Load balancing for coordinators

Cloudera recommends setting up a load balancer for incoming connections to Impala.

By: Manish Maheshwari, Data Architect and Data Scientist at Cloudera, Inc.

Load balancers such as HaProxy or F5 usually work well to load balance incoming JDBC and Impala shell connections. The load balancer hostname should also be configured for Hue to connect to Impala.

The rule of thumb for load balancer configurations are as follows:

• Always configure your load balancer to use balance source, which uses the source IP address, to ensure that connections from one client are always routed to the same coordinator node.
• The load balancer timeout should be set to a value that is greater than the Impala idle session timeout. For example, if you are using HaProxy and set it up with a timeout of 5 minutes using timeout server 300s, ensure that the Impala idle timeout is set to less than 300 seconds.