[
https://issues.apache.org/jira/browse/IMPALA-13307?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17874925#comment-17874925
]
Michael Smith commented on IMPALA-13307:
----------------------------------------
Good point, I'd forgotten about that one. We've separately discussed whether it
would be feasible to do whole-fragment codegen, but I don't think we're there
yet. I'll look at how different the implementation would be.
> disable_codegen_rows_threshold should apply to individual fragments
> -------------------------------------------------------------------
>
> Key: IMPALA-13307
> URL: https://issues.apache.org/jira/browse/IMPALA-13307
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Reporter: Michael Smith
> Priority: Major
> Labels: codegen
>
> The {{disable_codegen_rows_threshold}} is currently applied query-wide. So if
> a single scan deep in the query is expected to read a large number of rows,
> but everything else expects to see hundreds of rows because the results will
> be aggregated or heavily filtered, we still codegen the whole query.
> {{disable_codegen_rows_threshold}} should be evaluated for each fragment so
> we can determine whether to codegen based on the particular workload they'll
> be handling.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
