieved by cache_fragment (or alternatively cache_child). Once you've defined the Widget subclass, inserting the placeholder into your template is all you have to do to make use of the widget. Fragmentary takes care of detecting the placeholder and inserting the widget's non-cached content into the content retrieved from the cache.

You can also pass variable data into your widget by including parenthesized capture groups within the widget's regular expression pattern. For example, with a pattern like this,

class MenuWidget < Fragmentary::Widget
  def pattern
    Regexp.new('menu_for_(\w*)')
  end
end

you could specify a placeholder,

<%= "%{menu_for_tuesday}" %>

The Ruby MatchData object that results from matching the placeholder in your template against the widget's pattern is available via the read accessor Fragmentary::Widget#match. Any substrings matched against parenthesized capture groups are accessible as match[1], match[2]... etc. So in the example, the string "tuesday" would be available within your content method as match[1].

Internally, the widget is instantiated with the current template as an instance variable and you can access it within your content method via a read accessor, template. This means that within content you can use any methods normally available within an Action View template simply by calling them on the template object explicitly. So in the last example, the menu content could be inserted by rendering a partial:

class MenuWidget < Fragmentary::Widget
  def pattern
    Regexp.new('menu_for_(\w*)')
  end

  def content
    template.render 'menu_for', :day => match[1]
  end
end

With all of the foregoing, we can now see how to insert user-specific content. For example we could define a widget like this one,

class WelcomeMessage < Fragmentary::UserWidget
  def pattern
    Regexp('welcome_message')
  end

  def content
    "Welcome #{current_user.display_name}, member since #{current_user.membership_year}"
  end
end

and insert it into a template with,

<p><%= "%{welcome_message}" %></p>

Notice that in this case we subclassed UserWidget instead of Widget. The only differences between the two are that:

  • as a convenience, UserWidget provides a read accessor current_user to save you having to write template.send(Fragmentary.config.current_user_method).
  • if current_user is nil because no user is signed in, content will return an empty string.

The 'memo' Attribute and Conditional Content

With the exception of updated_at, the attributes of the fragment model we have discussed, type, record_id, parent_id, key, and user_type, are all designed for one purpose: to uniquely identify a fragment record. Occasionally, however, we may need to attach additional information to a fragment record, e.g. regarding the content contained within the fragment.

Suppose you wish the inclusion of a piece of cached content in your template to be conditional on the result of some computation that occurs within the process of rendering of that content. This could arise, for example, if database records needed for the fragment are retrieved within the body of the fragment (in order to avoid having to access the database when a cached version of the fragment is available), and the fragment is either shown or hidden selectively, depending on whether some variable derived from that data matches another parameter, say a user input.

Of course if you have to render the content in order to determine whether it needs to be included, you defeat the purpose of caching. However, the fragment memo attribute included in the database migration suggested earlier can be used to address this. The process involves three parts:

  • wrap the cache_fragment or cache_child method that defines the fragment content with a Rails capture block and assign the result to a local variable.
  • Inside the block passed to cache_fragment or cache_child, update the fragment's memo attribute with the data that will be used to determine whether the fragment will be displayed.
  • After calling the capture helper, test the fragment's memo attribute and conditionally insert the 'captured' content into the template.
<% conditional_content = capture do %>
  <% cache_fragment :type => 'ProductTemplate', :record_id => @product.id do |fragment| %>
    <%#  Content to be cached goes here  %>
    <%#  ...                             %>
    <% fragment.update_attribute(:memo, computed_memo_value) %>
  <% end %>
<% end %>

<% if Fragment.root(:type => 'ProductTemplate', :record_id => @product.id).memo == required_memo_value %>
  <%= conditional_content %>
<% end %>

Note that we update the fragment's memo attribute by calling update_attribute on the fragment object. Although this is actually a CacheBuilder object, that class passes any methods such as update_attribute to the underlying fragment.

