Open bobvawter opened 11 months ago
cc @cockroachdb/cdc
Thanks @bobvawter, we'll consider this as part of 24.1 planning
This could be achieved via CDC queries if we added support for functions that return the cluster ID and job ID. The trade off would be that this would be emitted per message instead of per batch, but it provides flexibility to the user.
CDC queries, though, are limited to a single table.
It would be helpful if the changefeed transport layer could include the source cluster uuid and the changefeed job id (e.g. a request HTTP header for ndjson or webhook deliveries). This would enable changefeed consumers to detect cases where an operator has recreated a feed or "crossed the streams" and fail in a somewhat informative manner when CDC invariants are violated (e.g. resolved timestamps not going backwards).
Jira issue: CRDB-32659
Epic CRDB-37343
gz#21394