PgParty

ActiveRecord migrations and model helpers for creating and managing PostgreSQL 10+ partitions!

Features

• Migration methods for partition specific database operations
• Model methods for querying partitioned data, creating adhoc partitions, and retreiving partition metadata

Limitations

• Partition tables are not represented correctly in db/schema.rb — please use the :sql schema format

Future Work

• Automatic partition creation (via cron or some other means)

Installation

Add this line to your application's Gemfile:

gem 'pg_party'

And then execute:

$ bundle

Or install it yourself as:

$ gem install pg_party

Note that the gemspec does not require pg, as some model methods may work for other databases. Migration methods will be unavailable unless pg is installed.

Configuration

These values can be accessed and set via PgParty.config and PgParty.configure.

• caching
  • Whether to cache currently attached partitions and anonymous model classes
  • Default: true
• caching_ttl
  • Length of time (in seconds) that cache entries are considered valid
  • Default: -1 (never expire cache entries)
• schema_exclude_partitions
  • Whether to exclude child partitions in rake db:structure:dump
  • Default: true
• create_template_tables
  • Whether to create template tables by default. Use the template: option when creating partitioned tables to override this default.
  • Default: true
• create_with_primary_key
  • Whether to add primary key constraints to partitioned (parent) tables by default.
    • This behavior is disabled by default as this configuration usually requires composite primary keys to be specified and ActiveRecord does not natively support composite primary keys. There are workarounds such as the composite_primary_keys gem.
    • This is not supported for Postgres 10 (requires Postgres 11+)
    • Primary key constraints must include all partition keys, for example: primary_key: [:id, :created_at], partition_key: :created_at
    • Partition keys cannot use expressions
    • Can be overridden via the create_with_primary_key: option when creating partitioned tables
  • Default: false
• include_subpartitions_in_partition_list
  • Whether to include nested subpartitions in the result of YourModelClass.partiton_list mby default. You can always pass the include_subpartitions: option to override this.
  • Default: false (for backward compatibility)

Note that caching is done in-memory for each process of an application. Attaching / detaching partitions will clear the cache, but only for the process that initiated the request. For multi-process web servers, it is recommended to use a TTL or disable caching entirely.

Example

# in a Rails initializer
PgParty.configure do |c|
  c.caching_ttl = 60
  c.schema_exclude_partitions = false
  c.include_subpartitions_in_partition_list = true
  # Postgres 11+ users starting fresh may consider the below options to rely on Postgres' native features instead of
  # this gem's template tables feature.
  c.create_template_tables = false
  c.create_with_primary_key = true
end

Usage

Migrations

Methods

These methods are available in migrations as well as ActiveRecord::Base#connection objects.

• create_range_partition
  • Create partitioned table using the range partitioning method
  • Required args: table_name, partitition_key:
• create_list_partition
  • Create partitioned table using the list partitioning method
  • Required args: table_name, partition_key:
• create_hash_partition (Postgres 11+)
  • Create partitioned table using the hash partitioning method
  • Required args: table_name, partition_key:
• create_range_partition_of
  • Create partition in range partitioned table with partition key between range of values
  • Required args: table_name, start_range:, end_range:
  • Create a subpartition by specifying a partition_type: of :range, :list, or :hash and a partition_key:
• create_list_partition_of
  • Create partition in list partitioned table with partition key in list of values
  • Required args: table_name, values:
  • Create a subpartition by specifying a partition_type: of :range, :list, or :hash and a partition_key:
• create_hash_partition_of (Postgres 11+)
  • Create partition in hash partitioned table for partition keys with hashed values having a specific remainder
  • Required args: table_name, modulus:, remainder
  • Create a subpartition by specifying a partition_type: of :range, :list, or :hash and a partition_key:
  • Note that all partitions in a hash partitioned table should have the same modulus. See Examples for more info.
• create_default_partition_of (Postgres 11+)
  • Create a default partition for values not falling in the range or list constraints of any other partitions
  • Required args: table_name
• attach_range_partition
  • Attach existing table to range partitioned table with partition key between range of values
  • Required args: parent_table_name, child_table_name, start_range:, end_range:
• attach_list_partition
  • Attach existing table to list partitioned table with partition key in list of values
  • Required args: parent_table_name, child_table_name, values:
• attach_hash_partition (Postgres 11+)
  • Attach existing table to hash partitioned table with partition key hashed values having a specific remainder
  • Required args: parent_table_name, child_table_name, modulus:, remainder
• attach_default_partition (Postgres 11+)
  • Attach existing table as the default partition
  • Required args: parent_table_name, child_table_name
• detach_partition
  • Detach partition from both range and list partitioned tables
  • Required args: parent_table_name, child_table_name
• create_table_like
  • Clone any existing table
  • Required args: table_name, new_table_name
• partitions_for_table_name
  • List all attached partitions for a given table
  • Required args: table_name, include_subpartitions: (true or false)
• parent_for_table_name
  • Fetch the parent table for a partition
  • Required args: table_name
  • Pass optional traverse: true to return the top-level table in the hierarchy (for subpartitions)
  • Returns nil if the table is not a partition / has no parent
• table_partitioned?
  • Returns true if the table is partitioned (false for non-partitioned tables and partitions themselves)
  • Required args: table_name
• add_index_on_all_partitions
  • Recursively add an index to all partitions and subpartitions of table_name using Postgres's ADD INDEX CONCURRENTLY algorithm which adds the index in a non-blocking manner.
  • Required args: table_name, column_name (all add_index arguments are supported)
  • Use the in_threads: option to add indexes in parallel threads when there are many partitions. A value of 2 to 4 may be reasonable for tables with many large partitions and hosts with 4+ CPUs/cores.
  • Use disable_ddl_transaction! in your migration to disable transactions when using this command with in_threads: or algorithm: :concurrently.

