model-writtenrun 2232 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Dec 2024 - Mar 2025 · 16 entries
- Melanie Plageman proposed logging total connection setup, process creation, and authentication durations in #1, passing timestamps through startup structures so the implementation would also work with EXEC_BACKEND.
- Jelte Fennema-Nio asked about statistics views and distinguishing network waits from server work in #2, while Guillaume Lelarge successfully tested the prototype in #3; Melanie later explained that historical log records better served the intended diagnosis than existing statistics views in #6.
- After a month, Bertrand Drouvot supplied a minimal rebase and warned about wall-clock adjustments in #4, and Jacob Champion supported monotonic timing while questioning EXEC_BACKEND boundaries in #5; Melanie switched to instr_time in #7.
- Fujii Masao and Bertrand identified missing WAL-sender coverage, unnecessary log details, and questions about timing overhead in #10 and #11; Melanie added WAL-sender timing, conditional collection, simpler output, and documentation in #13 and #15.
- Andres Freund argued that timestamp overhead was negligible compared with connection creation but that extra log volume mattered in #18 and #19; Melanie consequently split the work into selectable log_connections categories and duration logging in #20, leaving consolidated connection messages for a larger project.
- Bertrand's testing exposed broken receipt logging in #23, and Melanie corrected bitmasks and EXEC_BACKEND initialization in #25, moved socket timestamps into BackendStartupData, and temporarily merged disconnection logging into the list interface.
- Bertrand and Fujii reviewed parsing, option combinations, documentation, and tests in #27 and #28, while Melanie considered a general set-valued GUC type but judged it too large for the release schedule in #29 and #32.
- Fujii demonstrated an invalid total duration when log_connections was enabled through client startup options in #30; Melanie made timing collection unconditional and temporarily reduced the interface to an on/off/timings enum in #34.
- After recognizing that a string list could retain common boolean spellings, Melanie restored the two-patch list-based design in #41, preserved log_disconnections as a separate setting, and retained the unconditional measurements needed for startup-option correctness.
- Bertrand's review in #42 led to readability and compatibility cleanups, and Melanie added TAP coverage in #43, although those initial tests still failed in CI.
- Andres recommended retaining absolute timestamps for debugging and possible future SQL access in #44; Melanie switched back to TimestampTz, stored individual event times, moved tests into the authentication suite, and excluded EXEC_BACKEND initialization overhead from process-creation timing in #45.
- Fujii found a duplicate timestamp read and an authentication interval that included EXEC_BACKEND setup in #47; Melanie fixed both in #53, aligning authentication timing with the authentication timeout boundary.
- Discussion among Melanie, Bertrand, and Jacob distinguished logging aspects from chronological stages in #49, #51, and #52; the revision in #53 adopted receipt, authentication, authorization, and durations, while deliberately declining legacy boolean abbreviations that could constrain future option names.
- Daniel Gustafsson reviewed naming, compatibility comments, total-duration wording, and test placement in #58; Melanie's final series in #59 adopted setup_durations, expanded documentation, and replaced the separate first-report boolean with a timestamp sentinel.
- Melanie noted in #59 that startup options affect setup_durations after the earlier connection messages have already been emitted, leaving that existing ordering difference unchanged.
- The modularization component of #59 landed as 9219093cab with documentation and GUC-description edits, and its duration component landed as 18cd15e706 with the backend predicate renamed IsExternalConnectionBackend; Melanie announced completion in #60.
May 2025 · 4 entries
- Two months later, Peter Eisentraut reported that documentation, debugging configuration, and tests still used deprecated boolean-style values in #61, and Daniel suggested enabling all connection logging in SSL tests in #62.
- Melanie explained why earlier versions had left those settings alone in #63, then submitted a cleanup using either targeted aspects, the former equivalent combination, or all in #64.
- Jacob requested receipt logging in the pre-authentication test in #65, while Peter approved the cleanup and the debug option's use of all in #66 and #67; those changes landed as cb1456423d.
- At the thread's final message, Melanie confirmed the cleanup was pushed in #68, leaving both feature components and the follow-up cleanup committed, with no attached patch still awaiting review.