diff --git a/README.md b/README.md
index 1faf754..81a6f39 100644
--- a/README.md
+++ b/README.md
@@ -29,31 +29,47 @@ gem install ruby-druid
 
 ## Usage
 
+1. Connect:
+
 ```ruby
-Druid::Client.new('zk1:2181,zk2:2181/druid').query('service/source')
+client = Druid::Client.new('zk1:2181,zk2:2181/druid', opts)
+datasource = client.data_source('druid:broker/datasource_name')
 ```
 
-returns a query object on which all other methods can be called to create a full and valid Druid query.
+if broker is behind of load balancer you can connect to static host without service discovery:
 
-A query object can be sent like this:
 
 ```ruby
-client = Druid::Client.new('zk1:2181,zk2:2181/druid')
-query = Druid::Query.new('service/source')
-client.send(query)
+datasource = Druid::DataSource.new('datasource_name', 'http://broker-host:8080/druid/v2/', opts)
 ```
 
-The `send` method returns the parsed response from the druid server as an array.  If the response is not empty it contains one `ResponseRow` object for each row.  The timestamp by can be received by a method with the same name (i.e. `row.timestamp`), all row values by hashlike syntax (i.e. `row['dimension'])
+`opts` is an optional hash of connection options:
+
+| key                 | description                                        | type   | default      |
+| ------------------- | -------------------------------------------------- | ------ | ------------ |
+| :open_timeout       | open timeout for druid services (in seconds)       | int    | 60           |
+| :read_timeout       | read timeout for druid services (in seconds)       | int    | nil          |
+| :discovery_path     | druid service discovery path in zookeeper          | string | '/discovery' |
+
+3. Create query:
+
+```ruby
+query = Druid::Query::Builder.new
+```
+4. Build query, e.g.:
+```ruby
+query.granularity(:all)
+query.long_sum(:aggregate1)
+# ....
+```
 
-An options hash can be passed when creating `Druid::Client` instance:
+5. Send request:
 
 ```ruby
-client = Druid::Client.new('zk1:2181,zk2:2181/druid', http_timeout: 20)
+result = datasource.post(query)
 ```
 
-Supported options are:
-* `static_setup` to explicitly specify a broker url, e.g. `static_setup: { 'my/source_name' => 'http://1.2.3.4:8080/druid/v2/' }`
-* `http_timeout` to define a timeout for sending http queries to a broker (in minutes, default value is 2)
+The `post` method returns the parsed response from the druid server as an array.  If the response is not empty it contains one `ResponseRow` object for each row.  The timestamp by can be received by a method with the same name (i.e. `row.timestamp`), all row values by hashlike syntax (i.e. `row['dimension'])
 
 ### GroupBy
 
diff --git a/lib/druid/client.rb b/lib/druid/client.rb
index 1462cff..c050c6d 100644
--- a/lib/druid/client.rb
+++ b/lib/druid/client.rb
@@ -7,12 +7,13 @@ class Client
     attr_reader :zk
 
     def initialize(zookeeper, opts = {})
+      @opts = opts
       @zk = ZK.new(zookeeper, opts)
     end
 
     def data_source(source)
       uri = @zk.data_sources[source]
-      Druid::DataSource.new(source, uri)
+      Druid::DataSource.new(source, uri, @opts)
     end
 
     def data_sources
diff --git a/lib/druid/data_source.rb b/lib/druid/data_source.rb
index d821207..702dda5 100644
--- a/lib/druid/data_source.rb
+++ b/lib/druid/data_source.rb
@@ -6,8 +6,10 @@ class DataSource
 
     attr_reader :name, :uri, :metrics, :dimensions
 
-    def initialize(name, uri)
+    def initialize(name, uri, opt = {})
       @name = name.split('/').last
+      @open_timeout = opt[:open_timeout] || 10 # if druid is down fail fast
+      @read_timeout = opt[:read_timeout] # we wait until druid is finished
       uri = uri.sample if uri.is_a?(Array)
       if uri.is_a?(String)
         @uri = URI(uri)
@@ -32,11 +34,11 @@ def metadata!(opts = {})
       end
 
       req = Net::HTTP::Get.new(meta_path)
-      response = Net::HTTP.new(uri.host, uri.port).start do |http|
-        http.open_timeout = 10 # if druid is down fail fast
-        http.read_timeout = nil # we wait until druid is finished
-        http.request(req)
-      end
+
+      http = Net::HTTP.new(uri.host, uri.port)
+      http.open_timeout = @open_timeout
+      http.read_timeout = @read_timeout
+      response = http.start { |h| h.request(req) }
 
       if response.code != '200'
         raise "Request failed: #{response.code}: #{response.body}"
@@ -61,11 +63,10 @@ def post(query)
       req = Net::HTTP::Post.new(uri.path, { 'Content-Type' => 'application/json' })
       req.body = query.to_json
 
-      response = Net::HTTP.new(uri.host, uri.port).start do |http|
-        http.open_timeout = 10 # if druid is down fail fast
-        http.read_timeout = nil # we wait until druid is finished
-        http.request(req)
-      end
+      http = Net::HTTP.new(uri.host, uri.port)
+      http.open_timeout = @open_timeout
+      http.read_timeout = @read_timeout
+      response = http.start { |h| h.request(req) }
 
       if response.code != '200'
         # ignore GroupBy cache issues and try again without cached results