Examples

Create range partitioned table on created_at::date with two partitions:

class CreateSomeRangeRecord < ActiveRecord::Migration[5.1]
  def up
    # proc is used for partition keys containing expressions
    create_range_partition :some_range_records, partition_key: ->{ "(created_at::date)" } do |t|
      t.text :some_value
      t.timestamps
    end

    # optional name argument is used to specify child table name
    create_range_partition_of \
      :some_range_records,
      name: :some_range_records_a,
      start_range: "2019-06-07",
      end_range: "2019-06-08"

    # optional name argument is used to specify child table name
     create_range_partition_of \
       :some_range_records,
       name: :some_range_records_b,
       start_range: "2019-06-08",
       end_range: "2019-06-09"
  end
end

Create list partitioned table on id with two partitions:

class CreateSomeListRecord < ActiveRecord::Migration[5.1]
  def up
    # symbol is used for partition keys referring to individual columns
    create_list_partition :some_list_records, partition_key: :id do |t|
      t.text :some_value
      t.timestamps
    end

    # without name argument, child partition created as "some_list_records_<hash>"
    create_list_partition_of \
      :some_list_records,
      values: 1..100

    # without name argument, child partition created as "some_list_records_<hash>"
     create_list_partition_of \
       :some_list_records,
       values: 101..200
    
    # default partition support is available in Postgres 11 or higher
     create_default_partition_of \
       :some_list_records
  end
end

Create hash partitioned table on account_id with two partitions (Postgres 11+ required):

• A hash partition can be used to spread keys evenly(ish) across partitions
• modulus: should always equal the total number of partitions planned for the table
• remainder: is an integer which should be in the range of 0 to modulus-1

class CreateSomeHashRecord < ActiveRecord::Migration[5.1]
  def up
    # symbol is used for partition keys referring to individual columns
    # create_with_primary_key: true, template: false on Postgres 11 will rely on PostgreSQL's native partition schema
    # management vs this gem's template tables
    # Note composite primary keys will require a workaround in ActiveRecord, such as through the use of the composite_primary_keys gem
    create_hash_partition :some_hash_records, partition_key: :account_id, primary_key: [:id, :account_id],
    create_with_primary_key: true, template: false do |t|
      t.bigserial :id, null: false
      t.bigint :account_id, null: false
      t.text :some_value
      t.timestamps
    end

    # without name argument, child partition created as "some_list_records_<hash>"
    create_hash_partition_of \
      :some_hash_records,
      modulus: 2,
      remainder: 0

    # without name argument, child partition created as "some_list_records_<hash>"
    create_hash_partition_of \
      :some_hash_records,
      modulus: 2,
      remainder: 1
  end
end

Advanced example with subpartitioning: Create list partitioned table on account_id subpartitioned by range on created_at with default partitions. This example is for a table with no primary key... perhaps for some analytics use case.

• Default partitions are only supported in Postgres 11+

class CreateSomeListSubpartitionedRecord < ActiveRecord::Migration[5.1]
  def up
    create_list_partition :some_list_subpartitioned_records, partition_key: :account_id, id: false,
      template: false do |t|
      t.bigint :account_id, null: false
      t.text :some_value
      t.created_at
    end

    create_default_partition_of \
      :some_list_subpartitioned_records,
      name: :some_list_subpartitioned_records_default,
      partition_type: :range,
      partition_key: :created_at

    create_range_partition_of \
      :some_list_subpartitioned_records_default,
      name: :some_list_subpartitioned_records_default_2019,
      start_range: '2019-01-01',
      end_range: '2019-12-31T23:59:59'
    
