wgzhao opened a new issue, #12371:
URL: https://github.com/apache/seatunnel/issues/12371
### Context
#12355 rebuilt the source type of `SET` / `ENUM` columns on the DDL-event
path so the option list
survives (`SET('a','b','c')` instead of `SET(5)`). Its regression test is a
unit test
(`CustomMySqlAntlrDdlParserTest`, `connector-cdc-mysql`), which proves the
type string is
reconstructed correctly for `ADD COLUMN`, `MODIFY COLUMN` and `CHANGE
COLUMN`. It does not prove that
MySQL accepts the DDL the sink generates from that string. This issue tracks
that missing end-to-end
coverage.
### What is already covered, and what is not
Covered: the string produced by
`CustomAlterTableParserListener#getSourceColumnTypeWithLengthScale`
(`seatunnel-connectors-v2/connector-cdc/connector-cdc-mysql/src/main/java/org/apache/seatunnel/connectors/seatunnel/cdc/mysql/source/parser/CustomAlterTableParserListener.java`),
which renders `typeName(v1,v2,...)` from `Column#enumValues()`.
Not covered: that the generated statement is valid MySQL. The value flows
unchanged into the
generated DDL - `MysqlCreateTableSqlBuilder` uses `Column#getSourceType()`
verbatim for MySQL
(`seatunnel-connectors-v2/connector-jdbc/src/main/java/org/apache/seatunnel/connectors/seatunnel/jdbc/catalog/mysql/MysqlCreateTableSqlBuilder.java:195-197`)
- so a malformed type string only fails at the sink, against a real server.
That is exactly how the
original defect surfaced: `COMMON-17 unsupported convert type` for the `SET
UNSIGNED` form, and a
MySQL syntax error for the `SET(5)` form.
### Suggested coverage
Host it in `MysqlCDCWithSchemaChangeIT`
(`seatunnel-e2e/seatunnel-connector-v2-e2e/connector-cdc-mysql-e2e/src/test/java/org/apache/seatunnel/connectors/seatunnel/cdc/mysql/MysqlCDCWithSchemaChangeIT.java`)
with `mysqlcdc_to_mysql_with_schema_change.conf`, which already runs a zeta
job against a MySQL sink
with schema evolution:
1. Start the job on `shop.products`.
2. Run `ALTER TABLE products ADD COLUMN c_set SET('a','b','c') NULL` (plus
an `ENUM` variant).
3. Assert the job survives the DDL and the sink table gains the column,
comparing source and sink
`information_schema.COLUMNS.COLUMN_TYPE` the way the existing cases do.
4. Optionally repeat through `MODIFY` / `CHANGE COLUMN`, so all three DDL
forms are exercised
against a real server rather than only in the parser unit test.
### Points to settle while implementing
- `mysqlcdc_to_mysql_with_schema_change.conf` does not set
`schema_save_mode` explicitly. Confirm
which mode is in effect and whether the auto-create branch is the one
exercised; the user-visible
failure was in the generated `CREATE TABLE`, so a case that goes through
`CREATE_SCHEMA_WHEN_NOT_EXIST` is the one that matters.
- Decide whether to assert on the sink column type only, or also assert that
no unsupported-type
error is raised along the way.
### Scope notes
- The initial snapshot path is not affected: table structures discovered at
job start come from the
JDBC catalog and carry the full `information_schema.COLUMN_TYPE` literal.
- Only columns added or modified while the job is running go through the
code path this issue
targets.
### Related
- #12355 - the fix this coverage belongs to
- #12354 - the parsing-side defect fixed in #12355
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