Also note the use above of the method Fragment.root to retrieve the fragment matching the supplied parameters after caching has occurred. We have to retrieve the fragment explicitly since the variable fragment is local to the cache_fragment block to which it is yielded. The method takes the same parameters as cache_fragment with one exception: At present, if your fragment class declares needs_user_type or needs_user_id, you need to explicitly pass either the respective parameter or the current user object (this may change in a future release). e.g.

...
<% root_fragment = Fragment.root(:type => 'ProductTemplate', :record_id => @product.id, :user => current_user).memo == required_memo_value %>
...

The process for child fragments is similar, except that an instance method Fragment#child can be called on the parent fragment in order to retrieve the fragment to be tested. e.g.

<% conditional_content = capture do %>
  <% parent_fragment.cache_child :type => 'StoresAvailable' do |fragment| %>
    <%#  Content to be cached goes here  %>
    <%#  ...                             %>
    <% fragment.update_attribute(:memo, computed_memo_value) %>
  <% end %>
<% end %>

<% if parent_fragment.child(:type => 'StoresAvailable').memo == required_memo_value %>
  <%= conditional_content %>
<% end %>

In both of these examples, the fragment to be tested is retrieved after caching. If relying on side-effects doesn't make you too queasy, you can alternatively retrieve it before caching, in which case the retrieved fragment can be passed to cache_fragment or cache_child as appropriate instead of the set of fragment attributes we used before. This avoids the need for cache_fragment or cache_child to retrieve the fragment themselves internally, e.g.

<% root_fragment = Fragment.root(:type => 'ProductTemplate', :record_id => @product.id) %>
<% conditional_content = capture do  %>
  <% cache_fragment :fragment => root_fragment do |fragment| %>
    <%#  Content to be cached goes here  %>
    <%#  ...                             %>
    <% fragment.update_attribute(:memo, computed_memo_value) %>
  <% end %>
<% end %>

<% if root_fragment.memo == required_memo_value %>
  <%= conditional_content %>
<% end %>

Note that both Fragment.root and Fragment#child will instantiate a matching fragment if one doesn't already exist. If you prefer to handle the case where a matching fragment doesn't already exist separately, you can instead use Fragment.existing or Fragment#existing_child, which return nil if a matching fragment isn't found rather than instantiating one, e.g.

<% if child = parent_fragment.existing_child(:type => 'StoresAvailable') %>

  <% conditional_content = capture do %>
    <% parent_fragment.cache_child(:child => child) do |fragment| %>
      ...
    <% end %>
  <% end %>

  <% if child.memo == required_memo_value %>
    <%= conditional_content %>
  <% end %>

<% else %>
  <%# handle this case differently %>
  ...
<% end %>

Updating Cached Content Automatically

Internal Application Requests

When application data that a cached fragment depends upon changes, i.e. as a result of a POST, PATCH or DELETE request, the subscribe_to declarations in your Fragment subclass definitions ensure that the updated_at attribute of any existing fragment records affected will be updated. Then on subsequent browser requests, the cached content itself will be refreshed.

As part of this process, new fragment records may be created as children of an existing fragment if the existing fragment contains newly added list items. A new root fragment, on the other hand, will be created for any root fragment class defined with needs_record_id if an application data record of the associated record_type is created, as soon as a browser request is made for the new content (often a new page) containing that fragment.

Sometimes, however, it is desirable to avoid having to wait for an explicit request from a user in order to update the cache or to create new cached content. To deal with this, Fragmentary provides a mechanism to automatically create or refresh cached content preemptively, essentially as soon as a change in application data occurs. Rails' ActionDispatch::Integration::Session class provides an interface that allows requests to be sent directly to the application programmatically, without generating any external network traffic. Fragmentary uses this interface to automatically send requests needed to update the cache whenever changes in application data occur.

Creating these requests is handled slightly differently depending on whether they are designed to update content associated with an existing root fragment or to create content for a new root fragment that does not yet exist. In the case of an existing root fragment, if the fragment's class has a request_path instance method defined, a request will be sent to the application at that path (represented in the form of a string) whenever touch is called on the fragment record (in general, the request can be suppressed by passing :no_request => true if required). You simply need to define the request_path method in any individual Fragment subclass that you wish to generate requests for. For example:

class ProductTemplate < Fragment
  needs_record_id :type => 'Product'

  def request_path
    "/product/#{record_id}"
  end
end

Since nested child fragments automatically touch their parent fragments when they themselves are updated, internal requests can be initiated by an update to application data that affects a fragment anywhere within a page. Only the root fragment needs to define the request path.