    create_default_partition_of \
      :some_list_subpartitioned_records_default

    create_list_partition_of \
      :some_list_subpartitioned_records,
      name: :some_list_subpartitioned_records_1,
      values: 1..100,
      partition_type: :range,
      partition_key: :created_at
  
    create_range_partition_of \
      :some_list_subpartitioned_records_1,
      name: :some_list_subpartitioned_records_1_2019,
      start_range: '2019-01-01',
      end_range: '2019-12-31T23:59:59'

    create_default_partition_of
      :some_list_subpartitioned_records_1

     create_list_partition_of \
       :some_list_subpartitioned_records,
       name: :some_list_subpartitioned_records_2,
       values: 101..200,
       partition_type: :range,
       partition_key: :created_at

    create_range_partition_of \
      :some_list_subpartitioned_records_2,
      name: :some_list_subpartitioned_records_2_2019,
      start_range: '2019-01-01',
      end_range: '2019-12-31T23:59:59'

    create_default_partition_of \
      :some_list_subpartitioned_records_2
  end
end

Template Tables

Unfortunately, PostgreSQL 10 doesn't support indexes on partitioned tables. However, individual partitions can have indexes. To avoid explicit index creation for every new partition, we've introduced the idea of template tables. For every call to create_list_partition and create_range_partition, a clone <table_name>_template is created. Indexes, constraints, etc. created on the template table will propagate to new partitions in calls to create_list_partition_of and create_range_partition_of:

• Subpartitions will correctly clone from template tables if a template table exists for the top-level ancestor
• When using Postgres 11 or higher, you may wish to disable template tables and use the native features instead, see Configuration
  but this may result in you using composite primary keys, which is not natively supported by ActiveRecord.

class CreateSomeListRecord < ActiveRecord::Migration[5.1]
  def up
    # template table creation is enabled by default - use "template: false" or the config option to opt-out
    create_list_partition :some_list_records, partition_key: :id do |t|
      t.integer :some_foreign_id
      t.text :some_value
      t.timestamps
    end

    # create index on the template table
    add_index :some_list_records_template, :some_foreign_id

    # create partition with an index on "some_foreign_id"
    create_list_partition_of \
      :some_list_records,
      values: 1..100

    # create partition with an index on "some_foreign_id"
    create_list_partition_of \
      :some_list_records,
      values: 101..200
  end
end

Attaching Existing Tables as Partitions

Attach an existing table to a range partitioned table:

class AttachRangePartition < ActiveRecord::Migration[5.1]
  def up
    attach_range_partition \
      :some_range_records,
      :some_existing_table,
      start_range: "2019-06-09",
      end_range: "2019-06-10"
  end
end

Attach an existing table to a list partitioned table:

class AttachListPartition < ActiveRecord::Migration[5.1]
  def up
    attach_list_partition \
      :some_list_records,
      :some_existing_table,
      values: 200..300
  end
end

Attach an existing table to a hash partitioned table:

class AttachHashPartition < ActiveRecord::Migration[5.1]
  def up
    attach_hash_partition \
      :some_hash_records,
      :some_existing_table,
      modulus: 2,
      remainder: 1
  end
end

Detach a partition from any partitioned table:

class DetachPartition < ActiveRecord::Migration[5.1]
  def up
    detach_partition :parent_table, :child_table
  end
end

Safely cascading add_index commands

Postgres 11+ will automatically cascade CREATE INDEX operations to partitions and subpartitions, however CREATE INDEX CONCURRENTLY is not supported, meaning table locks will be taken on each table while the new index is built. Postgres 10 provides no way to cascade index creation natively.

• The add_index_on_all_partitions method solves for these limitations by recursively creating the specified index on all partitions and subpartitions. Index names on individual partitions will include a hash suffix to avoid conflicts.
• On Postgres 11+, the created indices are correctly attached to an index on the parent table
• On Postgres 10, if you are using Template Tables, you will want to add the index to the template table separately.
• This command can also be used on subpartitions to cascade targeted indices starting at one level of the table hierarchy

class AddSomeValueIndexToSomeListRecord < ActiveRecord::Migration[5.1]
  # add_index_on_all_partitions with in_threads option may not be used within a transaction
  # (also, algorithm: :concurrently cannot be used within a transaction)
  disable_ddl_transaction!

  def up
    add_index :some_records_template, :some_value # Only if using Postgres 10 with template tables

    # Pass the `in_threads:` option to create indices in parallel across multiple Postgres connections
    add_index_on_all_partitions :some_records, :some_value, algorithm: :concurrently, in_threads: 4
  end
end

For more examples, take a look at the Combustion schema definition and integration specs:

• https://github.com/rkrage/pg_party/blob/master/spec/dummy/db/schema.rb
• https://github.com/rkrage/pg_party/blob/master/spec/integration/migration_spec.rb

Models

Methods

Class methods available to all ActiveRecord models:

