LvYanquan created FLINK-35213:
---------------------------------
Summary: A potential inconsistent table structure issue
Key: FLINK-35213
URL: https://issues.apache.org/jira/browse/FLINK-35213
Project: Flink
Issue Type: Improvement
Components: Flink CDC
Affects Versions: 3.1.0
Reporter: LvYanquan
Currently, DataSinkWriterOperator will [request
CreateTableEvent|https://github.com/apache/flink-cdc/blob/313726b09690e82aa56fb5b42e89b535d24dadd7/flink-cdc-runtime/src/main/java/org/apache/flink/cdc/runtime/operators/sink/DataSinkWriterOperator.java#L149]
from SchemaRegistry when restarted.
However, If a SchemeChangeEvent is received during this process, SchemaOperator
will
1. [request
SchemaRegistry|https://github.com/apache/flink-cdc/blob/313726b09690e82aa56fb5b42e89b535d24dadd7/flink-cdc-runtime/src/main/java/org/apache/flink/cdc/runtime/operators/schema/SchemaOperator.java#L252]
to update the schema,
2. and then send FlushEvent.
As the network situation is quite complex, SchemaRegistry may update the schema
first, and then send a CreateTableEvent with the new schema, which is
incompatible with DatachangeEvent.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
