[
https://issues.apache.org/jira/browse/IMPALA-14456?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Steve Carlin resolved IMPALA-14456.
-----------------------------------
Resolution: Later
> Calcite planner: Need a better estimate for deciding if a runtime filter is
> needed
> ----------------------------------------------------------------------------------
>
> Key: IMPALA-14456
> URL: https://issues.apache.org/jira/browse/IMPALA-14456
> Project: IMPALA
> Issue Type: Sub-task
> Reporter: Steve Carlin
> Assignee: Steve Carlin
> Priority: Major
>
> There is some code in ImpalaLoptOptimizeJoinRule which decides if a runtime
> filter should cause the left side and a right side of a join to be swapped.
> It compares row counts from the deepest left side versus the row count on the
> right side.
> This isn't a good comparison because we really want to associate a cost with
> the right side.
> There was some initial simplistic code that also multipled the row counts
> with the average row size, but that was incorrect because only projected
> columns come out of the deepest left side, not the whole row.
> This should be re-examined.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
