[
https://issues.apache.org/jira/browse/KYLIN-4185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16968205#comment-16968205
]
ZhouKang commented on KYLIN-4185:
---------------------------------
In our production environment, the estimated size and real size is as follows：
cube_sample, segment: 20191102000000_20191103000000
estimate size: 1736731MB HTable size: 190701MB
cube_sample, segment: 20191001000000_20191008000000 (this is a merged segemtn)
estimate size: 49251MB HTable size: 1153314MB
> CubeStatsReader estimate wrong cube size
> ----------------------------------------
>
> Key: KYLIN-4185
> URL: https://issues.apache.org/jira/browse/KYLIN-4185
> Project: Kylin
> Issue Type: Improvement
> Reporter: ZhouKang
> Priority: Major
>
> CubeStatsReader estimate wrong cube size, which cause a lot of problems.
> when the estimated size is much larger than the real size, the spark
> application's executor number is small, and cube build step will take a long
> time. sometime the step will failed due to the large dataset.
> When the estimated size is much smaller than the real size. the cuboid file
> in HDFS is small, and there are much of cuboid file.
>
> In our production environment, both the two situation happened.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
