1
0
Fork 0
mirror of https://github.com/rails/rails.git synced 2022-11-09 12:12:34 -05:00
rails--rails/guides/source/active_record_multiple_databases.md
eileencodes de6b4efa3e
Add option to skip joins for associations.
In a multiple database application, associations can't join across
databases. When set, this option tells Rails to make 2 or more queries
rather than using joins for associations.

Set the option on a has many through association:

```ruby
class Dog
  has_many :treats, through: :humans, disable_joins: true
  has_many :humans
end
```

Then instead of generating join SQL, two queries are used for `@dog.treats`:

```
SELECT "humans"."id" FROM "humans" WHERE "humans"."dog_id" = ?  [["dog_id", 1]]
SELECT "treats".* FROM "treats" WHERE "treats"."human_id" IN (?, ?, ?)  [["human_id", 1], ["human_id", 2], ["human_id", 3]]
```

This code is extracted from a gem we use internally at GitHub which
means the implementation here is used in production daily and isn't
experimental.

I often get the question "why can't Rails do this automatically" so I
figured I'd include the answer in the commit. Rails can't do this
automatically because associations are lazily loaded. `dog.treats` needs
to load `Dog`, then `Human` and then `Treats`. When `dog.treats` is
called Rails pre-generates the SQL that will be run and puts that
information into a reflection object. Because the SQL parts are pre-generated,
as soon as `dog.treats` is loaded it's too late to skip a join. The join
is already available on the object and that join is what's run to load
`treats` from `dog` through `humans`. I think the only way to avoid setting
an option on the association is to rewrite how and when the SQL is
generated for associations which is a large undertaking. Basically the
way that Active Record associations are designed, it is currently
impossible to have Rails figure out to not join (loading the association
will cause the join to occur, and that join will raise an error if the
models don't live in the same db).

The original implementation was written by me and Aaron. Lee helped port
over tests, and I refactored the extraction to better match Rails style.

Co-authored-by: Lee Quarella <leequarella@gmail.com>
Co-authored-by: Aaron Patterson <aaron@rubyonrails.org>
2021-04-19 11:17:31 -04:00

21 KiB

DO NOT READ THIS FILE ON GITHUB, GUIDES ARE PUBLISHED ON https://guides.rubyonrails.org.

Multiple Databases with Active Record

This guide covers using multiple databases with your Rails application.

After reading this guide you will know:

  • How to set up your application for multiple databases.
  • How automatic connection switching works.
  • How to use horizontal sharding for multiple databases.
  • How to migrate from legacy_connection_handling to the new connection handling.
  • What features are supported and what's still a work in progress.

As an application grows in popularity and usage you'll need to scale the application to support your new users and their data. One way in which your application may need to scale is on the database level. Rails now has support for multiple databases so you don't have to store your data all in one place.

At this time the following features are supported:

  • Multiple writer databases and a replica for each
  • Automatic connection switching for the model you're working with
  • Automatic swapping between the writer and replica depending on the HTTP verb and recent writes
  • Rails tasks for creating, dropping, migrating, and interacting with the multiple databases

The following features are not (yet) supported:

  • Automatic swapping for horizontal sharding
  • Load balancing replicas
  • Dumping schema caches for multiple databases

Setting up your application

While Rails tries to do most of the work for you there are still some steps you'll need to do to get your application ready for multiple databases.

Let's say we have an application with a single writer database and we need to add a new database for some new tables we're adding. The name of the new database will be "animals".

The database.yml looks like this:

production:
  database: my_primary_database
  adapter: mysql2
  username: root
  password: <%= ENV['ROOT_PASSWORD'] %>

Let's add a replica for the first configuration, and a second database called animals and a replica for that as well. To do this we need to change our database.yml from a 2-tier to a 3-tier config.

If a primary configuration is provided, it will be used as the "default" configuration. If there is no configuration named "primary", Rails will use the first configuration as default for each environment. The default configurations will use the default Rails filenames. For example, primary configurations will use schema.rb for the schema file, whereas all the other entries will use [CONFIGURATION_NAMESPACE]_schema.rb for the filename.

production:
  primary:
    database: my_primary_database
    username: root
    password: <%= ENV['ROOT_PASSWORD'] %>
    adapter: mysql2
  primary_replica:
    database: my_primary_database
    username: root_readonly
    password: <%= ENV['ROOT_READONLY_PASSWORD'] %>
    adapter: mysql2
    replica: true
  animals:
    database: my_animals_database
    username: animals_root
    password: <%= ENV['ANIMALS_ROOT_PASSWORD'] %>
    adapter: mysql2
    migrations_paths: db/animals_migrate
  animals_replica:
    database: my_animals_database
    username: animals_readonly
    password: <%= ENV['ANIMALS_READONLY_PASSWORD'] %>
    adapter: mysql2
    replica: true

When using multiple databases, there are a few important settings.