The request that is generated will be sent to the application at the request_path specified, but it may also include additional request parameters and options. To send HTTP request parameters with the request, the Fragment subclass should define an additional instance method request_parameters that returns a hash of named parameters. To send an XMLHttpRequest, a class should define an instance method request_options that returns the hash {:xhr => true}.

In the case of a root fragment that does not yet exist, i.e. for a Fragment subclass defined with needs_record_id and an associated record_type when a new application record of that type is first created, a request will be generated in order to create the new fragment automatically if the subclass has a class method request_path defined that takes the id of the newly created application record and returns a string representing the path to which the request should be sent. For example:

class ProductTemplate < Fragment
  needs_record_id :type => 'Product'

  def self.request_path(record_id)
    "/product/#{record_id}"
  end

  def request_path
    self.class.request_path(record_id)
  end

So in this example, any time a new Product record is created, Fragmentary will send a request to the page for that new product, resulting in the corresponding ProductTemplate fragment being created. As shown here, in cases like this we generally choose to define the instance method request_path in terms of the corresponding class method.

Request Queues

A single external HTTP POST, PATCH or DELETE request can cause changes to application data affecting multiple fragments on multiple pages. One external request can therefore lead to multiple internal application requests. In addition, considering that different versions of some cached fragments exist for different user types, in order to ensure that all affected versions get refreshed we may need to send each internal request multiple times in the context of several different user sessions, each representing a different user type.

To achieve this, during handling of the initial external request in which changes in application data propagate to affected fragment records via the subscribe_to declarations in your fragment class definitions, multiple instances of class Fragmentary::RequestQueue, each corresponding to a different user type, are used to store a collection of Fragmentary::Request objects representing the internal requests generated during that process.

The set of user types that an individual Fragment subclass is expected to support is available via a configurable class method user_types, which returns an array of user type strings (Note this is different from the class method user_type discussed earlier that returns a single user type for a specific current user).

Configuration of the user_types method is discussed in the next section. Fragmentary uses the types returned by this method to identify the request queues that each internal application request needs to be added to when instances of each particular Fragment subclass are updated. Fragment subclasses inherit both class and instance methods request_queues that return a hash of queues keyed by each of the specific user type strings that that specific subclass supports.

The request queue for any given user type is shared across all Fragment subclasses within the application. So for example, a ProductTemplate fragment may have two different request queues, ProductTemplate.request_queues["admin"] and ProductTemplate.request_queues["signed_in"]. If a StoreTemplate fragment has one request queue StoreTemplate.request_queues["signed_in"], the "signed_in" queues for both classes represent the same object.

Configuring Internal Request Users

The user types each subclass supports can be configured in two ways. If all of the fragments in your application that declare need_user_type always use the same set of user types, you can configure these by setting Fragmentary.config.session_users in your initializers/fragementary.rb file:

Fragmentary.setup do |config|
  ...
  config.session_users = {
    'signed_in' => {:credentials => {:user => {:email => '[email protected]', :password => 'bobs_secret'},
    'admin' => {:credentials => {:user => {:email => '[email protected]', :password => 'alices_secret'}
  }
  config.get_sign_in_path = '/users/sign_in'
  config.post_sign_in_path = '/users/sign_in'
  config.sign_out_path = '/users/sign_out'
end

The value assigned to session_users is a hash whose keys are each a required user_type string, with values containing a hash of credentials needed to sign in as a stereotypical user for that type, i.e. the HTTP request parameters that need to be sent with a POST request to sign in. If authentication is not required, the value should be an empty hash. If you prefer not to put credentials in the configuration file directly, you can alternatively specify a Proc that returns the required hash; the Proc will be executed when sign-in actually occurs. We have used this, for example, to retrieve randomized single-use credentials from our User model for specific test users that have only read-access to the site.

