SEZ9 commented on PR #11206:
URL: https://github.com/apache/seatunnel/pull/11206#issuecomment-6995577080

   Thanks for the update. The repair to the failure path looks right: 
failed-writer quarantine now runs after the barrier is released, and the 
two-replica fixtures exercise it. Here is what I still need on the previously 
raised points.
   
   **Still open**
   
   - **JDBC capability check.** `JdbcSinkFactory.supportsNewlyCreatedTable()` 
still declares the capability globally, even for incompatible configs. Please 
make it config-aware, so an exactly-once or custom-`query` JDBC sink is 
rejected before any writer is built. That way the job fails instead of the 
table being quarantined under `CONTINUE_OTHER_TABLES`.
   - **Retry while the barrier is held.** Quarantine now happens after release, 
but the retries appear to run before it. Do the retry sleeps still happen while 
every queue worker is parked at the schema-change barrier? If so, one failing 
new table still stalls all tables and checkpoints for the whole retry window, 
so please bound that window or move the retry off the barrier.
   - **Skipped rows.** Rows skipped for a quarantined table under 
`CONTINUE_OTHER_TABLES` are not guaranteed to replay later. Please make sure 
this is clearly visible to users.
   - **Unbounded runtime admission.** I don't see an upper bound or a removal 
path for runtime-admitted tables, so sink writers and checkpoint state can 
still grow without limit. Please add a bound, or explain how this is 
constrained.
   
   **Please confirm**
   
   - **Zeta with a single initial table.** The collector now updates the row 
type and sends the event before rows. Does this also cover a job that starts 
with exactly one matched table, where the produced type is not a 
`MultipleRowType`? A test for that case would settle it.
   - **DROP followed by re-CREATE.** What happens now when a second `CREATE 
TABLE` arrives for an already-registered path with a different schema? Please 
confirm the converters and the sink schema are refreshed rather than the event 
being ignored.
   
   **Docs (`docs/en/connectors/source/MySQL-CDC.md`)**
   
   I can't tell whether these three were updated, so please point me to the 
changes or add them:
   
   - How the runtime option relates to `schema-changes.enabled`, and that it 
overrides a user-supplied `debezium { include.schema.changes }`.
   - That `scan.binlog.newly-added-table.enabled` is rejected at startup when 
`table-names` is configured, in both the option row and the "Newly Added 
Tables" section.
   - Which engines support runtime table registration.
   
   <!-- streview-comment:2452 -->


-- 
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]

Reply via email to