First, the database name for the primary and primary_replica should be the same because they contain the same data. This is also the case for animals and animals_replica.

Second, the username for the writers and replicas should be different, and the replica user's permissions should be set to only read and not write.

When using a replica database, you need to add a replica: true entry to the replica in the database.yml. This is because Rails otherwise has no way of knowing which one is a replica and which one is the writer.

Lastly, for new writer databases, you need to set the migrations_paths to the directory where you will store migrations for that database. We'll look more at migrations_paths later on in this guide.

Now that we have a new database, let's set up the connection model. In order to use the new database we need to create a new abstract class and connect to the animals databases.

class AnimalsRecord < ApplicationRecord
  self.abstract_class = true

  connects_to database: { writing: :animals, reading: :animals_replica }
end

Then we need to update ApplicationRecord to be aware of our new replica.

class ApplicationRecord < ActiveRecord::Base
  self.abstract_class = true

  connects_to database: { writing: :primary, reading: :primary_replica }
end

If you use a differently named class for your application record you need to set primary_abstract_class instead, so that Rails knows which class ActiveRecord::Base should share a connection with.

class PrimaryApplicationRecord < ActiveRecord::Base
  self.primary_abstract_class
end

Classes that connect to primary/primary_replica can inherit from your primary abstract class like standard Rails applications:

class Person < ApplicationRecord
end

By default Rails expects the database roles to be writing and reading for the primary and replica respectively. If you have a legacy system you may already have roles set up that you don't want to change. In that case you can set a new role name in your application config.

config.active_record.writing_role = :default
config.active_record.reading_role = :readonly

It's important to connect to your database in a single model and then inherit from that model for the tables rather than connect multiple individual models to the same database. Database clients have a limit to the number of open connections there can be and if you do this it will multiply the number of connections you have since Rails uses the model class name for the connection specification name.

Now that we have the database.yml and the new model set up, it's time to create the databases. Rails 6.0 ships with all the rails tasks you need to use multiple databases in Rails.

You can run bin/rails -T to see all the commands you're able to run. You should see the following:

$ bin/rails -T
rails db:create                          # Creates the database from DATABASE_URL or config/database.yml for the ...
rails db:create:animals                  # Create animals database for current environment
rails db:create:primary                  # Create primary database for current environment
rails db:drop                            # Drops the database from DATABASE_URL or config/database.yml for the cu...
rails db:drop:animals                    # Drop animals database for current environment
rails db:drop:primary                    # Drop primary database for current environment
rails db:migrate                         # Migrate the database (options: VERSION=x, VERBOSE=false, SCOPE=blog)
rails db:migrate:animals                 # Migrate animals database for current environment
rails db:migrate:primary                 # Migrate primary database for current environment
rails db:migrate:status                  # Display status of migrations
rails db:migrate:status:animals          # Display status of migrations for animals database
rails db:migrate:status:primary          # Display status of migrations for primary database
rails db:rollback                        # Rolls the schema back to the previous version (specify steps w/ STEP=n)
rails db:rollback:animals                # Rollback animals database for current environment (specify steps w/ STEP=n)
rails db:rollback:primary                # Rollback primary database for current environment (specify steps w/ STEP=n)
rails db:schema:dump                     # Creates a database schema file (either db/schema.rb or db/structure.sql  ...
rails db:schema:dump:animals             # Creates a database schema file (either db/schema.rb or db/structure.sql  ...
rails db:schema:dump:primary             # Creates a db/schema.rb file that is portable against any DB supported  ...
rails db:schema:load                     # Loads a database schema file (either db/schema.rb or db/structure.sql  ...
rails db:schema:load:animals             # Loads a database schema file (either db/schema.rb or db/structure.sql  ...
rails db:schema:load:primary             # Loads a database schema file (either db/schema.rb or db/structure.sql  ...

Running a command like bin/rails db:create will create both the primary and animals databases. Note that there is no command for creating the database users, and you'll need to do that manually to support the readonly users for your replicas. If you want to create just the animals database you can run bin/rails db:create:animals.

Generators and Migrations

Migrations for multiple databases should live in their own folders prefixed with the name of the database key in the configuration.

You also need to set the migrations_paths in the database configurations to tell Rails where to find the migrations.

For example the animals database would look for migrations in the db/animals_migrate directory and primary would look in db/migrate. Rails generators now take a --database option so that the file is generated in the correct directory. The command can be run like so:

$ bin/rails generate migration CreateDogs name:string --database animals

If you are using Rails generators, the scaffold and model generators will create the abstract class for you. Simply pass the database key to the command line.

