cpoms / innkeeper

Apartment clone with threadsafe multihost support
7 stars 2 forks source link

Innkeeper

Code Climate Build Status

Multitenancy for Rails and ActiveRecord

Innkeeper provides tools to help you deal with multiple tenants in your Rails application. If you need to have certain data sequestered based on account or company, but still allow some data to exist in a common tenant, Innkeeper can help.

Installation

Rails

Add the following to your Gemfile:

gem 'innkeeper'

Then generate your Innkeeper config file using

bundle exec rails generate innkeeper:install

This will create a config/initializers/innkeeper.rb initializer file. Configure as needed using the docs below.

That's all you need to set up the Innkeeper libraries. If you want to switch tenants on a per-user basis, look under "Usage - Switching tenants per request", below.

Usage

Creating new Tenants

Before you can switch to a new innkeeper tenant, you will need to create it. Whenever you need to create a new tenant, you can run the following command:

Innkeeper::Tenant.create('tenant_name')

If you're using PostgreSQL, this will create the database and schema from the derived tenant configuration. For example, if you're using the Schema resolver, this will create a schema named 'tenant_name', assuming you're not using a decorator. If you're customising the tenant name with a decorator, it is the decorated name that will be used.

When you create a new tenant, the schema is loaded on to that tenant, so it will be up to date when create returns.

Switching Tenants

To switch tenants using Innkeeper, use the following command:

Innkeeper::Tenant.switch!('tenant_name')

When switch is called, all requests coming to ActiveRecord will be routed to the tenant you specify (with the exception of excluded models, see below). To return to the 'root' tenant, call switch with no arguments.

Switching Tenants per request

You can have Innkeeper route to the appropriate tenant by adding some Rack middleware. Innkeeper can support many different "Elevators" that can take care of this routing to your data.

NOTE: when switching tenants per-request, keep in mind that the order of your Rack middleware is important. See the Middleware Considerations section for more.

The initializer above will generate the appropriate code for the Subdomain elevator by default. You can see this in config/initializers/innkeeper.rb after running that generator. If you're not using the generator, you can specify your elevator below. Note that in this case you will need to require the elevator manually in your application.rb like so:

# config/application.rb
require 'innkeeper/elevators/subdomain' # or 'domain' or 'generic'

Switch on subdomain

In house, we use the subdomain elevator, which analyzes the subdomain of the request and switches to a tenant schema of the same name. It can be used like so:

# application.rb
module MyApplication
  class Application < Rails::Application
    config.middleware.use 'Innkeeper::Elevators::Subdomain'
  end
end

If you want to exclude a domain, for example if you don't want your application to treat www like a subdomain, in an initializer in your application, you can set the following:

# config/initializers/innkeeper/subdomain_exclusions.rb
Innkeeper::Elevators::Subdomain.excluded_subdomains = ['www']

This functions much in the same way as Innkeeper.excluded_models. This example will prevent switching your tenant when the subdomain is www. Handy for subdomains like: "public", "www", and "admin" :)

Switch on domain

To switch based on full domain (excluding subdomains ie 'www' and top level domains ie '.com' ) use the following:

# application.rb
module MyApplication
  class Application < Rails::Application
    config.middleware.use 'Innkeeper::Elevators::Domain'
  end
end

Switch on full host using a hash

To switch based on full host with a hash to find corresponding tenant name use the following:

# application.rb
module MyApplication
  class Application < Rails::Application
    config.middleware.use 'Innkeeper::Elevators::HostHash', {'example.com' => 'example_tenant'}
  end
end

Custom Elevator

A Generic Elevator exists that allows you to pass a Proc (or anything that responds to call) to the middleware. This Object will be passed in an ActionDispatch::Request object when called for you to do your magic. Innkeeper will use the return value of this proc to switch to the appropriate tenant. Use like so:

# application.rb
module MyApplication
  class Application < Rails::Application
    # Obviously not a contrived example
    config.middleware.use 'Innkeeper::Elevators::Generic', Proc.new { |request| request.host.reverse }
  end
end

Your other option is to subclass the Generic elevator and implement your own switching mechanism. This is exactly how the other elevators work. Look at the subdomain.rb elevator to get an idea of how this should work. Basically all you need to do is subclass the generic elevator and implement your own parse_tenant_name method that will ultimately return the name of the tenant based on the request being made. It could look something like this:

