model-writtenrun 7185 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Sep 2020 · 2 entries
- Konstantin Knizhnik opened the thread with database login triggers for Oracle-migration customers in #1, deliberately leaving disconnect handling outside the proposal.
- Konstantin replaced the ordinary-trigger prototype with existing event-trigger machinery in #5; Pavel's review of startup permissions and session semantics led to corrections in #7 and Greg Nancarrow's client_connection naming and documentation revision in #11.
Dec 2020 - Jan 2021 · 3 entries
- Pavel proposed a generic disabling mechanism and a separate future session-reset event in #14, and Greg split the disabling GUC from login support in #15 while deferring the reset event.
- Pavel found roughly five percent connection overhead and an ownership-check crash in #19 and #23, prompting Konstantin to return to the earlier connection-specific GUC branch and add a pg_database fast-path flag in #37.
- Pavel's stale-flag concerns in #38 led to lazy clearing and rechecking in #41, followed by catalog documentation, regression and completion fixes culminating in Pavel's tested revision #50.
Mar 2021 · 1 entry
- Ivan Panchenko resumed the work with sample configuration and a practical session-initialization example in #56, then rebased it for PostgreSQL 14 in #57.
May - Dec 2021 · 4 entries
- Greg repaired memory leaks and snapshot handling in #59, adapted recovery tests to stricter stderr checking in #61, and kept the patch current in #63.
- Daniel Gustafsson's review in #66 challenged inconsistent naming, use of a plugin hook and special superuser bypasses, leading Greg to standardize login terminology and remove those mechanisms in #70.
- Teodor Sigaev demonstrated lockout risks and questioned failed-login auditing in #71; Ivan restored a recovery GUC in #75, but objections about bypassing audit infrastructure led Greg back to the earlier design in #81.
- Greg shortened the database flag name and repaired Windows authentication tests in #84, then maintained the series through rebases ending with #87.
Jan - Apr 2022 · 4 entries
- Daniel measured connection overhead in #88, while Robert Haas showed in #89 that interrupting psql during a sleeping login trigger could leave the server backend running.
- Andres Freund identified invalidation and locking errors and demanded standby safeguards and practical recovery in #90; Tom Lane questioned client-side audit bypasses in #95, while Robert argued against requiring a server shutdown for ordinary mistakes in #98.
- Daniel supplied a generic disabling GUC, the existing login patch and an integration delta in #99, then folded the integration and review fixes into #101.
- Andrey Sokolov's Windows failure report in #102 led to rewritten recovery tests in #103, and further standby discussion and Ivan's suggestions in #112 produced clearer warnings, safer examples and flag-clear tests in #113.
Sep 2022 - Mar 2023 · 8 entries
- After a long gap, Daniel rebased the work in #117, but Sergey's race, template-copy and reenabling reports in #119 and #121 exposed further fragility, and Daniel extracted the independently useful GUC into another thread in #122.
- Mikhail Gribkov took over in #124, fixing several flag bugs and offering a flagless alternative with benchmarks, while including a convenience backport of Daniel's separate GUC.
- Mikhail adopted the simpler flagless design in #125, accepting some connection overhead to eliminate catalog-flag complexity and leaving GUC development to its separate thread.
- Nikita's timezone-bypass report in #126 and Ted's review in #128 led to corrected examples and wording in #127 and #129.
- Pavel's testing and requests in #131 and #134 led Mikhail to add authentication-exception coverage in #132 and pg_dump coverage in #135.
- Following another rebase, Gregory Stark reported a pg_dump test failure in #142, which Mikhail repaired in #143 before a further maintenance rebase in #146.
- Robert reviewed an older GUC patch in #144, advocating positive naming and boolean values, and Daniel clarified in #145 that the independently developed version had moved to another thread.
- Alexander therefore revived the flagged implementation from #124, explaining the approach in #156 and attaching it in #158, with the flag read during InitPostgres and conditional locking for lazy cleanup.
Sep - Oct 2023 · 7 entries
- Alexander Korotkov revived the effort in #147, explicitly leaving failed-login auditing and psql cancellation behavior outside the immediate implementation while prioritizing recovery and connection overhead.
- Daniel committed the independently developed GUC from thread #46888 as 7750fefdb2, matching that thread's final patch #542987, and announced the prerequisite's availability here in #148; this supersedes attribution to the older convenience backport #124.
- Alexander integrated the committed GUC and documented slow-trigger cancellation limitations in #149, but Daniel's measurements in #150 and Alexander's confirmation in #151 established approximately five percent startup overhead.
- Robert's and Andres's concurrency reviews in #159 and #160 led to exclusive setter locking and a shared command-tag helper in #161, while conditional reset locking preserved the ability to connect without waiting for another transaction.
- After locking clarification and benchmarks in #167 and final test fixes and formatting in #168, Alexander committed login event triggers as e83d1b0c40, adding only the catalog-version bump to the attached patch.
- Michael Paquier identified immediate indentation and test-naming problems in #169, which Alexander reported fixing in #170; Mikhail separately reported a login-count race in #171 and moved its fix to another thread as recorded in #173.
- Tom Lane diagnosed a fatal-login TAP race involving IPC::Run in #174, approved Alexander's removal patch #175 in #176, and the patch landed with a comment edit as 06be01eb26.
Jan - Feb 2024 · 5 entries
- Alexander Lakhin reproduced an autovacuum-dependent regression failure in #177, and Daniel explained the nondeterministic flag cleanup in #180, leading directly to removal of the assertion in 4b885d01f9, announced in #181 without an attached patch.
- Lakhin then demonstrated a TOAST assertion with large ICU rules in #182, prompting Alexander's heap_inplace_update() fix in #184 to avoid both TOAST processing and row-lock waits.
- Lakhin verified that fix but raised concurrent-update concerns in #185, and Alexander agreed in #186 to await related discussion before proceeding.
- The last message, #187, supplied Bharath Rupireddy's documentation corrections and the fmgr.h include required to compile the C example; at that point these corrections and the in-place-update fix were still pending in the thread.
- Both final fixes subsequently landed: 8be93177c4 took #184 with an expanded concurrency comment, and 5fce30e77f took #187 unchanged, leaving the login feature and its follow-up fixes committed while the broader deferred features remained outside this work.