Package org.apache.hadoop.hbase.master.balancer

Class BalancerConditionals

java.lang.Object
org.apache.hadoop.hbase.master.balancer.BalancerConditionals
All Implemented Interfaces:
org.apache.hadoop.conf.Configurable
@Private final class BalancerConditionals extends Object implements org.apache.hadoop.conf.Configurable
Balancer conditionals supplement cost functions in the StochasticLoadBalancer. Cost functions are insufficient and difficult to work with when making discrete decisions; this is because they operate on a continuous scale, and each cost function's multiplier affects the relative importance of every other cost function. So it is difficult to meaningfully and clearly value many aspects of your region distribution via cost functions alone. Conditionals allow you to very clearly define discrete rules that your balancer would ideally follow. To clarify, a conditional violation will not block a region assignment because we would prefer to have uptime than have perfectly intentional balance. But conditionals allow you to, for example, define that a region's primary and secondary should not live on the same rack. Another example, conditionals make it easy to define that system tables will ideally be isolated on their own RegionServer (without needing to manage distinct RegionServer groups).