As shown in the example, we also need to assign sign-in and sign-out paths, with separate sign-in paths for GET and POST requests. The GET path is the address from which a user would retrieve the sign-in form. Fragmentary sends a GET request to this address first in order to retrieve the CSRF token that needs to be submitted with the user credentials in a POST request when signing in.

The configuration defined using Fragmentary.setup as shown above will be used as the default for any fragment subclass defined using needs_user_type; for those subclasses the class method user_types will return an array of the user_type strings so defined. However, if you need a specific subclass to support a different set of user types, you can configure that by passing additional options to needs_user_type in the class definition.

class MyFragment < Fragment
  needs_user_type
    :session_users => ['admin',
                       'paid' => {:credentials => ...},
                       'unpaid' => {:credentials => ...}]
end

The value of the :session_users option (you can also use :user_types or just :types instead) is an array containing either strings of user types that are defined elsewhere, e.g. using Fragmentary.setup (e.g. 'admin' in the example) or hashes of new session user definitions that include their required sign-in credentials.

Sending Queued Requests

A class method Fragmentary::RequestQueue.all returns an array of all request queues the application uses. The requests stored within a given RequestQueue can be sent to the application by calling the instance method RequestQueue#send. Calling send instantiates a Fragmentary::UserSession object representing a browser session for the particular type of user the queue is handling. For sessions representing user types that need to be authenticated, instantiating the UserSession will sign in to the application using the credentials configured for the particular user_type.

To send all requests once processing of each external browser request has been completed, add a method such as the following to your ApplicationController class and call it using a controller after_filter, e.g. for create, update and destroy actions:

def send_queued_requests
  delay = 0.seconds
  Fragmentary::RequestQueue.all.each{|q| q.send(:delay => delay += 10.seconds)}
end

The send method takes two optional named arguments, delay and between. If neither are present, all requests held in the queue are sent immediately. If either are present, sending of requests is off-loaded to an asynchronous process using the Delayed::Job gem (i.e. we are not currently using Active Job) and scheduled according to the parameters provided: delay represents the delay before the queue begins sending requests and between represents the interval between individual requests in the queue being sent. In the example above, we choose to delay the sending of requests from each queue by 10 seconds each. You may customize as appropriate.

Queuing Requests Explicitly

In the cases we've described so far, internal requests are added to request queues automatically, either as a result of updating an existing root fragment or because a new application record has been created that requires a new root fragment with a class declaration needs_record_id to be created. There are, however, sometimes cases in which you may wish to create Fragmentary::Request objects yourself and add them to the appropriate queues explicitly. This could arise, for example, with a fragment class that declares needs_key. If a change in application data results in a value of the 'key' that didn't previously exist, you may wish, in a subscribe_to block within your fragment subclass definition, to create an internal request explicitly that will result in new cached content being generated for that new 'key' value.

The process is straightforward. First instantiate a request by calling Fragmentary::Request.new. This method takes two required parameters and two optional parameters:

request = Fragmentary::Request.new(method, path, parameters, options)

method is a string, usually 'get', but in general 'post', 'patch', 'put' and 'delete' are also acceptable values.

path is a string representing the path you wish the request to be sent to.

parameters is an optional hash containing named HTTP request parameters to be sent with the request.

options is an optional hash {:xhr => true} if the request is to be sent as an XMLHttpRequest.

Once the request is instantiated, you can add it to all of the queues required by the fragment's subclass simply by calling the class method queue_request:

MyFragment.queue_request(request)

Asynchronous Fragment Updating

Off-loading internal application requests to an asynchronous process as noted in the previous section means they can occur without delaying the server's response to the user's initial external request. However, in a complex application there may also be some overhead just in the process of updating fragment records that is not critical in order to send a response back to the user. For example, some application data changes may require a large number of fragments to be updated on pages other than the one the user is going to be sent or redirected to right away. In this case, we may wish to offload the task of updating these fragments to an asynchronous process as well.

