ActiveRecord::Bogacs

ActiveRecord (all-year) pooling "alternatives" ... in a relaxed 'spa' fashion.

Bogács is a village in Borsod-Abaúj-Zemplén county, Hungary.

NOTE: do not put this on production if you do not understand the consequences!

Install

add this line to your application's Gemfile:

gem 'activerecord-bogacs', :require => 'active_record/bogacs'

... or install it yourself as:

$ gem install activerecord-bogacs

Setup

Bogacs' pools rely on a small monkey-patch that allows to change the AR pool. The desired pool class needs to be set before establish_connection happens, thus an initializer (under Rails) won't work, you might consider setting the pool class at the bottom of your application.rb e.g. :

Bundler.require(*Rails.groups)

module MyApp
  class Application < Rails::Application
    # config.middleware.delete 'ActiveRecord::QueryCache'
  end
end

# sample AR-Bogacs setup using the "default" pool :
if Rails.env.production?
  pool_class = ActiveRecord::Bogacs::DefaultPool
  ActiveRecord::ConnectionAdapters::ConnectionHandler.connection_pool_class = pool_class
end

Alternatively, for FalsePool, there's a configuration convention (no need to path and set the connection_pool_class) :

production:
  adapter: mysql2
  <% if $servlet_context %>
  jndi: java:comp/env/jdbc/mydb
  pool: false # use AR::Bogacs::FalsePool
  <% else %>
  port: <%= ENV['DATABASE_PORT'] || 3306 %>
  database: mydb
  # ...
  <% end %>

This works only as long as one doesn't set a custom connection handler, since Bogacs only sets up its custom ActiveRecord::Base.default_connection_handler.

Pools are expected to work with older ActiveRecord versions: 3.x as well as 4.x.

Default Pool

Meant, primarily, as a back-port for users stuck with old Rails versions (< 4.0) on production, facing potential (pool related) concurrency bugs e.g. with highly concurrent loads under JRuby, although it also enhances some of the thread locking issues present in 4.x's pool.

Based on pool code from 4.x (which works much better than any previous version), with a few minor tunings and extensions such as pool_initial: 0.5 which allows to specify how many connections to initialize in advance when the pool is created.

False Pool

The false pool won't do any actual pooling, it is assumed that an underlying pool is configured. Still, it does maintain a hash of AR connections mapped to threads. Ignores pool related configuration such as pool: 42 or checkout_timeout: 2.5.

NOTE: be sure to configure an underlying pool e.g. with Trinidad (using the default Tomcat JDBC pool) :

---
  http: # true
    port: 3000
    # ...
  extensions:
    mysql_dbpool:
      url: jdbc:mysql:///my_production
      username: root
      jndi: jdbc/MyDB
      initialSize: <%= ENV['POOL_INITIAL'] || 25 %> # connections created on start
      maxActive: <%= ENV['POOL_SIZE'] || 100 %> # default 100 (AR pool: size)
      maxIdle: <%= ENV['POOL_SIZE'] || 100 %> # max connections kept in the pool
      minIdle: <%= ENV['POOL_INITIAL'] || 50 %>
      # idle connections are checked periodically (if enabled) and connections
      # that been idle for longer than minEvictableIdleTimeMillis will be released
      minEvictableIdleTimeMillis: <%= 3 * 60 * 1000 %> # default 60s
      # AR checkout_timeout: 5
      maxWait: <%= (( ENV['POOL_TIMEOUT'] || 5.0 ).to_f * 1000).to_i %> # default 30s

ActiveRecord-JDBC adapter allows you to lookup connection from JNDI using the following configuration :

production:
  adapter: mysql2
  jndi: java:/comp/env/jdbc/MyDB

NOTE: when using FalsePool there's nothing to configure (in database.yml)!

Shareable Pool

This pool allows for a database connection to be "shared" among threads, this is very dangerous normally. You will need to understand the underlying driver's connection implementation (whether its thread-safe).

You'll need to manually declare blocks that run with a shared connection (make sure only read operations happen within such blocks) similar to the built-in with_connection e.g. :

cache_fetch( [ 'user', user_id ] ) do
  ActiveRecord::Base.with_shared_connection { User.find(user_id) }
end

The pool "might" share connections among such blocks but only if it runs out of all connections (pool size is reached), until than it will always prefer checking out a connection just like with_connection does.

Tested with ActiveRecord-JDBC-Adapter using the official Postgres' driver (< 42).