centronet-marketing/surus

PostgreSQL extensions for ActiveRecord

Surus

Description

Surus adds PostgreSQL specific functionality to ActiveRecord. It adds helper methods for searching PostgreSQL arrays and hstores. It also can control PostgreSQL synchronous commit behavior. By relaxing PostgreSQL's durability guarantee, transaction commit rate can be increased by 50% or more. It can also directly generate JSON in PostgreSQL which can be substantially faster than converting ActiveRecord objects to JSON.

Installation

gem install surus

Or add to your Gemfile.

gem 'surus'

Rails 3

This version of Surus only works on Rails 4. Use the 0.4 line for Rails 3

gem 'surus', '~> 0.4.2'

Hstore

Hstores can be searched with helper scopes.

User.hstore_has_pairs(:properties, "favorite_color" => "green")
User.hstore_has_key(:properties, "favorite_color")
User.hstore_has_all_keys(:properties, "favorite_color", "gender")
User.hstore_has_any_keys(:properties, "favorite_color", "favorite_artist")

Hstore is a PostgreSQL extension. You can generate a migration to install it.

rails g surus:hstore:install
rake db:migrate

Even though the underlying hstore can only use strings for keys and values (and NULL for values) Surus can successfully maintain type for integers, floats, bigdecimals, dates, and any value that YAML can serialize. It does this by storing an extra key value pair (or two) to maintain type information.

Because it falls back to YAML serialization for complex types, this means that nested data structures can be serialized to an hstore. In other words, any hash that can be serialized with the normal Rails YAML serialization can be serialized with Surus.

Array

Arrays can be searched with helper scopes.

User.array_has(:permissions, "admin")
User.array_has(:permissions, "manage_accounts", "manage_users")
User.array_has_any(:favorite_integers, 7, 11, 42)

Synchronous Commit

PostgreSQL can trade durability for speed. By disabling synchronous commit, transactions will return before the data is actually stored on the disk. This can be substantially faster, but it entails a short window where a crash could cause data loss (but not data corruption). This can be enabled for an entire session or per transaction.

User.synchronous_commit # -> true

User.transaction do
  User.synchronous_commit false
  @user.save
end # This transaction can return before the data is written to the drive

# synchronous_commit returns to its former value outside of the transaction
User.synchronous_commit # -> true

# synchronous_commit can be turned off permanently
User.synchronous_commit false

Read more in the PostgreSQL asynchronous commit documentation.

JSON

PostgreSQL 9.2 added row_to_json and array_to_json functions. These functions can be used to build JSON very quickly. Unfortunately, they are somewhat cumbersome to use. The find_json and all_json methods are easy to use wrappers around the lower level PostgreSQL functions that closely mimic the Rails to_json interface.

User.find_json 1
User.find_json 1, columns: [:id, :name, :email]
Post.find_json 1, include: :author
User.find_json(user.id, include: {posts: {columns: [:id, :subject]}})
User.all_json
User.where(admin: true).all_json
User.all_json(columns: [:id, :name, :email], include: {posts: {columns: [:id, :subject]}})
Post.all_json(include: [:forum, :post])

Benchmarks

JSON generation is with all_json and find_json is substantially faster than to_json.

jack@hk-47~/dev/surus$ ruby -I lib -I bench bench/json_generation.rb
Generating test data... Done.
                                                              user     system      total        real
find_json: 1 record 500 times                             0.140000   0.010000   0.150000 (  0.205195)
to_json:   1 record 500 times                             0.240000   0.010000   0.250000 (  0.287435)
find_json: 1 record with 3 associations 500 times         0.480000   0.010000   0.490000 (  0.796025)
to_json:   1 record with 3 associations 500 times         1.130000   0.050000   1.180000 (  1.500837)
all_json:  50 records with 3 associations 20 times        0.030000   0.000000   0.030000 (  0.090454)
to_json:   50 records with 3 associations 20 times        1.350000   0.020000   1.370000 (  1.710151)

Disabling synchronous commit can improve commit speed by 50% or more.

jack@moya:~/work/surus$ ruby -I lib -I bench bench/synchronous_commit.rb
Generating random data before test to avoid bias... Done.

Writing 1000 narrow records
                                     user     system      total        real
enabled                          0.550000   0.870000   1.420000 (  3.025896)
disabled                         0.700000   0.580000   1.280000 (  1.788585)
disabled per transaction         0.870000   0.580000   1.450000 (  2.072150)
enabled / single transaction     0.700000   0.330000   1.030000 (  1.280455)
disabled / single transaction    0.660000   0.340000   1.000000 (  1.252301)

Writing 1000 wide records
                                     user     system      total        real
enabled                          1.030000   0.870000   1.900000 (  3.559709)
disabled                         0.930000   0.780000   1.710000 (  2.259340)
disabled per transaction         0.970000   0.850000   1.820000 (  2.478290)
enabled / single transaction     0.890000   0.500000   1.390000 (  1.693629)
disabled / single transaction    0.820000   0.450000   1.270000 (  1.554767)

Many more benchmarks are in the bench directory. Most accept parameters to adjust the amount of test data.

Running the benchmarks

1. Create a database
2. Configure bench/database.yml to connect to it.
3. Load bench/database_structure.sql into your bench database.
4. Run benchmark scripts from root of gem directory (remember pass ruby the include paths for lib and bench)

License

MIT