# app/middleware/my_custom_elevator.rb
class MyCustomElevator < Innkeeper::Elevators::Generic

  # @return {String} - The tenant to switch to
  def parse_tenant_name(request)
    # request is an instance of Rack::Request

    # example: look up some tenant from the db based on this request
    tenant_name = SomeModel.from_request(request)

    return tenant_name
  end
end

Middleware Considerations

In the examples above, we show the Innkeeper middleware being appended to the Rack stack with

Rails.application.config.middleware.use 'Innkeeper::Elevators::Subdomain'

By default, the Subdomain middleware switches into a Tenant based on the subdomain at the beginning of the request, and when the request is finished, it switches back to the "public" Tenant. This happens in the Generic elevator, so all elevators that inherit from this elevator will operate as such.

It's also good to note that Innkeeper switches back to the "public" tenant any time an error is raised in your application.

This works okay for simple applications, but it's important to consider that you may want to maintain the "selected" tenant through different parts of the Rack application stack. For example, the Devise gem adds the Warden::Manager middleware at the end of the stack in the examples above, our Innkeeper::Elevators::Subdomain middleware would come after it. Trouble is, Innkeeper resets the selected tenant after the request is finish, so some edirects (e.g. authentication) in Devise will be run in the context of the "public" tenant. The same issue would also effect a gem such as the better_errors gem which inserts a middleware quite early in the Rails middleware stack.

To resolve this issue, consider adding the Innkeeper middleware at a location in the Rack stack that makes sense for your needs, e.g.:

Rails.application.config.middleware.insert_before 'Warden::Manager', 'Innkeeper::Elevators::Subdomain'

Now work done in the Warden middleware is wrapped in the Innkeeper::Tenant.switch context started in the Generic elevator.

Dropping Tenants

To drop tenants using Innkeeper, use the following command:

Innkeeper::Tenant.drop('tenant_name')

When method is called, the schema is dropped and all data from itself will be lost. Be careful with this method.

Config

The following config options should be set up in a Rails initializer such as:

config/initializers/innkeeper.rb

To set config options, add this to your initializer:

Innkeeper.configure do |config|
  # set your options (described below) here
end

Excluding models

If you have some models that should always access the 'public' tenant, you can specify this by configuring Innkeeper using Innkeeper.configure. This will yield a config object for you. You can set excluded models like so:

config.excluded_models = ["User", "Company"] # these models will not be multi-tenanted, but remain in the global (public) namespace

Note that a string representation of the model name is now the standard so that models are properly constantized when reloaded in development.

Rails will always access the 'public' tenant when accessing these models, but note that tables will be created in all schemas. This may not be ideal, but its done this way because otherwise rails wouldn't be able to properly generate the schema.rb file.

NOTE - Many-To-Many Excluded Models: Since model exclusions must come from referencing a real ActiveRecord model, has_and_belongs_to_many is NOT supported. In order to achieve a many-to-many relationship for excluded models, you MUST use has_many :through. This way you can reference the join model in the excluded models configuration.

Postgresql Schemas

Providing a Different default_schema

By default, ActiveRecord will use "$user", public as the default schema_search_path. This can be modified if you wish to use a different default schema be setting:

config.default_schema = "some_other_schema"

With that set, all excluded models will use this schema as the table name prefix instead of public and reset on Innkeeper::Tenant will return to this schema as well.

Persistent Schemas

Innkeeper will normally just switch the schema_search_path whole hog to the one passed in. This can lead to problems if you want other schemas to always be searched as well. Enter persistent_schemas. You can configure a list of other schemas that will always remain in the search path, while the default gets swapped out:

config.persistent_schemas = ['some', 'other', 'schemas']

Installing Extensions into Persistent Schemas

Persistent Schemas have numerous useful applications. Hstore, for instance, is a popular storage engine for Postgresql. In order to use extensions such as Hstore, you have to install it to a specific schema and have that always in the schema_search_path.

