On Fri, Sep 11, 2026 at 10:07:37AM -0700, Bharath Rupireddy wrote:
> Upon thinking more, I think we need a test case where pg_resetwal sets
> the OID8 counter beyond the OID4 limit of 4 billion and verify that
> the TOAST chunk_id goes beyond that to properly cover the OID8 case.
> The tests added by v17-0004-Add-battery-of-tests-related-oid8.patch,
> although they create a table with OID8 TOAST, all use chunk_ids within
> the OID4 limits.

Yes, Rui has proposed a few things to cross the gap here:
https://www.postgresql.org/message-id/CAHWVJhGkb1n_QSS4M=pkjd5hhcwk3kxpfwvxorvlzag3j+2...@mail.gmail.com

One thing that he has done is to check a reorderbuffer change with the
control file holding a value larger than UINT32_MAX.  I agree that
having a "normal" SQL case would be adapted, but why not just plug one
SELECT into pg_upgrade once varatt_external_oid8 is introduced?  We
hold into toasttest_oid8 and toasttest_oid in patch 0004 for this
purpose.  The check cannot be added before 0005, but the basics are
here.
--
Michael

Attachment: signature.asc
Description: PGP signature

Reply via email to