Fragmentary accomplishes this by means of the Fragmentary::Handler class. An individual task you wish to be handled asynchronously is created by defining a subclass of Fragmentary::Handler, typically within the scope of the Fragment subclass in which you wish to use it. The Handler subclass you define requires a single instance method call, which defines the task to be performed.

To use the handler, typically within a fragment's subscribe_to declaration, call the inherited class method create, passing it a hash of named arguments. Those arguments can then be accessed within your call method as the instance variable @args. Consider our earlier example in which each product template contains a list of stores in which the product is available. Suppose an administrator fixes a typographical error in the name of a store. That correction needs to propagate to every AvailableStore fragment for every product that is available in that store. We can do this using a Handler as follows:

class AvailableStore < Fragment
  needs_record_id :type => 'ProductStore'

  class UpdateStoreHandler < Fragmentary::Handler
    def call
      product_stores = ProductStore.where(:store_id => @args[:id])
      touch_fragments_for_record(product_stores)
    end
  end

  subscribe_to 'Store' do
    def update_store_successful(store)
      UpdateStoreHandler.create(store.attributes.symbolize_keys)
    end
  end
end

While a Handler subclass definition defines a task to be run asynchronously, and calling the class method create within a subscribe_to block causes the task to be instantiated, we still have to explicitly dispatch the task to an asynchronous process, which we again (currently) do using Delayed::Job.

To facilitate this we use the Fragmentary::Dispatcher class. A dispatcher object is instantiated with an array containing all existing handlers, retrieved using the class method Fragmentary::Handler.all.

dispatcher = Fragmentary::Dispatcher.new(Fragmentary::Handler.all)

The Dispatcher class defines an instance method perform, which invokes call on all the handlers provided when the dispatcher is instantiated. Delayed::Job uses the perform method to define the job to be run asynchronously, which is accomplished by passing the dispatcher object to Delayed::Job.enqueue:

Delayed::Job.enqueue(dispatcher)

As in the case of our queued background requests, we enqueue the dispatcher in an ApplicationController after_filter. In fact we can combine both the sending of background requests and the asynchronous fragment updating task in one method:

class ApplicationController < ActionController::Base

  after_filter :handle_asynchronous_tasks

  def handle_asynchronous_tasks
    send_queued_requests  # as defined earlier
    Delayed::Job.enqueue(Fragmentary::Dispatcher.new(Fragmentary::Handler.all))
    Fragmentary::Handler.clear
  end

end

Note that updating fragments in an asynchronous process like this will itself generate internal application requests beyond those generated in the course of handling the user's initial request. The Dispatcher takes care of sending these additional requests.

Updating Lists Asynchronously

As discussed earlier, if a fragment class is declared with acts_as_list_fragment, fragments of that class will be automatically touched whenever a new list item is created. If you want this process to take place asynchronously, simply pass the option :delay => true to acts_as_list_fragment, e.g.:

class StoresAvailable < Fragment
  acts_as_list_fragment :members => :product_stores, :list_record => :product, :delay => true
end

Updating Fragments Explicitly Within a Controller

The typical usage scenario for Fragmentary is for fragment records to be updated by the methods defined in the subscribe_to blocks you create in your fragment subclasses, in response to user requests that modify the application data. This involves very little coupling between your application's controllers and models and the fragment caching process. Models need only include the Fragmentary::Publisher module to ensure that Fragment subclasses can subscribe to the application data events that trigger fragment updates. Your ApplicationController only needs to ensure that any queued internal requests and asynchronous fragment update tasks are dispatched (if you choose to use either of those features) before sending a response to the user's browser. Application models and controllers generally have no need to access the Fragment class API.

Ocassionally, however, a situation may arise in which it is useful to touch fragments directly from a controller. For example, an application data event may occur that requires fragments to be updated on a large number of different pages within your site. This is the kind of scenario in which you might choose to offload fragment updating to a Fragmentary::Handler and execute the task asynchronously as described in the previous section. However, it may also be that one of those pages is the one that is going to be sent back to the user's browser immediately in response to the current request. You can't put off touching the affected fragment on that one page until the asynchronous process executes because then the user won't see the updated content. In a scenario like this, you also can't detect within the subscribe_to block in your fragment subclass definition which one fragment needs to be touched immediately rather than asynchronously, since the fragment class has no knowledge of the internal state of the controller.