• partitioned?
  • Check if a model is backed by either a list or range partitioned table
  • No arguments
• range_partition_by
  • Configure a model backed by a range partitioned table
  • Required arg: key (partition key column) or block returning partition key expression
• list_partition_by
  • Configure a model backed by a list partitioned table
  • Required arg: key (partition key column) or block returning partition key expression
• hash_partition_by
  • Configure a model backed by a hash partitioned table
  • Required arg: key (partition key column) or block returning partition key expression

Class methods available to both range and list partitioned models:

• partitions
  • Retrieve a list of currently attached partitions
  • Optional include_subpartitions: argument to include all subpartitions in the returned list
• in_partition
  • Retrieve an ActiveRecord model scoped to an individual partition
  • Required arg: child_table_name
• partition_key_eq
  • Query for records where partition key matches a value
  • Required arg: value

Class methods available to range partitioned models:

• create_partition
  • Dynamically create new partition with partition key in range of values
  • Required args: start_range:, end_range:
• partition_key_in
  • Query for records where partition key in range of values
  • Required args: start_range, end_range

Class methods available to list partitioned models:

• create_partition
  • Dynamically create new partition with partition key in list of values
  • Required arg: values:
• partition_key_in
  • Query for records where partition key in list of values
  • Required arg: list of values

Class methods available to hash partitioned models:

• create_partition
  • Dynamically create new partition with hashed partition key divided by modulus equals remainder
  • Required arg: modulus:, remainder:
• partition_key_in
  • Query for records where partition key in list of values (method operates the same as for list partitions above)
  • Required arg: list of values

Examples

Configure model backed by a range partitioned table to get access to the methods described above:

class SomeRangeRecord < ApplicationRecord
  # block is used for partition keys containing expressions
  range_partition_by { "(created_at::date)" }
end

Configure model backed by a list partitioned table to get access to the methods described above:

class SomeListRecord < ApplicationRecord
  # symbol is used for partition keys referring to individual columns
  list_partition_by :id
end

Configure model backed by a hash partitioned table to get access to the methods described above:

class SomeHashRecord < ApplicationRecord
  # symbol is used for partition keys referring to individual columns
  hash_partition_by :id
end

Dynamically create new partition from range partitioned model:

# additional options include: "name:" and "primary_key:"
SomeRangeRecord.create_partition(start_range: "2019-06-09", end_range: "2019-06-10")

Dynamically create new partition from list partitioned model:

# additional options include: "name:" and "primary_key:"
SomeListRecord.create_partition(values: 200..300)

Dynamically create new partition from hash partitioned model:

# additional options include: "name:" and "primary_key:"
SomeHashRecord.create_partition(modulus: 2, remainder: 1)

For range partitioned model, query for records where partition key in range of values:

SomeRangeRecord.partition_key_in("2019-06-08", "2019-06-10")

For list and hash partitioned models, query for records where partition key in list of values:

SomeListRecord.partition_key_in(1, 2, 3, 4)

For all partitioned models, query for records matching partition key:

SomeRangeRecord.partition_key_eq(Date.current)

SomeListRecord.partition_key_eq(100)

For all partitioned models, retrieve currently attached partitions:

SomeRangeRecord.partitions

SomeListRecord.partitions(include_subpartitions: true) # Include nested subpartitions

For both all partitioned models, retrieve ActiveRecord model scoped to individual partition:

SomeRangeRecord.in_partition(:some_range_records_partition_name)

SomeListRecord.in_partition(:some_list_records_partition_name)

To create range partitions by month for previous, current and next months it's possible to use this example. To automate creation of partitions, run Log.maintenance every day with cron:

class Log < ApplicationRecord
  range_partition_by { '(created_at::date)' }

  def self.maintenance
    partitions = [Date.today.prev_month, Date.today, Date.today.next_month]

    partitions.each do |day|
      name = Log.partition_name_for(day)
      next if ActiveRecord::Base.connection.table_exists?(name)
      Log.create_partition(
        name: name,
        start_range: day.beginning_of_month,
        end_range: day.next_month.beginning_of_month
      )
    end
  end

  def self.partition_name_for(day)
    "logs_y#{day.year}_m#{day.month}"
  end
end

For more examples, take a look at the model integration specs:

• https://github.com/rkrage/pg_party/tree/master/spec/integration/model

Development

The development / test environment relies heavily on Docker.

Start the containers in the background:

$ docker-compose up -d

Install dependencies:

$ bin/de bundle
$ bin/de appraisal

Run the tests:

$ bin/de appraisal rake

Open a Pry console to play around with the sample Rails app:

$ bin/de console

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/rkrage/pg_party. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the Contributor Covenant code of conduct.

License

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

Code of Conduct

Everyone interacting in the PgParty project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.