[
https://issues.apache.org/jira/browse/FLINK-36701?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated FLINK-36701:
-----------------------------------
Labels: pull-request-available (was: )
> Pipeline failover again after handling a schema change event as the first
> event after a failover
> ------------------------------------------------------------------------------------------------
>
> Key: FLINK-36701
> URL: https://issues.apache.org/jira/browse/FLINK-36701
> Project: Flink
> Issue Type: Bug
> Components: Flink CDC
> Affects Versions: cdc-3.2.0, cdc-3.3.0
> Reporter: zjjiang
> Priority: Major
> Labels: pull-request-available
> Attachments: image-2024-11-13-16-29-30-065.png
>
>
> Currently, directly after a failover, when the pipeline first handles a
> schema change event (e.g. addColumnEvent) and then a DataChangeEvent, it may
> cause the job to fail again as sink has repeatedly applied that schema change.
> The cause of the problem can be explained as follows:
> 1. SinkWriterOperator now requests the latest schema when it receives the
> first non-createTableEvent schema change event (assuming there is no schema
> in the local cache).
> 2. The schema manager applies the schema change after confirming flush
> success.
> 3. Assume that the sequence after failover is to process a schema change
> event first, followed by a data change event.
> On the schema manager side, the schema manager will apply the schema change
> event to its cached schema(i.e. evolvedSchema) after confirming a successful
> flush.
> On the SinkWriterOperator side, the processing flow is:
> 1) Handle the flushEvent;
> 2) Handle the schema change event (in this step, the latest schema will be
> fetched from the schema manager and sent downstream; then the schema change
> event will be emitted) – note that this step does not report an error;
> 3) Handle the data change – here the failover occurs because the data record
> column size does not match the schema.
> !image-2024-11-13-16-29-30-065.png|width=1436,height=960!
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
