model-writtenrun 7815 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane reported intermittent pageinspect failures where VACUUM FREEZE left a tuple unfrozen, and also identified a duplicate query in the test script #1.
- Alvaro Herrera explained that vacuum can skip pinned pages and confirmed that checkpoint writes pin buffers; Tom therefore proposed DISABLE_PAGE_SKIPPING and broadened the investigation to other tests that depend on vacuum processing every page #2 #3 #4 #5.
- Michael Paquier identified check_heap.sql and heap_surgery.sql as possible additional cases and agreed that the duplicate query could be removed, before a patch was posted #6.
- Tom posted a patch covering pageinspect, amcheck and pg_visibility, excluded heap_surgery because it uses a temporary table, and left other vacuum calls alone where skipped pages would not affect visible results #7.
- Andres Freund noted that ordinary buffer replacement could also pin pages, while Tom considered checkpoint writes the more plausible explanation because they require no competing demand for buffers #8 #9.
- Michael accepted the temporary-table explanation and leaving check_btree.sql unchanged; his post-patch response reported neither testing nor a concrete defect or adopted change #10.
- Tom committed the sole patchset unchanged as c2dc1a7976, with backpatching to applicable branches, and closed the thread by confirming the push #11.