[
https://issues.apache.org/jira/browse/FLINK-37388?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17930911#comment-17930911
]
zexin gong commented on FLINK-37388:
------------------------------------
I think the cause of this problem is that when starting with
*scan.startup.mode: initial*, if an exception occurs during the snapshot
reading phase and checkpoint/savepoint is used for recovery,
MySqlPipelineRecordEmitter will reread the table structure and resend
CreateTableEvent for the table that has not been synchronized after recovery
during the Snapshot -> Binlog conversion process, causing the downstream to be
unable to correctly process the ddl information in the earlier binlog.
> [BUG] Field names must be unique.
> ---------------------------------
>
> Key: FLINK-37388
> URL: https://issues.apache.org/jira/browse/FLINK-37388
> Project: Flink
> Issue Type: Bug
> Components: Flink CDC
> Affects Versions: 3.0.0
> Reporter: zexin gong
> Priority: Major
> Attachments: screenshot-20250226-171451.png
>
>
> If configured of schema.change.behavior: EVOLVE, an exception may be
> triggered when reading binlog split: Field names must be unique. Found
> duplicates: [xxx]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