A possible solution in this situation may be to touch the fragment representing the content that is about to be returned to the user in the controller. We're not convinced that this is good practice, but if you decide you need to, you can use the class method Fragment.existing to obtain the required fragment in the controller and call touch on it explicitly. If the fragment subclass involved is one that supports multiple user types, you will want to ensure that just the fragment corresponding to the current user's user_type is touched, which you can do by passing the current user object to existing. Internally, existing will map the user object to the corresponding user_type using the user_type_mapping you have configured in order to select the correct fragment. In addition, when you touch the fragment in this context, you should pass :no_request => true so that it doesn't generate an internal request to update the cache since responding to the user's actual request is going to accomplish this automatically.

class ProductStoresController < ApplicationController
  def create
    ...
    @product_store.save
    fragment = ProductTemplate.existing(:record_id => @product_store.product_id, :user => current_user)
    fragment.touch(:no_request => true)
  end
end

Removing Queued Requests in the Controller

Another scenario that occasionally arises is when internal requests created in the course of executing methods defined in a subscribe_to block include the path to the same page that the controller is going to render anyway in the normal course of responding to the user's current browser request. In effect, that one internal request is redundant. Although it generally won't cause any real harm to send a redundant request (remember the internal request is usually dispatched asynchronously, so it won't interfere with the synchronous response), redundancy is redundancy and you may wish to avoid it.

In this case, it is possible within the controller to remove a request that has already been queued, after the application data has been created, updated or destroyed, by using the class method Fragment.remove_queued_request. The method takes two named parameters, the path of the request to be removed and the current user object, with the latter allowing it to remove the request from the correct queue, e.g.

class ProductsController < ApplicationController
  def create
    ...
    if @product.save
      respond_to do |format|
        format.html do
          ProductTemplate.remove_queued_request(:request_path => "/products/#{@product.id}", :user => current_user)
          redirect_to @product
        end
        format.js
      end
    else
      ...
    end
  end
end

Handling AJAX Requests

Providing a Receiver for Calls to 'cache_child'

There are a couple of special issues to consider when using Fragmentary with templates designed to respond to partial page requests. An example would be a Javascript template used to insert content into a previously loaded page, such as when dynamically adding a new item to an existing list. The typical approach in this scenario is to use embedded Ruby (ERB) in the Javascript template to render a partial containing the required content, escape the resulting string for Javascript and insert it into the list using jQuery's append method, e.g.

$('ul.product_list').append('<%= j(render 'product/summary', :product => @product) %>')

However, it's possible for the partial to contain a cached fragment that happens to be a child of a parent containing the list as a whole. In this case, when the entire page containing the original list is first rendered and the parent fragment is retrieved using say cache_fragment (if it happens to be a root fragment), a CacheBuilder object is yielded to the block that renders the partial, and this object is passed to the partial to act as the receiver for cache_child.

<% cache_fragment :type =>`ProductList` do |parent_fragment| %>
  <ul class='product_list'>
    <% @products.each do |product| %>
      <% render 'product/summary', :product => product, :parent_fragment => parent_fragment %>
    <% end %>
  </ul>
<% end %>

In the Javascript case, however, since the template only renders the new list item and not the list as a whole, there is no cache_fragment method invoked in order to yield the CacheBuilder object, and so we have to construct it explicitly. To do this, Fragmentary provides a helper method fragment_builder that takes an options hash containing the parameters that define the fragment (the same ones passed to cache_fragment) and returns the CacheBuilder object that the partial needs.

<% parent_fragment = fragment_builder(:type => 'ProductList') %>
$('ul.product_list').append('<%= j(render 'product/summary', :product => @product,
                                                             :parent_fragment => parent_fragment) %>')

Inserting Widgets into Child Fragments

The second challenge in dealing with child fragments rendered without the context provided by a call to cache_fragment concerns the insertion of widgets into the generated content. Ordinarily, when a full page is rendered in response to the initial browser request, any widgets that are required are inserted after the root fragment has been either created or retrieved. This occurs within the cache_fragment method. But again, if we are only rendering part of the page containing a child fragment, there is no call to cache_fragment in which this can occur.

