Hi Zhong, Luke I wanted to bring your attention to https://issues.apache.org/jira/browse/KYLIN-1518 whih had been filed before and we are consistently facing this issue with cubes on our end with large number of measures. It worked with upto 45 measures and there was no consistent behavior after that and breaks consistently.
Prashant, has added a comment to the ticket, wanted to check if we could help you with some consistent replication of the scenario and provide specific details that may help get to a possible solution. We are now building more fat cubes instead of many smaller cubes as we realized that the scale out case helps analysis from single cube much more than requiring us to devote time into breaking the data into smaller cubes and then asking users to join data across result sets. Thanks
