postgres=# select * from postgres_hacking where visibility = 'public';
/* agentic Postgres hacking in public — we use AI heavily, inspect every detail, test the ugly edges, and take full responsibility for the result */
2026-10-14 · 50:19
Test Salma El-Sayed’s B-tree leaf-page merge prototype: snapshots, parallel scans, VACUUM, and standby safety.
A proposed AI-assisted review pipeline for a small CommitFest queue: reproducible builds, red-green tests, adversarial reviews, and public summaries linked to the evidence. Designed to help human reviewers, not replace committer judgment.
Low-overhead core timing, trace capture, coverage analysis, independent observer-effect benchmarks, and honest query/plan-node attribution.
Expose cumulative server-message counts by backend, database, user, severity, and SQLSTATE without parsing log files.
Reduce shutdown latency and turn binary replacement, release-note remediation, and extension compatibility into a repeatable fleet workflow.
Log the correct commit-record LSN for DROP TABLE and related destructive operations to make PITR recovery less guessy.
Keep logical slots usable on a standby fed by WAL archives by pausing and safely resuming recovery around conflicts.
Inspired by the October 29 Hacking Postgres session, viewer Adam Brusselback implemented WHERE support for targeted REFRESH MATERIALIZED VIEW and has continued revising it through review.
Compress or batch larger WAL records, originating in B-tree build experiments.
Negotiate custom command-completion tags through a _pq_ protocol capability.
Salma El-Sayed’s implementation, mentored by Kirk Wolak, merges sparse siblings and reclaims merged-away tombstone pages without rebuilding the entire index.
Keep loadable slots for excluded table data so replacement data is restored before dependent indexes and constraints.
Dmitry’s pg_wait_tracer and core wait-event timing work, supported here with design sessions, reviews, coverage analysis, and independent benchmarks.
A proposed GSoC project that continued independently: systematically identify invisible waits and add coverage, starting with blocking writes to server logging destinations.
Quan continued the proposed GSoC work on LWLock contention. In parallel, an AWS-led patch series implements a broader pgstat-based redesign for the same scalability problem.
Standby snapshot corruption, seven upstream bug fixes, and a plan for adversarial CommitFest review.
Nik Samokhvalov, Andrey Borodin, and Kirk Wolak discussed standby snapshot corruption and how to review AI-assisted patches without losing human judgment. Nik reported about 30 bug reports with proposed fixes, seven committed fixes, and two false positives. The main proposal was Committer in Training: a small, low-risk CommitFest queue reviewed by adversarial agents, with red-green tests and public evidence. The group agreed to start a static bug-fixing record and work toward a patch-addressable CommitFest overlay.
Summary from session notes; timecodes pending captions.
Review the PG20-2 backlog: minor upgrades, global temporary tables, wait instrumentation, and pg_stat_log.
The group turned a tour of the PG20-2 CommitFest into concrete review homework. They reconciled Andrey Borodin's prolific August as 25 currently authored entries, explored bounded pre-shutdown checkpoints and safe cache handoff for minor upgrades, and examined why global temporary tables require deep catalog, relcache, planner, statistics, and lifecycle work. They then compared Ilmar Yunusov's EXPLAIN wait-event patch with Dmitry Fomin's broader timing/tracing series, insisting on honest attribution and workload benchmarks. The most actionable outcome was pg_stat_log: Nikolay will independently benchmark and review its SQL-queryable error counters, Kirk will support the thread, Andrey will revisit Dmitry's series, and Kirk will revise his pg_dump placeholder interface.
Agentic Postgres hacking with Andrey Borodin, Kirk Wolak, and Nik Samokhvalov. Detailed notes pending captions.
Transcript summary and timecodes pending captions.
Fix correctness and coverage of recovery landmarks for destructive DDL.
The group returned to commit-LSN logging for DROP TABLE and related destructive DDL, using several AI agents to review and repair the revived patch. They rejected an extension split, found that the hook exposed the end rather than the start of the commit record, and revised the code to log the precise recovery boundary without a concurrency race. After considering restore points and event triggers as alternatives, they audited documentation and an end-to-end restore test, then prepared a PostgreSQL-style commit message, a new pgsql-hackers revision, and CI for the fork.