[
https://issues.apache.org/jira/browse/FLINK-38468?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jie Cheng updated FLINK-38468:
------------------------------
Attachment: 112.svg
> Performance drops when the coalesce parameter includes functions
> ----------------------------------------------------------------
>
> Key: FLINK-38468
> URL: https://issues.apache.org/jira/browse/FLINK-38468
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 1.18.0
> Reporter: Jie Cheng
> Priority: Major
> Labels: pull-request-available
> Fix For: 2.2.0
>
> Attachments: 112.svg, 1_12_StreamExecCalc$10.java,
> 1_18_StreamExecCalc$10.java
>
>
> I have two flink tasks, using 1.12 and 1.18 respectively. flinksql contains
> the COALESCE function. I found that the performance of 1.18 is not as good as
> that of 1.12
>
> use sql
> {code:java}
> select COALESCE(level_type, GET_JSON_OBJECT(order_n_tuple, '$.level_type'),
> '0') as level_type {code}
> 1.12 will convert COALESCE to case when. When the first parameter is not
> null, the GET_JSON_OBJECT function will not be called, while 1.18 will always
> call the GET_JSON_OBJECT function
>
> 1.12 Plan
> {code:java}
> Calc(select=[CASE(IS NOT NULL(level_type), CAST(level_type), IS NOT
> NULL(GET_JSON_OBJECT(order_n_tuple, _UTF-16LE'$.level_type')),
> CAST(GET_JSON_OBJECT(order_n_tuple, _UTF-16LE'$.level_type')),
> _UTF-16LE'0':VARCHAR(2147483647) CHARACTER SET "UTF-16LE") AS level_type]) +-
> TableSourceScan(table=[[default_catalog, default_database, shadow_source]],
> fields=[order_n_tuple, level_type])
> {code}
>
>
> 1.18 Plan
> {code:java}
> Calc(select=[COALESCE(level_type, GET_JSON_OBJECT(order_n_tuple,
> '$.level_type'), '0') AS level_type])
> +- TableSourceScan(table=[[default_catalog, default_database,
> shadow_source]], fields=[order_n_tuple, level_type]) {code}
>
> I plan to modify the optimization rule corresponding to the COALESCE function
> and convert it to case when the coalesce parameter contains the function
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
