Multitenancy for Rails and ActiveRecord
Apartment 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, Apartment can help.
After having reached out via github issues and email directly, no replies ever came. Since we wanted to upgrade our application to Rails 6 we decided to fork and start some development to support Rails 6. Because we don't have access to the apartment gem itself, the solution was to release it under a different name but providing the exact same API as it was before.
We were never involved with the development of Apartment gem in the first place and this project started out of our own needs. We will be more than happy to collaborate to maintain the gem alive and supporting the latest versions of ruby and rails, but your help is appreciated. Either by reporting bugs you may find or proposing improvements to the gem itself. Feel free to reach out.
Add the following to your Gemfile:
gem 'ros-apartment', require: 'apartment'Then generate your Apartment config file using
bundle exec rails generate apartment:installThis will create a config/initializers/apartment.rb initializer file.
Configure as needed using the docs below.
That's all you need to set up the Apartment libraries. If you want to switch tenants on a per-user basis, look under "Usage - Switching tenants per request", below.
NOTE: If using postgresql schemas you must use:
- for Rails 3.1.x: Rails ~> 3.1.2, it contains a patch that makes prepared statements work with multiple schemas
How to separate your application data into different accounts or companies. GoRails #47
Before you can switch to a new apartment tenant, you will need to create it. Whenever you need to create a new tenant, you can run the following command:
Apartment::Tenant.create('tenant_name')If you're using the prepend environment config option or you AREN'T using Postgresql Schemas, this will create a tenant in the following format: "#{environment}_tenant_name". In the case of a sqlite database, this will be created in your 'db/' folder. With other databases, the tenant will be created as a new DB within the system.
When you create a new tenant, all migrations will be run against that tenant, so it will be up to date when create returns.
PostgreSQL works slightly differently than other databases when creating a new tenant. If you are using PostgreSQL, Apartment by default will set up a new schema and migrate into there. This provides better performance, and allows Apartment to work on systems like Heroku, which would not allow a full new database to be created.
One can optionally use the full database creation instead if they want, though this is not recommended
To switch tenants using Apartment, use the following command:
Apartment::Tenant.switch('tenant_name') do
# ...
endWhen switch is called, all requests coming to ActiveRecord will be routed to the tenant you specify (with the exception of excluded models, see below). The tenant is automatically switched back at the end of the block to what it was before.
There is also switch! which doesn't take a block, but it's recommended to use switch.
To return to the default tenant, you can call switch with no arguments.
When using schemas, you can also pass in a list of schemas if desired. Any tables defined in a schema earlier in the chain will be referenced first, so this is only useful if you have a schema with only some of the tables defined:
Apartment::Tenant.switch(['tenant_1', 'tenant_2']) do
# ...
endYou can have Apartment route to the appropriate tenant by adding some Rack middleware. Apartment 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/apartment.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 'apartment/elevators/subdomain' # or 'domain', 'first_subdomain', 'host'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 Apartment::Elevators::Subdomain
end
endIf 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/apartment/subdomain_exclusions.rb
Apartment::Elevators::Subdomain.excluded_subdomains = ['www']This functions much in the same way as Apartment.excluded_models. This example will prevent switching your tenant when the subdomain is www. Handy for subdomains like: "public", "www", and "admin" :)
To switch on the first subdomain, which analyzes the chain of subdomains of the request and switches to a tenant schema of the first name in the chain (e.g. owls.birds.animals.com would switch to "owls"). It can be used like so:
# application.rb
module MyApplication
class Application < Rails::Application
config.middleware.use Apartment::Elevators::FirstSubdomain
end
endIf 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/apartment/subdomain_exclusions.rb
Apartment::Elevators::FirstSubdomain.excluded_subdomains = ['www']This functions much in the same way as the Subdomain elevator. NOTE: in fact, at the time of this writing, the Subdomain and FirstSubdomain elevators both use the first subdomain (#339). If you need to switch on larger parts of a Subdomain, consider using a Custom Elevator.
