On 2020-05-08 16:32:38 -0700, Cary Huang wrote: > I have added more regression test cases to the sequence replication > patch with emphasis on transactions and rollback per your > suggestions. I find that when a transaction is aborted with rollback, > the decoder plugin will not receive the change but the sequence value > will in fact advance if nextval() or setval() were called.
Right. The sequence advances shouldn't be treated transactionally. That's already (optionally) done similarly for messages, so you should be able to copy that code.