postgres=# select * from sessions where video_id = 'pVFH6VwjgwM';
Test Salma El-Sayed’s B-tree leaf-page merge prototype: snapshots, parallel scans, VACUUM, and standby safety.
Nik Samokhvalov, Kirk Wolak, and Salma El-Sayed tested the B-tree leaf-page merge prototype and agreed to prioritize correctness before benchmarks. The discussion covered boundary pivots, merge-group metadata, concurrent readers, and deferred VACUUM cleanup. Nik reported standby and parallel-scan failures; the group examined index-only-scan snapshot safety and the lifetime of pinned pages. The next step is focused reproduction, small fixes, and retesting against exact revisions, not a claim that the prototype is ready to ship.
Based on October 14 session summary and action items. Caption-backed timecodes are not available yet.
Salma described a boundary pivot and merge-group metadata replacing the earlier fix-up approach. The current prototype merges left into right under the same parent. Merges are explicitly invoked; VACUUM performs later cleanup. Automatic initiation and final SQL syntax remain design questions.
Merge, half-dead, deleted, and reusable describe logical states, not a requirement for separate VACUUM runs. The group emphasized retaining merge-group relationships until cleanup is safe, including interrupted cleanup and intermediate states.
Long-lived snapshots, index-only and parallel scans, direction changes, rescans, and concurrent VACUUM need independent result comparisons. A successful amcheck run or absence of a crash is not sufficient evidence of correctness.
Nik reported a standby PANIC and parallel-scan wrong results. Salma suggested missing first-page merge-state publication as one possible explanation, not a verified fix. The group located Oleg’s index-only-scan reproducer and discussed the reverted cleanup-lock change.
The session notes also link follow-up reports recorded after the live discussion. These cover parallel merge-skip state, cursor direction changes, and merge-join mark/restore. This is not a claim that every report was discussed on air; each issue has its own revision, evidence, and validation status.
Follow-up reports address VACUUM rechecks after lock waits, merges into already-vacuumed pages, interrupted cleanup, adjacent tombstones, merge-group validation, and parent-check locking. Use the campaign as the evolving inventory rather than a fixed defect count.
The campaign also tracks WAL-format compatibility and pageinspect extension versioning. The WAL report distinguishes source observations from behavior not independently reproduced. These are review reports, not claims of released Postgres defects.
Owners and targets from the session notes. Targets are plans, not completed releases.