[ https://issues.apache.org/jira/browse/HBASE-24138?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sean Busbey updated HBASE-24138: -------------------------------- Description: Ran into a customer case where the StochasticLoadBalancer was consistently deciding not to balance when bringing new region servers on line. Even setting the class to TRACE logging would only log a summary statement like: {code} 2020-04-03 00:29:55,133 TRACE org.apache.hadoop.hbase.master.balancer.StochasticLoadBalancer: Skipping load balancing because balanced cluster; total cost is 25.24853189705185, sum multiplier is 602.0 min cost which need balance is 0.05 {code} Without any details about what went into that decision it's really hard to figure out what we need to tune to get the behavior we want. was: Ran into a customer case where the StochasticLoadBalancer was consistently deciding not to balance when bringing new region servers on line. Even setting the class to TRACE logging would only log a summary statement like: {code} 2020-04-03 00:29:55,133 TRACE org.apache.hadoop.hbase.master.balancer.StochasticLoadBalancer: Skipping load balancing because balanced cluster; total cost is 25.24853189705185, sum multiplier is 602.0 min cost which need balance is 0.05 {code} Without any details about what went into that decision it's really hard to figure out if what we need to tune to get the behavior we want. > Ensure StochasticLoadBalancer can log details of decision to not run balancer > ----------------------------------------------------------------------------- > > Key: HBASE-24138 > URL: https://issues.apache.org/jira/browse/HBASE-24138 > Project: HBase > Issue Type: Task > Components: Balancer, Operability > Reporter: Sean Busbey > Assignee: Sean Busbey > Priority: Major > > Ran into a customer case where the StochasticLoadBalancer was consistently > deciding not to balance when bringing new region servers on line. Even > setting the class to TRACE logging would only log a summary statement like: > {code} > 2020-04-03 00:29:55,133 TRACE > org.apache.hadoop.hbase.master.balancer.StochasticLoadBalancer: Skipping load > balancing because balanced cluster; total cost is 25.24853189705185, sum > multiplier is 602.0 min cost which need balance is 0.05 > {code} > Without any details about what went into that decision it's really hard to > figure out what we need to tune to get the behavior we want. -- This message was sent by Atlassian Jira (v8.3.4#803005)