When using extensions, keep in mind:

1. Ensure the extensions schema is created when the database is created

# lib/tasks/db_enhancements.rake

####### Important information ####################
# This file is used to setup a shared extensions #
# within a dedicated schema. This gives us the   #
# advantage of only needing to enable extensions #
# in one place.                                  #
#                                                #
# This task should be run AFTER db:create but    #
# BEFORE db:migrate.                             #
##################################################

namespace :db do
  desc 'Also create shared_extensions Schema'
  task :extensions => :environment  do
    # Create Schema
    ActiveRecord::Base.connection.execute 'CREATE SCHEMA IF NOT EXISTS shared_extensions;'
    # Enable Hstore
    ActiveRecord::Base.connection.execute 'CREATE EXTENSION IF NOT EXISTS HSTORE SCHEMA shared_extensions;'
    # Enable UUID-OSSP
    ActiveRecord::Base.connection.execute 'CREATE EXTENSION IF NOT EXISTS "uuid-ossp" SCHEMA shared_extensions;'
  end
end

Rake::Task["db:create"].enhance do
  Rake::Task["db:extensions"].invoke
end

Rake::Task["db:test:purge"].enhance do
  Rake::Task["db:extensions"].invoke
end

2. Ensure the schema is in Rails' default connection

Next, your database.yml file must mimic what you've set for your default and persistent schemas in Innkeeper. When you run migrations with Rails, it won't know about the extensions schema because Innkeeper isn't injected into the default connection, it's done on a per-request basis, therefore Rails doesn't know about hstore or uuid-ossp during migrations. To do so, add the following to your database.yml for all environments

# database.yml
...
adapter: postgresql
schema_search_path: "public,shared_extensions"
...

This would be for a config with default_schema set to public and persistent_schemas set to ['shared_extensions'].

3. Ensure the schema is in the innkeeper config

# config/initializers/innkeeper.rb
...
config.persistent_schemas = ['shared_extensions']
...

Alternative: Creating schema by default

Another way that we've successfully configured hstore for our applications is to add it into the postgresql template1 database so that every tenant that gets created has it by default.

One caveat with this approach is that it can interfere with other projects in development using the same extensions and template, but not using innkeeper with this approach.

You can do so using a command like so:

psql -U postgres -d template1 -c "CREATE SCHEMA shared_extensions AUTHORIZATION some_username;"
psql -U postgres -d template1 -c "CREATE EXTENSION IF NOT EXISTS hstore SCHEMA shared_extensions;"

The ideal setup would actually be to install hstore into the public schema and leave the public schema in the search_path at all times. We won't be able to do this though until public doesn't also contain the tenanted tables, which is an open issue with no real milestone to be completed. Happy to accept PR's on the matter.

Managing Migrations

In order to migrate all of your tenants (or postgresql schemas) you need to provide a list of dbs to Innkeeper. You can make this dynamic by providing a Proc object to be called on migrations. This object should yield an array of string representing each tenant name. Example:

# Dynamically get tenant names to migrate
config.tenant_names = lambda{ Customer.pluck(:tenant_name) }

# Use a static list of tenant names for migrate
config.tenant_names = ['tenant1', 'tenant2']

You can then migrate your tenants using the normal rake task:

rake db:migrate

This just invokes Innkeeper::Tenant.migrate(#{tenant_name}) for each tenant name supplied from Innkeeper.tenant_names.

Parallel Migrations

Innkeeper supports parallelizing migrations into multiple threads when you have a large number of tenants. By default, parallel migrations is turned off. You can enable this by setting parallel_migration_threads to the number of threads you want to use in your initializer.

Keep in mind that because migrations are going to access the database, the number of threads indicated here should be less than the pool size that Rails will use to connect to your database.

Tenants on different servers

Innkeeper supports tenant-based sharding at the application level. The switch, create, and drop methods all support full database configurations (as a hash) as well as tenant names. In fact, even when you pass a tenant name, it gets resolved to a full configuration using the configured tenant_resolver. If you wish to switch to a tenant on a different host, you can pass the full config with the host key.

Innkeeper will compare the config to it's current one and work out whether it needs to switch host, database, schema, etc, and only do the minimal switch. For tenants (databases for mysql, schemas for pg) on the same host, the switch will be a lightweight 'local' switch, which is one that occurs as a SQL query only, rather than a re-establishment of the database connection.

You could make use of a custom resolver to do multi-host tenant switching by name. You could map tenant names to a host (shard) IP via the hash of the host name, or something similar, and divide the hash space across available hosts.

Sidekiq

See innkeeper-sidekiq

Contributing

License

Innkeeper is released under the MIT License.

Bitdeli Badge