[
https://issues.apache.org/jira/browse/FLINK-30771?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17680064#comment-17680064
]
jingjing dong commented on FLINK-30771:
---------------------------------------
function one and two are overloaded functions . in the script named
my_issue.sql , i create a table with 3 columns . i find the function one was
executed 3 times . just see the source code as below:
when version <= 8.0.22:
Lines from 1218 to 1220 in method 'Object
com.mysql.cj.jdbc.result.ResultSetImpl.getObject(int columnIndex)' .
case TIMESTAMP:
case DATETIME:
return this.getTimestamp(columnIndex);
when version >= 8.0.23:
Lines from 1230 to 1233 in method 'Object
com.mysql.cj.jdbc.result.ResultSetImpl.getObject(int columnIndex)'
case TIMESTAMP:
return this.getTimestamp(columnIndex);
case DATETIME:
return this.getLocalDateTime(columnIndex);
> different performance of conversion between datetime type in mysql and
> timestamp in flink
> -----------------------------------------------------------------------------------------
>
> Key: FLINK-30771
> URL: https://issues.apache.org/jira/browse/FLINK-30771
> Project: Flink
> Issue Type: Bug
> Components: Connectors / JDBC
> Affects Versions: 1.16.0
> Environment: System timezone is UTC and mysql version is 5.7.37;
> my laptap is window and i build my flink in wsl 2(Windows Subsystem for
> Linux) ;
> my mysql client is dbeaver;
> Reporter: jingjing dong
> Priority: Major
> Attachments: my_issue.sql
>
> Original Estimate: 504h
> Remaining Estimate: 504h
>
> in the attachment below named 'my_issue.sql', you can repeat my problem in
> eight steps。in simple terms，i think the conversion between datetime type in
> mysql and timestamp in flink need redefine carefully . after the research of
> source code ，i find something as below:
> # the timestime type in flink represent localdatetime in java
> # in class 'com.mysql.cj.jdbc.result.ResultSetImpl' , there are two
> overloaded function :function one is 'Object getObject(int columnIndex)' ,
> function two is '<T> T getObject(int columnIndex, Class<T> type)' . the
> chosen is function one by flink ,however the most popular ORM project like
> JPA/Mybatis choose function two.
> # since version 8.0.23 of mysql connector ，the function one has been
> refactored. and this refactor affect the conversion from datetime type in
> mysql to localdatetime in java. this conversion means fetch data from source
> mysql .
> # when the version of mysql connector is less than or equal with 8.0.22, the
> conversion is affected by the JDBC connection property "serverTimezone" which
> is defined in the WITH clause in the CREATE TABLE statment.
> # But when the version is greater than or equal with 8.0.23 , the result is
> different
> # in class
> 'org.apache.flink.connector.jdbc.internal.converter.AbstractJdbcRowConverter'
> , i am not sure if the localdatime type is cast into timestamp type, However
> i am pretty sure that no matter the version of mysql connector ,the
> conversion from localdatetime in java to datetime type in mysql is always
> affected by the JDBC connection property "serverTimezone" which is defined in
> the WITH clause in the CREATE TABLE statment. this conversion means sink
> data to mysql .
> in my option ,i suggest we should refer to the ORM project , neither
> the fetch from mysql or sink to mysql proccess should affect by the property
> "serverTimezone" .
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