To address this, cache_child can take an additional boolean option, :insert_widgets that can be used to force the insertion of widgets into the child. Typically a local variable containing the value of this option would be passed to the partial in which cache_child appears. In the Javascript template:

<% parent_fragment = fragment_builder(:type => 'ProductList') %>
$('ul.product_list').append('<%= j(render 'product/summary', :product => @product,
                                                             :parent_fragment => parent_fragment,
                                                             :insert_widgets => true) %>')

Then in the partial itself, something like this:

<% insert_widget ||= nil %>
<% parent_fragment.cache_child :type => 'ProductSummary', :insert_child => insert_child do |fragment| %>
  <li>
    ...
  </li>
<% end %>

Note that if the partial page content being generated contains several nested child fragments, the :insert_widgets option only needs to be passed to the outer-most call to cache_child.

Suppressing Cache Storage

It is possible to define a fragment without actually storing its content in the cache store. This can be useful, for example if you wish to cache several sibling children within a page but don't need to store the entire root fragment that contains them. Simply include the option :no_cache => true in the hash passed to cache_fragment or cache_child.

Support for Multiple Application Instances

In many practical deployment scenarios it is desirable to maintain a live pre-release version of the application online separate from the public-facing production website. This allows new software releases to be staged for either internal or beta testing prior to final deployment to the production environment. For example, if the public-facing application is accessed at a root URL of http://myapp.com/, a separate pre-release version might be deployed say to http://prerelease.myapp.com/ or http://myapp.com/prerelease/.

In the specific scenario in which Fragmentary was developed, it was important for the pre-release version of the application to share the same application database as the production site, i.e. both production and pre-release versions render exactly the same data, and any changes to application data initiated by a user of one version of the application will be reflected in the content seen by a user of the other version as well. For caching, this leads to some additional challenges which we discuss below.

Storing Multiple Versions of Content in the Cache

There is an important difference between the content rendered by each instance of an application: any HTML links to other pages on the site must be to URLs representing the particular version being rendered. For example, on a 'products' index page, a link to an individual product page on the production website might look like http://myapp.com/products/123, while the same link on the index page of the pre-release version might be http://myapp.com/prerelease/products/123.

In general, as long as links contained in the view are created using Rails' standard path or url helpers, e.g. product_path(@product) etc, they will automatically be based on the root URL of the application instance (production or pre-release) in which they are generated. However, in the context of caching, the fact that differences exist in the content generated by different application instances implies that for each fragment defined in the view by calling cache_fragment or cache_child, different versions of fragment content need to be stored in the cache. Also, each of these distinct versions needs to be associated with a unique record in the fragments database table identifying which application instance generated that content.

In Fragmentary we can accomplish this simply by adding a column to the fragments table to store the root URL of the particular application instance that created the fragment content.

class AddAppUrlToFragment < ActiveRecord::Migration
  def change
    change_table :fragments do |t|
      t.string :application_root_url
    end
  end
end

The column name application_root_url shown above is the default assumed by Fragmentary. You can use a different column name if you wish, as long as you tell Fragmentary in initializers/fragmentary.rb, e.g.:

Fragmentary.setup do |config|
  ...
  config.application_root_url_column = 'app_root'
end

Once this column has been added to the fragments table, any fragment records subsequently created by calls to cache_fragment or cache_child will have the column populated automatically based on the particular application instance in which the code is executed, and the content stored in the cache for each fragment record will be that generated by the corresponding instance. As long as any links rendered within the content are generated using Rails' path or url helpers, the cached content will be rendered correctly both when initially created and when retrieved subsequently.

Note that it is possible to store cached content for each of the different application instances in different places. Simply set config.cache_store in config/environments/production.rb (or alternative environment file) as required.