To switch based on full domain (excluding the 'www' subdomains and top level domains ie '.com' ) use the following:
# application.rb
module MyApplication
class Application < Rails::Application
config.middleware.use Apartment::Elevators::Domain
end
endNote that if you have several subdomains, then it will match on the first non-www subdomain:
- example.com => example
- www.example.com => example
- a.example.com => a
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 Apartment::Elevators::HostHash, {'example.com' => 'example_tenant'}
end
endTo switch based on full host to find corresponding tenant name use the following:
# application.rb
module MyApplication
class Application < Rails::Application
config.middleware.use Apartment::Elevators::Host
end
endIf you want to exclude a first-subdomain, for example if you don't want your application to include www in the matching, in an initializer in your application, you can set the following:
Apartment::Elevators::Host.ignored_first_subdomains = ['www']With the above set, these would be the results:
- example.com => example.com
- www.example.com => example.com
- a.example.com => a.example.com
- www.a.example.com => a.example.com
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. Apartment 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 Apartment::Elevators::Generic, proc { |request| request.host.reverse }
end
endYour 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 < Apartment::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
endIn the examples above, we show the Apartment middleware being appended to the Rack stack with
Rails.application.config.middleware.use Apartment::Elevators::SubdomainBy 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 Apartment 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 Apartment::Elevators::Subdomain middleware would come after it. Trouble is, Apartment resets the selected tenant after the request is finish, so some redirects (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 Apartment middleware at a location in the Rack stack that makes sense for your needs, e.g.:
Rails.application.config.middleware.insert_before Warden::Manager, Apartment::Elevators::SubdomainNow work done in the Warden middleware is wrapped in the Apartment::Tenant.switch context started in the Generic elevator.
To drop tenants using Apartment, use the following command:
Apartment::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.
ros-apartment console configures two helper methods:
tenant_list- list available tenants while using the consolest(tenant_name:String)- Switches the context to the tenant name passed, if it exists.
ros-apartment also has a custom prompt that gives a bit more information about
the context in which you're running. It shows the environment as well as the tenant
that is currently switched to. In order for you to enable this, you need to require
the custom console in your application.
In application.rb add require 'apartment/custom_console'.
Please note that we rely on pry-rails to edit the prompt, thus your project needs
to install it as well. In order to do so, you need to add gem 'pry-rails' to your
project's gemfile.
The following config options should be set up in a Rails initializer such as:
config/initializers/apartment.rb
To set config options, add this to your initializer:
Apartment.configure do |config|
# set your options (described below) here
endThis is configurable by setting: tenant_presence_check. It defaults to true
in order to maintain the original gem behavior. This is only checked when using one of the PostgreSQL adapters.
The original gem behavior, when running switch would look for the existence of the schema before switching. This adds an extra query on every context switch. While in the default simple scenarios this is a valid check, in high volume platforms this adds some unnecessary overhead which can be detected in some other ways on the application level.
Setting this configuration value to false will disable the schema presence check before trying to switch the context.
Apartment.configure do |config|
config.tenant_presence_check = false
endEnabling this configuration will output the database that the process is currently connected to as well as which
schemas are in the search path. This can be enabled by setting to true the active_record_log configuration.
Please note that our custom logger inherits from ActiveRecord::LogSubscriber so this will be required for the configuration to work.
Example log output:
Apartment.configure do |config|
config.active_record_log = true
endIf you have some models that should always access the 'public' tenant, you can specify this by configuring Apartment using Apartment.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) namespaceNote 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_manyis NOT supported. In order to achieve a many-to-many relationship for excluded models, you MUST usehas_many :through. This way you can reference the join model in the excluded models configuration.
Apartment can be forced to use raw SQL dumps insted of schema.rb for creating new schemas. Use this when you are using some extra features in postgres that can't be represented in schema.rb, like materialized views etc.
This only applies while using postgres adapter and config.use_schemas is set to true.
(Note: this option doesn't use db/structure.sql, it creates SQL dump by executing pg_dump)
Enable this option with:
config.use_sql = trueBy 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_tenant = "some_other_schema"With that set, all excluded models will use this schema as the table name prefix instead of public and reset on Apartment::Tenant will return to this schema as well.
Apartment 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']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:
- Extensions can only be installed into one schema per database, so we will want to install it into a schema that is always available in the
schema_search_path - The schema and extension need to be created in the database before they are referenced in migrations, database.yml or apartment.
- There does not seem to be a way to create the schema and extension using standard rails migrations.
- Rails db:test:prepare deletes and recreates the database, so it needs to be easy for the extension schema to be recreated here.
