Xin Gong created FLINK-36247:
--------------------------------
Summary: Potential transaction leak during MySQL snapshot phase
Key: FLINK-36247
URL: https://issues.apache.org/jira/browse/FLINK-36247
Project: Flink
Issue Type: Bug
Components: Flink CDC
Reporter: Xin Gong
copy from
[https://github.com/apache/flink-cdc/issues/1552]
We have found that during the MySQL snapshot phase, a transaction is initiated
but never committed, resulting in the transaction being left there for a long
time.
For instance, the binlog is like
{code:java}
SET autocommit=0 // a transaction is started by
com.ververica.cdc.connectors.mysql.source.utils.StatementUtils#initStatement
SELECT * FROM `database`.`table` WHERE id >= 1111 AND NOT (id = 2222) AND id <=
2222
SHOW MASTER STATUSSHOW GLOBAL VARIABLES LIKE 'GTID_MODE'SHOW MASTER STATUS
SELECT @@global.gtid_purgedrollback // the transaction is aborted without a
commit after a long wait {code}
Steps to reproduce the behavior:
1、The test data : Any MySQL database with existing data in the tableThe
1、test code : INSERT INTO any_sink SELECT * FROM mysql_cdc_source_table;
3、The error : A transaction is left without being committed.
>From the code, maybe it is the connection.setAutoCommit(false) that causes
>this issue, but we are still not sure why it is there in the first place.
!https://user-images.githubusercontent.com/5620838/189831936-522c8ae7-a116-4948-8b5d-41b4d0eaab72.png!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