Automatically Refreshing Multiple Versions of Cached Content

The fact that two versions of content exist in the cache for each fragment means that whenever a change in application data occurs that triggers the generation of an internal application request to refresh a piece of cached content (i.e. for any fragment that has a request_path method defined), both (in general all) versions need to be refreshed. So for example, a change to application data caused by a user action on the production website needs to trigger internal requests to both the production and the pre-release instances in order to refresh their respective content. The converse is true for changes initiated from the pre-release website.

(Note: for better or worse, we've stuck with the term 'internal request' to mean any request delivered to the application programmatically in order to refresh cached content. This includes requests created by one application instance that are intended to be processed by another instance.)

Our approach to sending requests between application instances relies on requests being processed asynchronously (i.e. by passing a delay value to RequestQueue#send in the controller method send_queued_requests discussed earlier). The reason is that asynchronous tasks used to process internal requests can be directed to specific task queues associated with the application instance they are intended to be processed by. Each application instance has an associated asynchronous task process. By configuring that process to run tasks from just the queue(s) designated for it, we ensure that any application instance can direct requests to any other.

As noted earlier, our current implementation relies on Delayed::Job for creating and processing asynchronous tasks. Queued tasks are stored in a database table, so as long as each application and Delayed::Job instance have access to the same database, this approach will be successful.

To configure Fragmentary to automatically refresh cached content for multiple instances, first set remote_urls in Fragmentary.config. This is an array of root URLs for all other instances of the application that requests should be sent to. For example, in order to allow requests to be sent from the production instance to the pre-release instance, in initializers/fragmentary.rb in the production code we would add the following configuration:

Fragmentary.setup do |config|
  ...
  config.remote_urls << 'http://myapp.com/prerelease/'
end

Our current approach to application deployment is to maintain different branches in our source repository for each application instance. This allows us to keep custom configurations like config.remote_urls above for each instance on their own branches. So in contrast to the case above, to allow requests to be sent from the pre-release instance to the production instance, in initializers/fragmentary.rb on the pre-release branch the remote_urls array would be set to ['http://myapp.com/'] .

As an alternative, it may be possible to create a separate environment for the pre-release deployment in your config/environments directory and thus maintain the configuration for all application instances in a single repository branch. We have not investigated this approach.

By specifying remote_urls as described above, for each internal application request Fragmentary will automatically create tasks to process the request not only on the application instance where it was created but also on all other instances corresponding to the elements in remote_urls. Fragmentary sends these tasks to Delayed::Job queues that have names formed from the domain name and path of both the root URL of the current instance and the values in remote_urls. So in the example, a request to '/products/123', created say by editing a product name on the production website, would be sent to Delayed::Job queues with names 'myapp.com' and 'myapp.com/prerelease/'.

The final step needed to allow internal requests to be processed by the instance they are intended for is to configure the Delayed::Job process associated with each instance to process tasks from the correct queue.

In our case we use Capistrano 2 for deployment and use delayed_job_args in config/deploy.rb to configure queue names (see documentation here):

queue_prefix = "myapp.com/prerelease/"
set :delayed_job_args, "--queue=#{queue_prefix},#{queue_prefix}_overnight"
after "deploy:stop",    "delayed_job:stop"
after "deploy:start",   "delayed_job:start"
after "deploy:restart", "delayed_job:restart"

Again, this configuration will be different for each instance and so in our case each one will be stored in different source repository branches.

Note that if you configure Delayed::Job to work only from specific queues, you'll need to make sure that any asynchronous tasks created by your application are submitted to one of those queues.

Dependencies

  • The current implementation of Fragmentary has been tested using Rails 5. It does not work with earlier versions of Rails due to a change in the API for Rails ActionDispatch::Integration::Session class. We do have a 'Rails.4.2' branch that uses the older API in the github repository. However this branch is no longer maintained.
  • As noted, Fragmentary uses the Delayed::Job gem to execute background tasks asynchronously.

Contributing

Bug reports and usage questions are welcome at https://github.com/MarkMT/fragmentary.

Testing

You're welcome to write some tests!!!

License

The gem is available as open source under the terms of the MIT License.