$ bin/rails generate scaffold Dog name:string --database animals

A class with the database name and Record will be created. In this example the database is Animals so we end up with AnimalsRecord:

class AnimalsRecord < ApplicationRecord
  self.abstract_class = true

  connects_to database: { writing: :animals }
end

The generated model will automatically inherit from AnimalsRecord.

class Dog < AnimalsRecord
end

Note: Since Rails doesn't know which database is the replica for your writer you will need to add this to the abstract class after you're done.

Rails will only generate the new class once. It will not be overwritten by new scaffolds or deleted if the scaffold is deleted.

If you already have an abstract class and its name differs from AnimalsRecord, you can pass the --parent option to indicate you want a different abstract class:

$ bin/rails generate scaffold Dog name:string --database animals --parent Animals::Record

This will skip generating AnimalsRecord since you've indicated to Rails that you want to use a different parent class.

Activating automatic connection switching

Finally, in order to use the read-only replica in your application, you'll need to activate the middleware for automatic switching.

Automatic switching allows the application to switch from the writer to replica or replica to writer based on the HTTP verb and whether there was a recent write by the requesting user.

If the application is receiving a POST, PUT, DELETE, or PATCH request the application will automatically write to the writer database. For the specified time after the write, the application will read from the primary. For a GET or HEAD request the application will read from the replica unless there was a recent write.

To activate the automatic connection switching middleware, add or uncomment the following lines in your application config.

config.active_record.database_selector = { delay: 2.seconds }
config.active_record.database_resolver = ActiveRecord::Middleware::DatabaseSelector::Resolver
config.active_record.database_resolver_context = ActiveRecord::Middleware::DatabaseSelector::Resolver::Session

Rails guarantees "read your own write" and will send your GET or HEAD request to the writer if it's within the delay window. By default the delay is set to 2 seconds. You should change this based on your database infrastructure. Rails doesn't guarantee "read a recent write" for other users within the delay window and will send GET and HEAD requests to the replicas unless they wrote recently.

The automatic connection switching in Rails is relatively primitive and deliberately doesn't do a whole lot. The goal is a system that demonstrates how to do automatic connection switching that was flexible enough to be customizable by app developers.

The setup in Rails allows you to easily change how the switching is done and what parameters it's based on. Let's say you want to use a cookie instead of a session to decide when to swap connections. You can write your own class:

class MyCookieResolver
  # code for your cookie class
end

And then pass it to the middleware:

config.active_record.database_selector = { delay: 2.seconds }
config.active_record.database_resolver = ActiveRecord::Middleware::DatabaseSelector::Resolver
config.active_record.database_resolver_context = MyCookieResolver

Using manual connection switching

There are some cases where you may want your application to connect to a writer or a replica and the automatic connection switching isn't adequate. For example, you may know that for a particular request you always want to send the request to a replica, even when you are in a POST request path.

To do this Rails provides a connected_to method that will switch to the connection you need.

ActiveRecord::Base.connected_to(role: :reading) do
  # all code in this block will be connected to the reading role
end

The "role" in the connected_to call looks up the connections that are connected on that connection handler (or role). The reading connection handler will hold all the connections that were connected via connects_to with the role name of reading.

Note that connected_to with a role will look up an existing connection and switch using the connection specification name. This means that if you pass an unknown role like connected_to(role: :nonexistent) you will get an error that says ActiveRecord::ConnectionNotEstablished (No connection pool for 'ActiveRecord::Base' found for the 'nonexistent' role.)

Horizontal sharding

Horizontal sharding is when you split up your database to reduce the number of rows on each database server, but maintain the same schema across "shards". This is commonly called "multi-tenant" sharding.

The API for supporting horizontal sharding in Rails is similar to the multiple database / vertical sharding API that's existed since Rails 6.0.

Shards are declared in the three-tier config like this:

production:
  primary:
    database: my_primary_database
    adapter: mysql2
  primary_replica:
    database: my_primary_database
    adapter: mysql2
    replica: true
  primary_shard_one:
    database: my_primary_shard_one
    adapter: mysql2
  primary_shard_one_replica:
    database: my_primary_shard_one
    adapter: mysql2
    replica: true

Models are then connected with the connects_to API via the shards key:

class ApplicationRecord < ActiveRecord::Base
  self.abstract_class = true

  connects_to shards: {
    default: { writing: :primary, reading: :primary_replica },
    shard_one: { writing: :primary_shard_one, reading: :primary_shard_one_replica }
  }
end

Then models can swap connections manually via the connected_to API. If using sharding, both a role and a shard must be passed:

ActiveRecord::Base.connected_to(role: :writing, shard: :default) do
  @id = Person.create! # Creates a record in shard default
end

ActiveRecord::Base.connected_to(role: :writing, shard: :shard_one) do
  Person.find(@id) # Can't find record, doesn't exist because it was created
                   # in the default shard
end

The horizontal sharding API also supports read replicas. You can swap the role and the shard with the connected_to API.

