model-writtenrun 8700 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Mar 2020 ยท 16 entries
- Fujii Masao proposed pg_stat_progress_basebackup so administrators could inspect backup progress through SQL, supplying the initial implementation in #1.
- Kyotaro Horiguchi questioned client identification, phase naming, and a streaming-phase transition that preceded total initialization in #2; Fujii kept client details obtainable through a join with pg_stat_replication in #3 and corrected the transition and regression expectations in #5.
- Masahiko Sawada advocated phase names describing time-consuming operations in #4, while Amit Langote refined documentation and column names in #6, leading to the revised submission in #8.
- Amit proposed always calculating tablespace sizes in #10, but Fujii raised the cost of scanning the database before transferring data in #11, and Amit acknowledged the value of retaining an opt-out in #12; the unconditional server-side patch was not adopted.
- Amit requested a separate size-estimation phase in #18, and Fujii incorporated it alongside explicit checkpoint and WAL-archiving wait phases in #19; subsequent documentation review in #21 produced the corrections in #24.
- Fujii changed related progress values to update atomically in #28, preventing readers from seeing intermediate states, and corrected the regression expectations in #29.
- Sergei Kornilov and Kyotaro reviewed the final progress patch in #30 and #31; Fujii committed it as e65497df8f with formatting adjustments and a catalog-version bump, announcing the push in #32.
- Noriyoshi Shinoda supplied a post-commit documentation spelling fix in #34, which Fujii reported pushing in #35.
- Magnus Hagander reopened the coupling between server-side size estimates and client progress output in #36; after discussing estimation overhead in #42, Fujii proposed requesting estimates by default while retaining an explicit opt-out in #43.
- Magnus proposed the shorter --no-estimate-size option name in #44, and Kyotaro and Magnus favored rejecting its combination with --progress in #45 and #46; Fujii implemented that design in #48.
- Amit's documentation review in #49 led to #50, and Magnus's review in #51 led to #54, which removed the server-version gate so the client controls the new default behavior.
- Magnus also proposed NULL for an unknown total in #51; Fujii separated that change into #56, a patch applied on top of the client-option work.
- Magnus questioned placing NULL conversion in the SQL view in #57, but Fujii considered broader infrastructure changes excessive in #58, and code review by Alvaro Herrera and Amit supported retaining the command-specific -1 sentinel and view-local conversion in #62 and #64.
- Fujii committed the client-option patch from #54 unchanged apart from rebasing as fab13dc50b, announcing completion in #65.
- Andres Freund warned that default estimation could nearly double backup work on high-latency filesystems with many small relations in #67; Fujii left the default in place while suggesting measurement and reconsideration if beta feedback warranted it in #68.
- Fujii committed the NULL-reporting patch from #56 with only an added catalog-version bump as 67e0adfb3f and announced it in #69, leaving all three changes landed and the estimation-performance concern open for further evaluation.