[
https://issues.apache.org/jira/browse/IMPALA-13191?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Noemi Pap-Takacs updated IMPALA-13191:
--------------------------------------
Labels: performance (was: easyfix performance)
> Do Not Create Sort Node with Constant Ordering Expression
> ---------------------------------------------------------
>
> Key: IMPALA-13191
> URL: https://issues.apache.org/jira/browse/IMPALA-13191
> Project: IMPALA
> Issue Type: Bug
> Components: fe, Frontend
> Reporter: Noemi Pap-Takacs
> Priority: Major
> Labels: performance
>
> Rows are sorted before inserting into partitioned Iceberg tables. See
> Planner.createPreDmlSort().
> If we update the partitioning column of the table, the sort ordering
> expression will be the partition column. If we set the new value to a
> constant (writing only to 1 partition), it will be evaluated as a constant.
> Therefore the Sort Node will get a constant ordering expression and works
> unnecessarily ordering indistinct values.
> For example:
> {code:java}
> create table ice_part partitioned by spec (l_discount) stored by iceberg
> tblproperties('format-version'='2') as select * from tpch_parquet.lineitem
> where l_linenumber=1;
> explain update ice_part set l_discount=0.11 where l_discount>0.07;
> {code}
> The output of explain - the plan - contains a Sort Node with the following
> ordering expression:
> {code:java}
> 04:SORT
> | order by: 0.11 ASC NULLS LAST
> {code}
> It is unnecessary to create a Sort Node to sort rows by a constant ordering
> expression. Constant expressions should be omitted, just like empty ones.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
