Fix/init data race

[hwchen]

@jazzido @jspeis just wanted to start the conversation about patching the race condition on connection init.

As far as I can tell, this patch fixes NilClass errors that were occurring when there were near-simultaneous requests to MondrianRest.

Previously, constructing the connection and connecting occurred on the first api request. Since connecting took some time, there would be a moment when @@olap was no longer nil, but the new mondrian connection was not yet ready for use (resulting in, we think, the NilClass errors).

With this patch, now those initialization steps are done in the config.ru.

mondrian_conn = Mondrian::OLAP::Connection.new(PARAMS)
mondrian_conn.connect
Mondrian::REST::APIHelpers.class_variable_set('@@olap', mondrian_conn)

use Rack::Config do |env|
  env['mondrian-olap.conn'] = mondrian_conn
end

There's probably a better way to do this, so please let us know what you think @jazzido . This would be a pretty big breaking change if it had to be done in config.ru

[hwchen]

Oops, doesn't fix. Increasing the threads and the NilClass behavior returns.

[hwchen]

Ok, so some modifications and it looks like nilClass has not returned.

Changing the reference from olap to @@olap in get_cubes_or_404 appeared to get rid of the last NilClass.