model-writtenrun 8236 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alexander Korotkov supplied a reproduction and patch for an amcheck false corruption report spotted by Konstantin Knizhnik: WAL replay could leave deleted B-tree pages empty, violating amcheck's minimum-item checks #1; Alexander identified PostgreSQL 11 onward as affected #2.
- Peter Geoghegan initially suggested changing page-deletion replay and considering a related backward-scan problem #4, but Alexander distinguished the tuple-content mismatch from the sibling-link and locking issue and proposed clearing deleted pages on the primary too #5.
- Alexander proposed backpatching the amcheck fix while restricting the primary-side consistency change to master #6; Peter agreed that deleted pages must tolerate either historical contents and suggested extending WAL consistency checks to deleted pages #7.
- Alexander posted a two-patch series with expanded amcheck comments and the primary-side page-clearing change #8; Peter reviewed both and renewed his request to remove deleted-page masking from WAL consistency checking #9.
- The first patch from #8 landed unchanged as 34dae902ca, with backports through PostgreSQL 11, while the companion page-clearing patch remained pending.
- After a summer gap, Peter asked Alexander to finish the remaining work alongside the related backward-scan fix #10; Alexander submitted the remainder with deleted-page masking removed and reported that installcheck passed with B-tree WAL consistency checking enabled #11.
- Peter reviewed the revised remainder and requested a comment explaining the WAL consistency benefit #12; Alexander incorporated that suggestion, clarified that all remaining tuples are removed, and committed f47b5e1395 with only that comment changed from #11.
- Alexander confirmed the second commit and its wording adjustment #13, and Peter accepted the clarification #14, leaving both parts of the series committed.