ActiveRecord::Base.connected_to(role: :reading, shard: :shard_one) do
  Person.first # Lookup record from read replica of shard one
end

Migrate to the new connection handling

In Rails 6.1+, Active Record provides a new internal API for connection management. In most cases applications will not need to make any changes except to opt-in to the new behavior (if upgrading from 6.0 and below) by setting config.active_record.legacy_connection_handling = false. If you have a single database application, no other changes will be required. If you have a multiple database application the following changes are required if you application is using these methods:

  • connection_handlers and connection_handlers= no longer works in the new connection handling. If you were calling a method on one of the connection handlers, for example, connection_handlers[:reading].retrieve_connection_pool("ActiveRecord::Base") you will now need to update that call to be connection_handlers.retrieve_connection_pool("ActiveRecord::Base", role: :reading).
  • Calls to ActiveRecord::Base.connection_handler.prevent_writes will need to be updated to ActiveRecord::Base.connection.preventing_writes?.
  • If you need all the pools, including writing and reading, a new method has been provided on the handler. Call connection_handler.all_connection_pools to use this. In most cases though you'll want writing or reading pools with connection_handler.connection_pool_list(:writing) or connection_handler.connection_pool_list(:reading).
  • If you turn off legacy_connection_handling in your application, any method that's unsupported will raise an error (i.e. connection_handlers=).

Granular Database Connection Switching

In Rails 6.1 it's possible to switch connections for one database instead of all databases globally. To use this feature you must first set config.active_record.legacy_connection_handling to false in your application configuration. The majority of applications should not need to make any other changes since the public APIs have the same behavior. See the above section for how to enable and migrate away from legacy_connection_handling.

With legacy_connection_handling set to false, any abstract connection class will be able to switch connections without affecting other connections. This is useful for switching your AnimalsRecord queries to read from the replica while ensuring your ApplicationRecord queries go to the primary.

AnimalsRecord.connected_to(role: :reading) do
  Dog.first # Reads from animals_replica
  Person.first  # Reads from primary
end

It's also possible to swap connections granularly for shards.

AnimalsRecord.connected_to(role: :reading, shard: :shard_one) do
  Dog.first # Will read from shard_one_replica. If no connection exists for shard_one_replica,
  # a ConnectionNotEstablished error will be raised
  Person.first # Will read from primary writer
end

To switch only the primary database cluster use ApplicationRecord:

ApplicationRecord.connected_to(role: :reading, shard: :shard_one) do
  Person.first # Reads from primary_shard_one_replica
  Dog.first # Reads from animals_primary
end

ActiveRecord::Base.connected_to maintains the ability to switch connections globally.

Handling associations with joins across databases

As of Rails 7.0+, Active Record has an option for handling associations that would perform a join across multiple databases. If you have a has many through association that you want to disable joining and perform 2 or more queries, pass the disable_joins: true option.

For example:

class Dog < AnimalsRecord
  has_many :treats, through: :humans, disable_joins: true
  has_many :humans
end

Previously calling @dog.treats without disable_joins would raise an error because databases are unable to handle joins across clusters. With the disable_joins option, Rails will generate multiple select queries to avoid attempting joining across clusters. For the above association @dog.treats would generate the following SQL:

SELECT "humans"."id" FROM "humans" WHERE "humans"."dog_id" = ?  [["dog_id", 1]]
SELECT "treats".* FROM "treats" WHERE "treats"."human_id" IN (?, ?, ?)  [["human_id", 1], ["human_id", 2], ["human_id", 3]]

There are some important things to be aware of with this option:

  1. There may be performance implications since now two or more queries will be performed (depending on the association) rather than a join. If the select for humans returned a high number of IDs the select for treats may send too many IDs.
  2. Since we are no longer performing joins a query with an order or limit is now sorted in-memory since order from one table cannot be applied to another table.
  3. This setting must be added to all associations that you want joining to be disabled. Rails can't guess this for you because association loading is lazy, to load treats in @dog.treats Rails already needs to know what SQL should be generated.

Caveats

Automatic swapping for horizontal sharding

While Rails now supports an API for connecting to and swapping connections of shards, it does not yet support an automatic swapping strategy. Any shard swapping will need to be done manually in your app via a middleware or around_action.

Load Balancing Replicas

Rails also doesn't support automatic load balancing of replicas. This is very dependent on your infrastructure. We may implement basic, primitive load balancing in the future, but for an application at scale this should be something your application handles outside of Rails.

Schema Cache

If you use a schema cache and multiple databases, you'll need to write an initializer that loads the schema cache from your app. This wasn't an issue we could resolve in time for Rails 6.0 but hope to have it in a future version soon.