Improve handling of multiple servers

[jingta]

Right now easy_stalk still seems to only connect to a single server instead of distribute across them.

[bardia-key]

we have also noticed that when the number of workers is smaller than the number of available servers, its taking a significant time for the worker pool to cycle through the servers and reserve jobs resulting in some jobs staying in ready for more than a few minutes.

[lanej]

max_age determines the time it takes to get a new connection. Dropping max_age will cause the workers to rotate servers more frequently. This is due to the fact that the pool is limited to 1.

conns = [1,2,3]
pool = EzPool.new(size: 1, timeout: 1, max_age: 300) { conns.sample }
5.times { pool.with { |c| puts c } }

1
1
1
1
1

conns = [1,2,3]
pool = EzPool.new(size: 1, timeout: 1, max_age: 0.01) { conns.sample }
5.times { pool.with { |c| puts c; sleep 2 } }

2
2
1
3
3

Something like:

    def self.create_worker_pool(tubes, config=EasyStalk.configuration)
      raise "beanstalkd_urls not specified in config" unless config.beanstalkd_urls
      EzPool.new(size: 1, timeout: 1, max_age: config.worker_reconnect_seconds, disconnect_with: lambda{ |client| client.close }) do
        beanstalkd = config.beanstalkd_urls.sample

        EasyStalk.logger.info "Watching tubes #{tubes} via #{beanstalkd} for jobs"

        Beaneater.new(beanstalkd).tap do |conn|
          conn.tubes.watch!(*tubes)
        end
      end
    end

would be a little more simple and random.