# 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;'
# Grant usage to public
ActiveRecord::Base.connection.execute 'GRANT usage ON SCHEMA shared_extensions to public;'
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
endNext, your database.yml file must mimic what you've set for your default and persistent schemas in Apartment. When you run migrations with Rails, it won't know about the extensions schema because Apartment 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_tenant set to public and persistent_schemas set to ['shared_extensions']. Note: This only works on Heroku with Rails 4.1+. For apps that use older Rails versions hosted on Heroku, the only way to properly setup is to start with a fresh PostgreSQL instance:
- Append
?schema_search_path=public,hstoreto yourDATABASE_URLenvironment variable, by this you don't have to revise thedatabase.ymlfile (which is impossible since Heroku regenerates a completely different and immutabledatabase.ymlof its own on each deploy) - Run
heroku pg:psqlfrom your command line - And then
DROP EXTENSION hstore;(Note: This will drop all columns that usehstoretype, so proceed with caution; only do this with a fresh PostgreSQL instance) - Next:
CREATE SCHEMA IF NOT EXISTS hstore; - Finally:
CREATE EXTENSION IF NOT EXISTS hstore SCHEMA hstore;and hit enter (\qto exit)
To double check, login to the console of your Heroku app and see if Apartment.connection.schema_search_path is public,hstore
# config/initializers/apartment.rb
...
config.persistent_schemas = ['shared_extensions']
...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 apartment 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.
In order to migrate all of your tenants (or postgresql schemas) you need to provide a list of dbs to Apartment. 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:migrateThis just invokes Apartment::Tenant.migrate(#{tenant_name}) for each tenant name supplied
from Apartment.tenant_names
Note that you can disable the default migrating of all tenants with db:migrate by setting
Apartment.db_migrate_tenants = false in your Rakefile. Note this must be done
before the rake tasks are loaded. ie. before YourApp::Application.load_tasks is called
Apartment 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.
By default, when not using postgresql schemas, Apartment will prepend the environment to the tenant name to ensure there is no conflict between your environments. This is mainly for the benefit of your development and test environments. If you wish to turn this option off in production, you could do something like:
config.prepend_environment = !Rails.env.production?You can store your tenants in different databases on one or more servers.
To do it, specify your tenant_names as a hash, keys being the actual tenant names,
values being a hash with the database configuration to use.
Example:
config.with_multi_server_setup = true
config.tenant_names = {
'tenant1' => {
adapter: 'postgresql',
host: 'some_server',
port: 5555,
database: 'postgres' # this is not the name of the tenant's db
# but the name of the database to connect to, before creating the tenant's db
# mandatory in postgresql
}
}
# or using a lambda:
config.tenant_names = lambda do
Tenant.all.each_with_object({}) do |tenant, hash|
hash[tenant.name] = tenant.db_configuration
end
endBoth these gems have been forked as a side consequence of having a new gem name. You can use them exactly as you were using before. They are, just like this one a drop-in replacement.
See apartment-sidekiq or apartment-activejob.
You can execute callbacks when switching between tenants or creating a new one, Apartment provides the following callbacks:
- before_create
- after_create
- before_switch
- after_switch
You can register a callback using ActiveSupport::Callbacks the following way:
require 'apartment/adapters/abstract_adapter'
module Apartment
module Adapters
class AbstractAdapter
set_callback :switch, :before do |object|
...
end
end
end
endBy default, once apartment starts, it establishes a connection to the database. It is possible to
disable this initial connection, by running with APARTMENT_DISABLE_INIT set to something:
$ APARTMENT_DISABLE_INIT=true DATABASE_URL=postgresql://localhost:1234/buk_development bin/rails runner 'puts 1'
# 1-
In both
spec/dummy/configandspec/config, you will seedatabase.yml.samplefiles- Copy them into the same directory but with the name
database.yml - Edit them to fit your own settings
- Copy them into the same directory but with the name
-
Rake tasks (see the Rakefile) will help you setup your dbs necessary to run tests
-
Please issue pull requests to the
developmentbranch. All development happens here, master is used for releases. -
Ensure that your code is accompanied with tests. No code will be merged without tests
-
If you're looking to help, check out the TODO file for some upcoming changes I'd like to implement in Apartment.
mysql2 gem in some cases fails to install. If you face problems running bundle install in OSX, try installing the gem running:
gem install mysql2 -v '0.5.3' -- --with-ldflags=-L/usr/local/opt/openssl/lib --with-cppflags=-I/usr/local/opt/openssl/include
Apartment is released under the MIT License.