model-writtenrun 1245 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Melanie Plageman reported that pruning redo leaves stale pd_prune_xid values on all-visible pages, creating confusing page state, and proposed either logging the prune XID or clearing the hint during recovery in #1.
- Heikki Linnakangas suggested logging a boolean indicating remaining prunable tuples in #2, while Andres Freund clarified that hot standbys cannot perform on-access pruning, placing the wasted pruning cycles after promotion in #4.
- Melanie explored using the WAL record's transaction ID in #5, but Heikki explained that it can be invalid and suggested a recent transaction ID in #6; Andres instead favored logging the exact prune XID to keep primary and standby pages consistent in #7.
- Returning in February, Melanie submitted #8 to clear prune hints when setting pages all-visible across redo, COPY FREEZE, and vacuum paths, addressing stale hints on ordinary primaries as well as promoted standbys without expanding WAL records.
- Andres supported the patch in #9 but identified a missing lazy_scan_prune() call and continued to advocate preserving accurate prune XIDs through WAL.
- Melanie committed 8b9d42bf6b from #8 with the additional lazy_scan_prune() clearing call, confirmed the fix in #10, and left logging pd_prune_xid in pruning WAL records as worthwhile follow-up work.