Why not use shared ActiveRecord connections for Rs

Question

It seems the most commonly accepted way to deal with Selenium and tests is to avoid using transactional fixtures and then using something like database_cleaner between tests/scenarios. I recently ran into the following article which suggested doing the following:

spec_helper.rb

class ActiveRecord::Base
  mattr_accessor :shared_connection
  @@shared_connection = nil

  def self.connection
    @@shared_connection || retrieve_connection
  end
end

# Forces all threads to share the same connection. This works on
# Capybara because it starts the web server in a thread.
ActiveRecord::Base.shared_connection = ActiveRecord::Base.connection

This seems loads better for performance than the alternatives. Does anyone have any reason why this shouldn't be used?

Answer 1

Actually there are issues with it. If you use the gem mysql2, for example, you'll start seeing some errors like:

Mysql2::Error This connection is still waiting for a result

Please use this instead. It was written by Mike Perham, all credits to him.

class ActiveRecord::Base
  mattr_accessor :shared_connection
  @@shared_connection = nil

  def self.connection
    @@shared_connection || ConnectionPool::Wrapper.new(:size => 1) { retrieve_connection }
  end
end

ActiveRecord::Base.shared_connection = ActiveRecord::Base.connection

You'll need to install gem connection_pool too. This will spare you from many headaches.

Answer 2

I have encountered a problem using the code you mentioned in my spec_helper.rb file.

What happens when your tests depend on using connections to multiple databases? I have two databases I need to connect to when I run my tests. I did a simple test to check what was happening to the database connections I establish.

class ActiveRecord::Base
   mattr_accessor :shared_connection
   @@shared_connection = nil

   def self.connection
     @@shared_connection || retrieve_connection
  end
end

# Forces all threads to share the same connection. This works on
# Capybara because it starts the web server in a thread.
puts "First Record cxn: #{FirstDatabase::Record.connection}"
# => First Record cxn: #<ActiveRecord::ConnectionAdapters::Mysql2Adapter:0xe59b524>
puts "AR Base cxn: #{ActiveRecord::Base.connection}"
# => AR Base cxn: #<ActiveRecord::ConnectionAdapters::Mysql2Adapter:0xc52761c>
ActiveRecord::Base.shared_connection = ActiveRecord::Base.connection

puts "First Record cxn: #{FirstDatabase::Record.connection}"
# => First Record cxn: #<ActiveRecord::ConnectionAdapters::Mysql2Adapter:0xc52761c>
puts "AR Base cxn: #{ActiveRecord::Base.connection}"
# => AR Base cxn: #<ActiveRecord::ConnectionAdapters::Mysql2Adapter:0xc52761c>

As you can see, before I call the shared connection method, I have two different database connections. After, the shared connection method call, I have only one.

So any test that requires going to the second database connection to retrieve information will fail. :(

I'm going to post this problem and see if anyone has arrived at a solution.

Answer 3

There's a good thing at the end of this post. It may explain why I get a MALLOC error when I attempt to a very simple threading script.

http://apidock.com/rails/ActiveRecord/Base/connection

leente - March 15, 2011 0 thanks
Don't cache it!

Don’t store a connection in a variable, because another thread might try to use it when it’s already checked back in into the connection pool. See: ActiveRecord::ConnectionAdapters::ConnectionPool

connection = ActiveRecord::Base.connection

threads = (1..100).map do
  Thread.new do
    begin
      10.times do
        connection.execute("SELECT SLEEP(1)")  # WRONG
        ActiveRecord::Base.connection.execute("SELECT SLEEP(1)")  # CORRECT
      end
      puts "success"
    rescue => e
      puts e.message
    end
  end
end

threads.each(&:join)

Answer 4

I was just doing a little reading on this myself. I discovered the snippet you shared here in this blog post:

http://blog.plataformatec.com.br/2011/12/three-tips-to-improve-the-performance-of-your-test-suite/

To answer your question directly, the database cleaner github page cautions that it can " result in non-deterministic failures". I'd go right ahead and use it, but if you start running into weird failures, maybe this is a good place to start looking.

Answer 5

This solution was written by Jose Valim - well respected in the Rails community and a member of the Rails core team. I doubt he would recommend using it if there were issues with it. I personally haven't had any issues.

Just be aware that if you use Spork this needs to be in the each_run block to work.

FWIW - I have had intermittent capybara test issues with the above patch on Postgres. The Mike Perham solution that @hsgubert has below appears to have solved those issues. I am now use that solution.

Answer 6

The DatabaseCleaner gem readme answers your "why not" question this way:

One common approach is to force all processes to use the same database connection (common ActiveRecord hack) however this approach has been reported to result in non-deterministic failures.